NVIDIA GeForce RTX 4090 Founders Edition Review & Benchmarks - Gaming, Power, & Thermals

The Not-So-Exotic DL ss3: A Cautionary Tale

In the world of high-performance computing, some technologies are touted as revolutionary, only to prove to be more hype than substance. The latest example is Nvidia's DL ss3, a technology that was heavily promoted during the company's keynote presentation. While there was indeed a lot of marketing talk surrounding this feature, we're not convinced that it will truly make an impact on the market.

In fact, we believe that the launch of the DL ss3 is more of a showcase for Nvidia's architecture generationally speaking. The company has been in a strong spot with their previous architectures, and it's only natural to want to build upon that success. However, by launching only the absolute flagship model, Nvidia is creating a perception that this technology is only for enthusiasts who can afford to pay top dollar.

This approach is not unique to Nvidia, as we've seen time and time again in the gaming industry. Companies often release high-end hardware at launch, only to lower the bar with more affordable options later on. Historically, these flagship models have been followed by mid-range and budget-friendly alternatives, which are typically priced significantly lower.

The Titan Class: A Realm of High-End Gaming

So, what does it mean to be in the "Titan class"? For Nvidia, this means that their graphics cards fall outside of the realm of mainstream gaming. While they're certainly capable of handling 4K resolutions and high frame rates, they're not necessarily designed for everyday gaming use cases.

In fact, if you're looking to play games at 1440p or lower with relatively high graphics quality, a more affordable option might be worth considering. The 30 series from Nvidia has already seen significant price drops, making it an attractive alternative to the latest flagship models. Similarly, AMD's 6,000 series has also become more competitive in terms of pricing, offering a viable option for those who don't want to pay top dollar for high-end hardware.

Power Consumption: A Concerning Trend

One area where Nvidia is showing significant growth is in power consumption. The company's flagship model, the DL ss3, consumes an astonishing 450-500 watts of power – and that's not even considering potential transients or overclocking. This kind of power draw can be a concern for builders, especially those who are looking to build high-performance gaming PCs.

Transients: A New Normal?

The company has assured us that the transients seen during testing were not concerning, but we're still skeptical. The duration of these transients – up to 11.1 milliseconds – is unusually long and could potentially be a cause for concern. We'll need to monitor this trend closely to see if it becomes an issue down the line.

The Thermal Design: A Mixed Bag

On the bright side, Nvidia's thermal design has finally been improved upon with the DL ss3. The company's use of massive fans and clever cooling solutions helps to mitigate the heat generated by this powerful GPU. However, we're still concerned about the potential for partner models to struggle with thermal management.

After all, Nvidia has infinite resources at their disposal – something that most partners simply can't match. This puts them in a difficult position when it comes to competing with the company on price and performance. It's an interesting dynamic that will likely play out over time as we see more partner models enter the market.

A Word of Caution: Value is Key

As always, value is key when it comes to purchasing high-end hardware. While the DL ss3 may be a powerful GPU, its high price point makes it difficult to recommend for all but the most serious enthusiasts.

We're still waiting to see how the 4080 model will shake things up in terms of pricing and performance. In the meantime, we'll continue to monitor the market and provide our take on which options are worth considering.

"WEBVTTKind: captionsLanguage: enit's time for early RTX 390 TI buyers to feel really bad because the short version of the review of the RTX 490 is that we're seeing rasterization uplift at 4K anywhere between 40 and 75% versus a 3090 ti so a big jump there and although Nvidia isn't allowing anyone to review the partner model cards until tomorrow uh the founders Edition can be reviewed today thinking of course about EVGA recently saying that they felt like they were competing with Nvidia this kind of rings true there either way the founders Edition is in for testing rumors leading up to launch of the $1,600 card were largely centered around thermals and power which are closely related so we'll be looking at those in depth today with transient spikes uh in the scope for analysis as well as the usual gaming and R tracing performance tests let's get started before that this video is brought to you by be quiet and their silent Wings 4 fans the silent Wings 4 fans Market themselves as being useful on radiators Tower coolers and cases alike the fans have a six-pole motor and use a fluid Dynamic bearing which helps with the noise profile and with longevity the fans use anti vibration mounts for reduced the noise transfer to the case and have a rated lifespan of $300,000s learn more at the link in the description below so first that we captured too much data for this review we're actually going to run another video with pressure tests some flatness testing of the cold plate we'll looking at frequency curves uh noise and Acoustics and more in a separate piece but this one is is already pretty detailed for power thermals and gaming uh we have over 30 charts just for gaming in here about 30 anyway and remember that as we ramp into this the cost of Alternatives currently is more important than the original MSRP of things like a 6950 XT or like a 3090 ti so just quickly glancing at Amazon as we're writing this review we saw that an RTX 380 today new in box cost about $700 to $740 on average in the US the 3090 is about $980 or so right now just on Amazon quick search and the 390 TI is in the range of $1,100 to $1,300 the $490 at $1,600 obviously very expensive you're looking at a 60% increase in price over say a 3090 right now today new in box and Nvidia still and its Partners have too many 30 series cards so that's going to be true for at least a little while so that Nvidia is launching the RTX 490s Standalone without the 480s with it should give everyone pause on pulling the trigger at least moment parly the company is clearly desperate to dump R tx30 inventory so every play is a move to ensure that we don't end up with the market that's overloaded with both 40 series and 30 series jamming the channel and the reason we're bringing that up is because they have two more cards both called 480 coming up in probably about November and AMD also has rdna A3 coming up shortly as well so uh we'll want to see how close the 480s fall to this and things like rasterization because if you look at like the a750 versus the a77 they're so close the 770 might as well not exist for most people so that's all the sort of disclaimers out of the way now we can really get into the performance and talk about how this thing does because genuinely this is one of the largest performance uplifts we've seen generationally in many years at in multiple Generations uh 1080ti saw huge jump similar to this but upwards of 75% is a massive Spike over the 390 ti so there is real power behind this card the the power numbers that have been discussed those are fueling something and it is an increase in performance now it's time to look at it though and uh we'll start with thermals we are leading our review with thermal and power over here stone is working on setting up the transients right now so he's doing some transient testing to see what the spikes look like and how bad they are we'll come back to that and then on my end I've wired up the card with all of these little wires running out of it these are thermocouples that we have attached to key components like mosfets memory components and we're also monitoring the GPU with software and you can see all these cards here we've done a lot of this type of testing in the past so actually I'll I'll show you some of the other cards we've done this on this is actually one of them that we're working on right now that's an RK 770 and we're actually keeping this shot for additional climate control so we can keep it flat for ambient and for others we have tested many of these on the GPU shelf here so a lot of these have been wired in the past in much the same way but our methods have improved now and we're using better hardware for the monitoring and I'm excited to see what the results are so let's get back in there so we have I think it's six thermocouples hooked up right now we're monitoring the ambient actively every second of the test it's something we normally do and what we're doing at the moment is this is our ambient Thermo couple probe it's up and away from any heat sources how we do a lot of testing so it's a really steady 20.0 de C plus or minus about 0.5 and then I've got a bunch of these wired up to memory specifically and then these two are wired up to ver mosfet we're running Port Royal on Loop at the moment but that's not the only test I'm doing and it's also a custom Port Royal Loop so we've increased the settings to load the GPU specifically in different areas and all of this will go into the thermal charts but to show you some context for what this actually takes to wire up because it's a nightmare let's go over to what I call the Arts and Crafts bench so this is a huge project to wire up cards with thermal couples where I had some print outs from our tear down video so I could just keep it keep track of where the thermal pads are and where everything goes we have the exact pads brand new clean that Nvidia used so we have the supply of the pads we have the supply of the uh phase change thermal interface material so that we're able to replace everything as needed as we do these tests with an exact like for like that Nvidia is using so enough of all the setup let's get into some of the thermal charts all right so the first test is Port Royal as a workload with some modified settings on the graphic side but full stock settings on the Nvidia side in this one the GPU core seemed to follow a temperature Target of 70° C it fluctuated between 66 and 72 and the GPU hotspot temperature ran in the 80s also fine but pushing it if you're in more of a hot boox case the fan speed would increase to compensate so your noise would be higher for the memory modules that we manually probed we mapped the thermocouples along the edge of the module that bordered the hottest components nearby memory module R2 held steady state at about 65° C that's case temperature so for actual internal silicon of the memory it could be as much as about 10° higher according to research papers previously published within the thero couple industry even at 75 that'd be acceptable the bottom module on the PCI slot ran in the low 60s with memory L2 in the same area as for the mosfets on the left and the right of the board those are running in the upper 50s and the low 60s since fets can take around 100 to 125 C on average without big problems this is all completely acceptable they're in good range here we'll also plot our ambient line just to sort of be transparent and show you how controlled our lab is you can also use this to follow the dips and see how they're Associated here's the bar chart with two different test scenarios both of these were with auto fan settings however the fans spun at the same speed for each was 45.4% average for the blue line and 47.0 for the red line which is within usual fluctuation for fan curve furm Mark with stock settings put more load on the GPU and the board which makes sense and it ran hotter for the GPU core in memory the other components impressively maintain the same temperature like for the memory and for the vrm components and that was due to good thermal balancing across the board so the GPU temperature following that 66 to 70° C range that indicates that is the vbios or the firmware that Nvidia has programmed being told to basically set the fans along a known curve to hit that temperature Target and that's completely acceptable as GPU temperature for silicon the hot spot's getting a little warm if you put it in a a really bad poor air flow case you will be on the higher side it'll run louder and uh the best solution is buy better case but some other types of designs may do better with that for example an AIO solution now since Nvidia talk such big game when going over the prototypes for this cooler we went ham wiring it up for thermal testing it took about 6 hours of wiring and validation work plus four additional of testing but we wanted to challenge everything that Nvidia claimed and from what we can see thermally this card is proving itself if you look at the lines in that first line graph we plotted you can see how tightly together all the memory modules and the fets are for temperature and that's scattered across the board so when we wired this up it wasn't all clustered in one spot it was across the whole PCB and that means there's good thermal load balancing across the PCB and some of that comes from design changes this generation like putting in blockers those little North and South fins internally that stop air recirculation along with obviously the increase in fan size that's a huge part of it it has downsides like case fitment uh but that's part of the thermal Improvement and then obviously the fin stack being as much surface area as it is so purely speaking thermal ignoring everything to do with value with gaming performance R tracing Performance dss3 Marketing and whatever other that gets talked about relating to this card when it launches purely thermally speaking the founders Edition here seems to actually be well-designed that actually is critical because that is abnormal that's the first time we've ever said that about an AMD card or an envidia card and certainly we've never said that about an Intel card and this one is actually able to compete with partner so Nvidia is now getting to a point where it is able to compete with Partners which is also scary because it starts to look like maybe EVGA knew something and just got out at the right time if their partner is going to be competing with them this well so that's it for The Thermals for now and we'll come back to The a770 Thermals later now we're going to look at Power transients where Patrick stone is setting up in testing the 490 for transient spikes and previously we saw something like 2 2 and a half X and transients for something like a 390 master from gigabyte and that could be potentially bad to knock your system offline so I'll let Stone explain what it is transients are what we're testing for yes so when we talk about transient spikes what we're looking for on the hardware level is current and we want to see how much current is coming across these wires and uh to get a baseline for this particular test we're using firmar then we'll jump to video games as well um the transients uh on this card are going to be a little more interesting because instead of having three cables to work with or two cables to work with now we got four cables to work with and so we've got to jump around here and there to figure out which one's delivering the most current and then we'll we'll use that to capture our transient spikes um right now we're just looking at micros seconds we're looking for these frames of high microsc so we got it set on a 200 microc per Division and in fermar we're catching things that are like 600 micr seconds for power supply over current protection you're really not worried about it until you get into like the millisecond territory but um you know well that's why we're testing things to see if it gets there so we'll see what happens on the 4090 so as for what all that means for the end user the reason we look at Power transients is because if you get a really bad transient spikes and you either have a lowquality power supply or one that's just not quite up to the capacity needs to cover that Spike you can end up with the system shutting down while you're playing a game or something something so that's why we're looking at it let's get into the charts we ultimately decided to clamp all four cables using three clamps which is fine since the rated range of the clamps supports that division we lose a marginal amount of accuracy in the low single digit percentages but nothing that'll actually affect the results we divided the fourth Cable's 12vt lines evenly between the three clamps for this testing overall in our testing the power Behavior seems well controlled the maximum current being pulled by the card is consistently about 33 to 40% over the nominal power which is 450 Watts most critically these zoomed in scope shots show a change in Behavior looking at the millisecond level at most we typically see spikes inside of 100 microsc on cards with the 4090 however the graphs almost look inverted the longest transient Spike we observed was 1,110 micros seconds or about 1.1 milliseconds that's long enough to potentially knock a borderline or lower-end power supply offline and much longer than we typically see in Rainbow Six Siege on Doom we saw approximately 500 microc La transients this is different from what we typically see where this one since the max Spike we're seeing is about 40% over nominal and for a longer period it's contrary to the usual much shorter much spikier and way higher power Spike like with a 3090 master from gigabyte where we saw spikes upwards of something like 2 and 1 12x or something like that if memory serves we have video on it uh so this is different and rumors had indicated spikes upwards of 2.5 to 3x over nominal for the 4090 for this card and we're not seeing that we weren't able to reproduce it doesn't mean it doesn't happen we only ran this for about 8 hours for transient testing then we had to do all the other stuff so someone might find something more interesting but from what we've seen for eight hours of work on different games and applications it's completely acceptable for the transients so far it's just that duration might kick more power spies offline uh if they don't have the capacity to soak now some of the rumors did come from Power spy manufacturers and it's possible and maybe likely that since they had early insight to what the card would require Nvidia probably provided them the feedback hey it's going to be 2.5 3x transend they freaked out said no way we can't do it and then maybe there was some design revision and rework there in the time since uh also we know that the GPS for the 490 were on 3090 TI pcbs and the 3090 TI PC was a test platform for the 490 maybe it wasn't up to the task we also noticed that initial small spikes led to the power delivery circuit balancing the load as time progressed and it does so in a really unique way we saw the highest load shifting from one cable to the next effectively Distributing the transient heat load as well across the individual cables whether this was total chance or part of the design we're not sure we'd have to do a lot more testing but we do know that this cable this uh adapter cable is really important so it does have the four sense pins in it and one of the things that we observed was if you connect three of these we tried getting rid of one to see what would happen it will still boot however you'll lose the ability to boost the power Target so normally you can go from 100% to 133% Max with all four connected with only three it drops down to just 100% with two we weren't able to get it to boot and post image to the screen and that's because Nvidia is actively sensing how many cables are connected to this ad adapter so it is a smart adapter in that way this isn't just a dumb plug them in and then it just pulls the power it wants to pull from whatever's connected it knows what's on the other side that's important we noticed that the power connector responsible for the highest current draw across that varied from test to test which is uncommon typically we see a dominant connector then a secondary and a tertiary and so forth this one demonstrated Behavior almost of a like a feedback loop that helps to regulate the card and its power drop so you'll definitely want to use all four if you get one of these and as for the aftermarket options we haven't tested them yet we don't know if it'll still pull the full power if you're only running two even though technically the cables can handle it or what'll happen that'll be a future thing but there's a lot more power testing we could do here for now let's just look at a nominal load chart and then if anyone discovers anything really unique on launch day we'll continue to investigate it but for now transience of 33% to 40% seems to be about the norm now for the averaged power consumption this shows the actual power utilization over longer periods so rather than the transient spikes upwards of 630 WS we're looking at sustained power tested in firmar for a torture load the RTX 490 stock card pulls 496 Watts on average we observed 450 to 490 watts in some gaming scenarios but closer to that 500 Mark and fur Mark that aligns with the 390 TI results previously and remember that this is just the card power so slot plus pcie cables we don't measure the whole system in this as for the 33% power in increase for an overclock the RTX 490 pushes to that's unfortunate let's move on to gaming benchmarks in Total War Warhammer 3 at 4K and high settings the RTX 490 is entirely GPU bound which is good for testing and runs at 125 FPS average that has it outdoing the RTX 390 TI by 59% and the RTX 390 by 81% the lows are also improved significantly on the 490 however they're mostly proportional to the average so it's nothing special here if you bought 390 TI though for $2,000 early this year it's probably feeling pretty bad right now they don't normally launch them this close together the 6950 XT Trails both and the RTX 380 but will do proportionately better at 1080P and 1440p we've also got the 2080 TI on here where the 490 outperforms it by an increase of 172% these are some of the largest gains we've ever seen generationally so regardless of price that at least is impressive on its own we'll talk talk about the value though and the conclusion at 1440p the 490 isn't yet CPU bottleneck so it's still actually showing scaling here the card maintains a 64% improvement over the 390 TI and an 83% improvement over the 3090 similar to the 4K results the 2 136 FPS average result has it about 100 FPS or so over in average over the 6950 XT from Sapphire which is a 90% Improvement finally for the 1080p results the 490 gets cut down to a 54% lead over the 390 TI that much of a s downwards isn't because of architecture alone but also again because we're starting to hit that CPU limitation this game in F1 2022 at 4K the 490 ran at 230 FPS average and held a 70% lead over the 390 TI that's similar to what we saw in total Warhammer uh although a bit better for the 490 here so it's shaping up to be the average of 60 to 70% right now for a rasterization lead the Boost over the 390 non TI is a 100% jump or a doubling of frame rate performance but the 6 950x he lands between the two in this game and cuts the 490s lead down to 73% as compared to its 133 FPS average entry those are proportionately scaled on the 390 TI the 390 and the 6950 XT they're still good on the 490 but they're not proportionately scaled with the average frame rate this chart will show us frame time consistency it's the most empirical way to look at the user experience as it shows the frame to frame time required in milliseconds to render each of those frames which just shown as one point on the line the RTX 490 mostly clusters around 4 to 5 milliseconds but regularly spikes to 9 milliseconds these aren't noticeable to the player typically we see that people observe 8 to 12 millisecond jumps as potential stutters but they are enough to throw that 1% out of proportion that we saw earlier there's a single Spike to 26 milliseconds that is noticeable to the player as a slight hitch or stutter but we normally have to see multiple of these in a test period to actually become concerned about them for now it's something to pay attention to as we iterate through the next charts the 6950 XT plots just as kind of a baseline for another device this one is mostly consistent throughout the test with fewer excursions though it does have a few excursions of its own moving to 1440p now the RTX 490 is approaching the CPU bind which is in the 290s for FPS average that means we're CPU bound on and off throughout this test but not for the entirety of it the 490 holds a 27% lead still over the 6950 XD but that's cut down from 73% previously some of that is AMD gaining at these lower resolutions but most of that is a CPU bind we won't learn much new at 1080p other than a reminder that the 6950 XE and the 6900 XE have an advantage over the 390 TI at this resolution Shadow the tomb Raiders up now at 4K the 490 maintains a 70% lead over the RTX 390 TI it's consistent with the previous results the 6950 XT and the 390 are both similarly ranked and if you were on an RTX 280 TI previously you'd see a 164% increase in frame rate moving to the 4090 almost leaked a new card there it's just called the 490 it's not the ti let's all move along before anybody from Nvidia hears that this is where we jump to a regular reminder that if you're happy with your computer today there's no need to upgrade right now even though something like this is shiny new and impressive because if you look at that chart we just saw everything on there is in playable territory you'll do just fine the frame rates are getting ridiculous at this point with these types of cards until you enable features like RT so it is something to really think about out do you actually need to upgrade and especially do you need to go this high end if there's stuff like 307s out there for cheaper now than they used to be 1440p reduces the 490 knot TI which doesn't exist lead over the 390 Ti from 70% to 27% a clear CPU cap the 6950 XT also encroaches on the 490 and it now outranks the 390 TI as well as we're bound here we'll move on but the value is clearly dismal if you're playing under a constraint like this and buying a 490 just does not make any sense just to show it since we have it here's the 1080p chart it's also bound obviously and everything above the 380 Mark is incomparable and useless because it's being limited by the CPU but gives you some perspective of where your limitations might lie Rainbow Six Siege at 4K gives us an older dx11 title to look at but one that's still popular this game has the 490 at 355 FPS average leading the 3090 TI by only about 47% only there compared to the previous results it's not spectacular for the price considering the 390 TI is currently closer to like 900 to 1,000 bucks and the 490 outperforms the 6950 XT by 86% here in average FPS which sets a mark to watch for lower resolutions at 1440p it's still not CPU bound the 490 leads the 390 TI by 41% consistent with the 4K result previously the gains in Rainbow Six Siege are just that much worse than in other games which shows that the pi and the game itself May dictate some of your scaling in a way that can wipe out most of the value of the 490 and again we won't spend much time on 1080p it's CPU bound now it did climb another 50 FPS first before hitting the CPU bind with these graphic settings uh but it's bound nonetheless the gap between the 6950 XT and the 390 TI has shrunken here although the 390 TI carries better lows than the AMD Hardware not that it is noticeable at this frame rate anyway and if you want better value you buy something further down the stack here's another game where the scaling isn't as impressive as those earlier 70% numbers in Horizon zero Dawn at 4K we saw a limitation scaling at 53% gain over the 39 DTI and were're not CPU bound that's uncharacteristic of the earlier results it's close but it's just not quite there 1440p shows us the limit at 180 FPS average with the 6950 XT also bouncing off of that limit the 6900xt and the 390 TI are close enough that some frames are likely getting capped but not all of them so these are all BAS basically the same clearly you should not buy a 4090 if you're going to play in a scenario like this it's waste of money 1080p is fully capped same thing nothing new to see here 490 looks like it's actually doing worse than the 390 in this one in reality it isn't it's just that they're all limited that we can't see scaling they're tripping over their own frames as the CPU tries to keep up and you end up with a stack that you can't differentiate Final Fantasy at 4K is up now and this one maybe shockingly we were still GPU bound the game is so easy to run that we thought we might actually be totally CPU limited even at 4K but it's still GPU targeted at high resolutions the RTX 490 ran at 214 FPS average which positions it 75% ahead of the 390 TI's 122 FPS average that's better scaling than we've seen elsewhere maybe partly thanks to dx11 Nvidia has always done disproportionately well on dx11 since that API requires more driver work to run well the lead over the 6950 XT is 88% with the 3090 getting doubled cleanly by the 490 at 1440p we're hitting the CPU buying the 490 runs at 235 FPS average which is functionally the same as the 220 result we had on the 390 ti so there's no real difference here and we'll skip 1080 this time it's the same we also ran GTA 5 at all three resolutions but only 4K matters since the engine is capped at 187.5 FPS average makes basically everything the same the RTX 490 ran at the full cap here dropping almost zero frames you have to realize we've been running this Benchmark since 2015 now on gpus we've never seen a card to hold base basically the engine cap like this the lead over the 39 TI is only 38% so in that regard it's not worth it the value doesn't scale like in other games because we're artificially limited on the engine the CPU isn't capped either just to be clear it's only Rage that's capped and not mine but like the name of the game engine my rage is uncapped there's no cap No Cap okay no capping now we get into the really complicated testing which is Ray tracing and dlss or FSR XS test and those three sets of initialisms are why this has gotten so complex so not only do we have to test for just pure rasterization we also need to test for rate tracing but then because features like dlss are so prominent now we have to enable those and there's different ones for each vendor so this has become a massive Matrix of really messy tests to run we've done our best to distill it down into a few sets of games to give you a wide picture performance let's take a look we'll start with control which came back as one of the most played RT games when he pulled our community you all recently on YouTube Community posts cyberp Punk was one of the other most popular we'll test that too starting with 4K and without dlss the RTX 490 ran at 76 FPS average here that's with rton now the Lowe's are spaced out extremely well on all of these devices keeping strong rank thanks to the game engine and the build itself from the developers that is the 490 outperforms the 390 TI by about 60% here here and the 3090 by a staggering 86% the RTX 380 is next in the list with the 6950 XT at 31 FPS average which allows the 490 and Improvement of about 144% int didn't make the cut for this game but it's in some of our other tests at 1440p the 490 jumped to 148 FPS average roughly doubling from its 4K frame rate previously that has it 61% ahead of the RTX 390 TI about the same as the 4K result they're scaling in Step so the jump over the 6950 XT would scale and step as well except for amd's architectural Advantage where the Improvement is reduced to 127% from the 144 previously and now we're at 1080p the 490 had its lead reduced to 29% over the 390 TI a clear Mark of a bottleneck elsewhere in the system everything tested thus far is playable with these settings at 1080p even without the support of dlss or FSR and there's just no benefit to a 490 here that's meaningful at this resolution f 2022 is back now this time with rate Trace settings remember that this chart cannot be compared to the previous F1 charts those do not use RT settings we get emails about this every time we publish reviews there are RT charts and non-rt charts they can't be compared the rate tracing result at 4K has the 490 running an incredible Gap over the 390 TI forming a depressing Rift for pre-priced drop buyers the Improvement is 75% over the 3090 TI which came out this year the 390 is next and generationally against the non-ti the Improvement is 105% the closest AMD card is currently the 6950 XT at 31 FPS average Intel's Arc makes an appear here as well but it's nearly fallen off the chart already at 17 FPS 1440p the 490 leads the 390 TI at 70% the lead against the 280 TI is 200% and that card wasn't too dissimilar of a launch price at $1,200 when factoring in inflation over the last few years even been ignoring inflation we'd be at a 33% price increase for MSRP for the 4090 for a 70% RT performance jump and not that the 2080 TI was ever on a recommended list anyway things matter last at 1080 so the 490 is only 27% ahead of the 390 TI here there's a loss of scaling we have an increase in CPU workload as well with rt and so we're seeing a CPU bind cyberpunk 2077 is up next and this was one of the most played RT games in our community questionnaire that we posted recently we used dlss and FSR for this one and then is using FSR at the moment instead of xss cuz it's not really ready we're using the quality settings at equal resolution scaling so in other words at 4K it's scaling up for 1440p at 4K the RTX 490 ran at 79 FPS average with dlss 2 the existing one this is without D lss3 testing because that's basically an alpha or a beta depending on how you look at it and the cyberpunk team isn't really ready with it either so that has the 490 again at about 76% ahead of the 390 TI it's about 170% ahead of the 6950 XT so nvidia's hardware and dlss combination pushes it pretty far ahead Intel's a770 technically is on the chart but let's move on at 1440p it runs at 136 FPS average to the 390 TI 80 FPS that's with the ultra settings and dlss for both so the 390 TI ends up allowing the 490 a big lead again the 390 though the non-ti wasn't too far back from the ti at 1080p we're hitting a CPU bind again increased work on the CPU the 490 still holds a significant lead but it's getting diminished here finally here's Tomb Raider this one has Ray tracing enabled as well we'll just flash through the results on the screen you can pause if you're interested this review is long enough as is but the 490 saw about the same percent scaling of the others uh 75% over the 390 TI once again at least until scaling got caught at 1080 okay so wrapping up then first of all DL ss3 was a lot of marketing talk and there was a good amount of in that presentation the keynote because kind of like RT we're not really going to see dss3 truly hitting the market uh with this launch there's some stuff out there not a lot it's also a lot of Alpha and beta and it doesn't work perfectly so just let's just factor that out right now and ignore it much like with the 20 series launch you ignore RT in most cases we're seeing 40 to 75% gains over something like a 3090 TI in 4k depends on the game some of them are last like Total War Warhammer but in a lot of them we're seeing those higher values now at lower resolutions this is truly pointless this this is like you can be on the best CPU and if you're not playing at 4K or 1440 with a relatively high graphics quality game you are going to be CPU bound so that's where you need to know your use case if you are not really pushing Graphics quality and resolution this is a waste of money for you plain and simple just like the Titans always have been as well at 4K that's where you start to see the actual gains and obviously with features like R tracing so uh whether this is actually worth it we have not really talked about that yet this is an expensive card it is functionally in that Titan class where historically you go this isn't a gaming GPU and that's true if you look at something like a 3090 a 3090 TI we've never said buy this one we've always said grab the one lower down from it so in those examples the lower end stuff launched first so like a 3080 or uh a 980 a 1080 stuff like that came out before these monster flagships did like this one Nvidia switched it this time and this came out first which we think probably means if we had the data for a 4080 today we'd probably be taking the same stance as previously where we go save the money buy the cheaper non Titan non TI class 10 a TI notwithstanding that was a mistake they'll never make again uh and move with one of those so right now we're not really comfortable making a full value judgment until we have the 4080 so we're we're kind of in that let's I'll just wait and see be patient see what happens with the 480s and with rdna 3 cuz that's going to be really interesting with multi chip but Nvidia is in a strong strong spot with the architecture generationally it's just the value comes into question when you launch only the absolute Flagship and traditionally that's not what you do as a company like Nvidia as for the thermal design it is actually excellent it's weird to say like we said earlier Founders edition cards reference cards we have never been in favor of the thermal design on those Nvidia really uh did kind of fix this one it's massive as a result but then so are the partner models and that begins the interesting discussion of can Partners truly compete with a company that has infinite resources as one of nvidia's thermal Engineers recently said uh to do things like fan simulations when the partners are just sort of picking from the supply that's out there on the market so that's going to be a really interesting fight to see we won't actually the answer to that until we get partner models in power so the transients were not bad from what we saw it's possible there's something hiding there that we didn't discover please keep us informed if you see something from someone we'll look into it but for now we didn't see any transients that were concerning the power consumption is high it's 450 to 500 watts and you're going to need to accommodate transients of at least 30 to 40% uh for pretty long durations of 500 micros seconds up to 11 100 mic 1.1 milliseconds so that's abnormal that's kind of new is that duration of transients so that's it for this one we'll talk more about value once we can see more of the market shape up but right now keep in mind the 30 series which Nvidia is happy to sell you as well has come down in price as have the 6,000 series from AMD those are seriously worth considering at the moment uh even if you buy it new in box and paying that slight premium over something used if you don't feel comfortable to used market and we don't blame you thanks for watching as always you can subscribe for more go to store. Gamers access.net to help us out directly by grabbing things like our mod mats to build your PCS on or you can go to patreon.com / Gamers Nexus to help fund our reporting check back for more we'll see you all next timeit's time for early RTX 390 TI buyers to feel really bad because the short version of the review of the RTX 490 is that we're seeing rasterization uplift at 4K anywhere between 40 and 75% versus a 3090 ti so a big jump there and although Nvidia isn't allowing anyone to review the partner model cards until tomorrow uh the founders Edition can be reviewed today thinking of course about EVGA recently saying that they felt like they were competing with Nvidia this kind of rings true there either way the founders Edition is in for testing rumors leading up to launch of the $1,600 card were largely centered around thermals and power which are closely related so we'll be looking at those in depth today with transient spikes uh in the scope for analysis as well as the usual gaming and R tracing performance tests let's get started before that this video is brought to you by be quiet and their silent Wings 4 fans the silent Wings 4 fans Market themselves as being useful on radiators Tower coolers and cases alike the fans have a six-pole motor and use a fluid Dynamic bearing which helps with the noise profile and with longevity the fans use anti vibration mounts for reduced the noise transfer to the case and have a rated lifespan of $300,000s learn more at the link in the description below so first that we captured too much data for this review we're actually going to run another video with pressure tests some flatness testing of the cold plate we'll looking at frequency curves uh noise and Acoustics and more in a separate piece but this one is is already pretty detailed for power thermals and gaming uh we have over 30 charts just for gaming in here about 30 anyway and remember that as we ramp into this the cost of Alternatives currently is more important than the original MSRP of things like a 6950 XT or like a 3090 ti so just quickly glancing at Amazon as we're writing this review we saw that an RTX 380 today new in box cost about $700 to $740 on average in the US the 3090 is about $980 or so right now just on Amazon quick search and the 390 TI is in the range of $1,100 to $1,300 the $490 at $1,600 obviously very expensive you're looking at a 60% increase in price over say a 3090 right now today new in box and Nvidia still and its Partners have too many 30 series cards so that's going to be true for at least a little while so that Nvidia is launching the RTX 490s Standalone without the 480s with it should give everyone pause on pulling the trigger at least moment parly the company is clearly desperate to dump R tx30 inventory so every play is a move to ensure that we don't end up with the market that's overloaded with both 40 series and 30 series jamming the channel and the reason we're bringing that up is because they have two more cards both called 480 coming up in probably about November and AMD also has rdna A3 coming up shortly as well so uh we'll want to see how close the 480s fall to this and things like rasterization because if you look at like the a750 versus the a77 they're so close the 770 might as well not exist for most people so that's all the sort of disclaimers out of the way now we can really get into the performance and talk about how this thing does because genuinely this is one of the largest performance uplifts we've seen generationally in many years at in multiple Generations uh 1080ti saw huge jump similar to this but upwards of 75% is a massive Spike over the 390 ti so there is real power behind this card the the power numbers that have been discussed those are fueling something and it is an increase in performance now it's time to look at it though and uh we'll start with thermals we are leading our review with thermal and power over here stone is working on setting up the transients right now so he's doing some transient testing to see what the spikes look like and how bad they are we'll come back to that and then on my end I've wired up the card with all of these little wires running out of it these are thermocouples that we have attached to key components like mosfets memory components and we're also monitoring the GPU with software and you can see all these cards here we've done a lot of this type of testing in the past so actually I'll I'll show you some of the other cards we've done this on this is actually one of them that we're working on right now that's an RK 770 and we're actually keeping this shot for additional climate control so we can keep it flat for ambient and for others we have tested many of these on the GPU shelf here so a lot of these have been wired in the past in much the same way but our methods have improved now and we're using better hardware for the monitoring and I'm excited to see what the results are so let's get back in there so we have I think it's six thermocouples hooked up right now we're monitoring the ambient actively every second of the test it's something we normally do and what we're doing at the moment is this is our ambient Thermo couple probe it's up and away from any heat sources how we do a lot of testing so it's a really steady 20.0 de C plus or minus about 0.5 and then I've got a bunch of these wired up to memory specifically and then these two are wired up to ver mosfet we're running Port Royal on Loop at the moment but that's not the only test I'm doing and it's also a custom Port Royal Loop so we've increased the settings to load the GPU specifically in different areas and all of this will go into the thermal charts but to show you some context for what this actually takes to wire up because it's a nightmare let's go over to what I call the Arts and Crafts bench so this is a huge project to wire up cards with thermal couples where I had some print outs from our tear down video so I could just keep it keep track of where the thermal pads are and where everything goes we have the exact pads brand new clean that Nvidia used so we have the supply of the pads we have the supply of the uh phase change thermal interface material so that we're able to replace everything as needed as we do these tests with an exact like for like that Nvidia is using so enough of all the setup let's get into some of the thermal charts all right so the first test is Port Royal as a workload with some modified settings on the graphic side but full stock settings on the Nvidia side in this one the GPU core seemed to follow a temperature Target of 70° C it fluctuated between 66 and 72 and the GPU hotspot temperature ran in the 80s also fine but pushing it if you're in more of a hot boox case the fan speed would increase to compensate so your noise would be higher for the memory modules that we manually probed we mapped the thermocouples along the edge of the module that bordered the hottest components nearby memory module R2 held steady state at about 65° C that's case temperature so for actual internal silicon of the memory it could be as much as about 10° higher according to research papers previously published within the thero couple industry even at 75 that'd be acceptable the bottom module on the PCI slot ran in the low 60s with memory L2 in the same area as for the mosfets on the left and the right of the board those are running in the upper 50s and the low 60s since fets can take around 100 to 125 C on average without big problems this is all completely acceptable they're in good range here we'll also plot our ambient line just to sort of be transparent and show you how controlled our lab is you can also use this to follow the dips and see how they're Associated here's the bar chart with two different test scenarios both of these were with auto fan settings however the fans spun at the same speed for each was 45.4% average for the blue line and 47.0 for the red line which is within usual fluctuation for fan curve furm Mark with stock settings put more load on the GPU and the board which makes sense and it ran hotter for the GPU core in memory the other components impressively maintain the same temperature like for the memory and for the vrm components and that was due to good thermal balancing across the board so the GPU temperature following that 66 to 70° C range that indicates that is the vbios or the firmware that Nvidia has programmed being told to basically set the fans along a known curve to hit that temperature Target and that's completely acceptable as GPU temperature for silicon the hot spot's getting a little warm if you put it in a a really bad poor air flow case you will be on the higher side it'll run louder and uh the best solution is buy better case but some other types of designs may do better with that for example an AIO solution now since Nvidia talk such big game when going over the prototypes for this cooler we went ham wiring it up for thermal testing it took about 6 hours of wiring and validation work plus four additional of testing but we wanted to challenge everything that Nvidia claimed and from what we can see thermally this card is proving itself if you look at the lines in that first line graph we plotted you can see how tightly together all the memory modules and the fets are for temperature and that's scattered across the board so when we wired this up it wasn't all clustered in one spot it was across the whole PCB and that means there's good thermal load balancing across the PCB and some of that comes from design changes this generation like putting in blockers those little North and South fins internally that stop air recirculation along with obviously the increase in fan size that's a huge part of it it has downsides like case fitment uh but that's part of the thermal Improvement and then obviously the fin stack being as much surface area as it is so purely speaking thermal ignoring everything to do with value with gaming performance R tracing Performance dss3 Marketing and whatever other that gets talked about relating to this card when it launches purely thermally speaking the founders Edition here seems to actually be well-designed that actually is critical because that is abnormal that's the first time we've ever said that about an AMD card or an envidia card and certainly we've never said that about an Intel card and this one is actually able to compete with partner so Nvidia is now getting to a point where it is able to compete with Partners which is also scary because it starts to look like maybe EVGA knew something and just got out at the right time if their partner is going to be competing with them this well so that's it for The Thermals for now and we'll come back to The a770 Thermals later now we're going to look at Power transients where Patrick stone is setting up in testing the 490 for transient spikes and previously we saw something like 2 2 and a half X and transients for something like a 390 master from gigabyte and that could be potentially bad to knock your system offline so I'll let Stone explain what it is transients are what we're testing for yes so when we talk about transient spikes what we're looking for on the hardware level is current and we want to see how much current is coming across these wires and uh to get a baseline for this particular test we're using firmar then we'll jump to video games as well um the transients uh on this card are going to be a little more interesting because instead of having three cables to work with or two cables to work with now we got four cables to work with and so we've got to jump around here and there to figure out which one's delivering the most current and then we'll we'll use that to capture our transient spikes um right now we're just looking at micros seconds we're looking for these frames of high microsc so we got it set on a 200 microc per Division and in fermar we're catching things that are like 600 micr seconds for power supply over current protection you're really not worried about it until you get into like the millisecond territory but um you know well that's why we're testing things to see if it gets there so we'll see what happens on the 4090 so as for what all that means for the end user the reason we look at Power transients is because if you get a really bad transient spikes and you either have a lowquality power supply or one that's just not quite up to the capacity needs to cover that Spike you can end up with the system shutting down while you're playing a game or something something so that's why we're looking at it let's get into the charts we ultimately decided to clamp all four cables using three clamps which is fine since the rated range of the clamps supports that division we lose a marginal amount of accuracy in the low single digit percentages but nothing that'll actually affect the results we divided the fourth Cable's 12vt lines evenly between the three clamps for this testing overall in our testing the power Behavior seems well controlled the maximum current being pulled by the card is consistently about 33 to 40% over the nominal power which is 450 Watts most critically these zoomed in scope shots show a change in Behavior looking at the millisecond level at most we typically see spikes inside of 100 microsc on cards with the 4090 however the graphs almost look inverted the longest transient Spike we observed was 1,110 micros seconds or about 1.1 milliseconds that's long enough to potentially knock a borderline or lower-end power supply offline and much longer than we typically see in Rainbow Six Siege on Doom we saw approximately 500 microc La transients this is different from what we typically see where this one since the max Spike we're seeing is about 40% over nominal and for a longer period it's contrary to the usual much shorter much spikier and way higher power Spike like with a 3090 master from gigabyte where we saw spikes upwards of something like 2 and 1 12x or something like that if memory serves we have video on it uh so this is different and rumors had indicated spikes upwards of 2.5 to 3x over nominal for the 4090 for this card and we're not seeing that we weren't able to reproduce it doesn't mean it doesn't happen we only ran this for about 8 hours for transient testing then we had to do all the other stuff so someone might find something more interesting but from what we've seen for eight hours of work on different games and applications it's completely acceptable for the transients so far it's just that duration might kick more power spies offline uh if they don't have the capacity to soak now some of the rumors did come from Power spy manufacturers and it's possible and maybe likely that since they had early insight to what the card would require Nvidia probably provided them the feedback hey it's going to be 2.5 3x transend they freaked out said no way we can't do it and then maybe there was some design revision and rework there in the time since uh also we know that the GPS for the 490 were on 3090 TI pcbs and the 3090 TI PC was a test platform for the 490 maybe it wasn't up to the task we also noticed that initial small spikes led to the power delivery circuit balancing the load as time progressed and it does so in a really unique way we saw the highest load shifting from one cable to the next effectively Distributing the transient heat load as well across the individual cables whether this was total chance or part of the design we're not sure we'd have to do a lot more testing but we do know that this cable this uh adapter cable is really important so it does have the four sense pins in it and one of the things that we observed was if you connect three of these we tried getting rid of one to see what would happen it will still boot however you'll lose the ability to boost the power Target so normally you can go from 100% to 133% Max with all four connected with only three it drops down to just 100% with two we weren't able to get it to boot and post image to the screen and that's because Nvidia is actively sensing how many cables are connected to this ad adapter so it is a smart adapter in that way this isn't just a dumb plug them in and then it just pulls the power it wants to pull from whatever's connected it knows what's on the other side that's important we noticed that the power connector responsible for the highest current draw across that varied from test to test which is uncommon typically we see a dominant connector then a secondary and a tertiary and so forth this one demonstrated Behavior almost of a like a feedback loop that helps to regulate the card and its power drop so you'll definitely want to use all four if you get one of these and as for the aftermarket options we haven't tested them yet we don't know if it'll still pull the full power if you're only running two even though technically the cables can handle it or what'll happen that'll be a future thing but there's a lot more power testing we could do here for now let's just look at a nominal load chart and then if anyone discovers anything really unique on launch day we'll continue to investigate it but for now transience of 33% to 40% seems to be about the norm now for the averaged power consumption this shows the actual power utilization over longer periods so rather than the transient spikes upwards of 630 WS we're looking at sustained power tested in firmar for a torture load the RTX 490 stock card pulls 496 Watts on average we observed 450 to 490 watts in some gaming scenarios but closer to that 500 Mark and fur Mark that aligns with the 390 TI results previously and remember that this is just the card power so slot plus pcie cables we don't measure the whole system in this as for the 33% power in increase for an overclock the RTX 490 pushes to that's unfortunate let's move on to gaming benchmarks in Total War Warhammer 3 at 4K and high settings the RTX 490 is entirely GPU bound which is good for testing and runs at 125 FPS average that has it outdoing the RTX 390 TI by 59% and the RTX 390 by 81% the lows are also improved significantly on the 490 however they're mostly proportional to the average so it's nothing special here if you bought 390 TI though for $2,000 early this year it's probably feeling pretty bad right now they don't normally launch them this close together the 6950 XT Trails both and the RTX 380 but will do proportionately better at 1080P and 1440p we've also got the 2080 TI on here where the 490 outperforms it by an increase of 172% these are some of the largest gains we've ever seen generationally so regardless of price that at least is impressive on its own we'll talk talk about the value though and the conclusion at 1440p the 490 isn't yet CPU bottleneck so it's still actually showing scaling here the card maintains a 64% improvement over the 390 TI and an 83% improvement over the 3090 similar to the 4K results the 2 136 FPS average result has it about 100 FPS or so over in average over the 6950 XT from Sapphire which is a 90% Improvement finally for the 1080p results the 490 gets cut down to a 54% lead over the 390 TI that much of a s downwards isn't because of architecture alone but also again because we're starting to hit that CPU limitation this game in F1 2022 at 4K the 490 ran at 230 FPS average and held a 70% lead over the 390 TI that's similar to what we saw in total Warhammer uh although a bit better for the 490 here so it's shaping up to be the average of 60 to 70% right now for a rasterization lead the Boost over the 390 non TI is a 100% jump or a doubling of frame rate performance but the 6 950x he lands between the two in this game and cuts the 490s lead down to 73% as compared to its 133 FPS average entry those are proportionately scaled on the 390 TI the 390 and the 6950 XT they're still good on the 490 but they're not proportionately scaled with the average frame rate this chart will show us frame time consistency it's the most empirical way to look at the user experience as it shows the frame to frame time required in milliseconds to render each of those frames which just shown as one point on the line the RTX 490 mostly clusters around 4 to 5 milliseconds but regularly spikes to 9 milliseconds these aren't noticeable to the player typically we see that people observe 8 to 12 millisecond jumps as potential stutters but they are enough to throw that 1% out of proportion that we saw earlier there's a single Spike to 26 milliseconds that is noticeable to the player as a slight hitch or stutter but we normally have to see multiple of these in a test period to actually become concerned about them for now it's something to pay attention to as we iterate through the next charts the 6950 XT plots just as kind of a baseline for another device this one is mostly consistent throughout the test with fewer excursions though it does have a few excursions of its own moving to 1440p now the RTX 490 is approaching the CPU bind which is in the 290s for FPS average that means we're CPU bound on and off throughout this test but not for the entirety of it the 490 holds a 27% lead still over the 6950 XD but that's cut down from 73% previously some of that is AMD gaining at these lower resolutions but most of that is a CPU bind we won't learn much new at 1080p other than a reminder that the 6950 XE and the 6900 XE have an advantage over the 390 TI at this resolution Shadow the tomb Raiders up now at 4K the 490 maintains a 70% lead over the RTX 390 TI it's consistent with the previous results the 6950 XT and the 390 are both similarly ranked and if you were on an RTX 280 TI previously you'd see a 164% increase in frame rate moving to the 4090 almost leaked a new card there it's just called the 490 it's not the ti let's all move along before anybody from Nvidia hears that this is where we jump to a regular reminder that if you're happy with your computer today there's no need to upgrade right now even though something like this is shiny new and impressive because if you look at that chart we just saw everything on there is in playable territory you'll do just fine the frame rates are getting ridiculous at this point with these types of cards until you enable features like RT so it is something to really think about out do you actually need to upgrade and especially do you need to go this high end if there's stuff like 307s out there for cheaper now than they used to be 1440p reduces the 490 knot TI which doesn't exist lead over the 390 Ti from 70% to 27% a clear CPU cap the 6950 XT also encroaches on the 490 and it now outranks the 390 TI as well as we're bound here we'll move on but the value is clearly dismal if you're playing under a constraint like this and buying a 490 just does not make any sense just to show it since we have it here's the 1080p chart it's also bound obviously and everything above the 380 Mark is incomparable and useless because it's being limited by the CPU but gives you some perspective of where your limitations might lie Rainbow Six Siege at 4K gives us an older dx11 title to look at but one that's still popular this game has the 490 at 355 FPS average leading the 3090 TI by only about 47% only there compared to the previous results it's not spectacular for the price considering the 390 TI is currently closer to like 900 to 1,000 bucks and the 490 outperforms the 6950 XT by 86% here in average FPS which sets a mark to watch for lower resolutions at 1440p it's still not CPU bound the 490 leads the 390 TI by 41% consistent with the 4K result previously the gains in Rainbow Six Siege are just that much worse than in other games which shows that the pi and the game itself May dictate some of your scaling in a way that can wipe out most of the value of the 490 and again we won't spend much time on 1080p it's CPU bound now it did climb another 50 FPS first before hitting the CPU bind with these graphic settings uh but it's bound nonetheless the gap between the 6950 XT and the 390 TI has shrunken here although the 390 TI carries better lows than the AMD Hardware not that it is noticeable at this frame rate anyway and if you want better value you buy something further down the stack here's another game where the scaling isn't as impressive as those earlier 70% numbers in Horizon zero Dawn at 4K we saw a limitation scaling at 53% gain over the 39 DTI and were're not CPU bound that's uncharacteristic of the earlier results it's close but it's just not quite there 1440p shows us the limit at 180 FPS average with the 6950 XT also bouncing off of that limit the 6900xt and the 390 TI are close enough that some frames are likely getting capped but not all of them so these are all BAS basically the same clearly you should not buy a 4090 if you're going to play in a scenario like this it's waste of money 1080p is fully capped same thing nothing new to see here 490 looks like it's actually doing worse than the 390 in this one in reality it isn't it's just that they're all limited that we can't see scaling they're tripping over their own frames as the CPU tries to keep up and you end up with a stack that you can't differentiate Final Fantasy at 4K is up now and this one maybe shockingly we were still GPU bound the game is so easy to run that we thought we might actually be totally CPU limited even at 4K but it's still GPU targeted at high resolutions the RTX 490 ran at 214 FPS average which positions it 75% ahead of the 390 TI's 122 FPS average that's better scaling than we've seen elsewhere maybe partly thanks to dx11 Nvidia has always done disproportionately well on dx11 since that API requires more driver work to run well the lead over the 6950 XT is 88% with the 3090 getting doubled cleanly by the 490 at 1440p we're hitting the CPU buying the 490 runs at 235 FPS average which is functionally the same as the 220 result we had on the 390 ti so there's no real difference here and we'll skip 1080 this time it's the same we also ran GTA 5 at all three resolutions but only 4K matters since the engine is capped at 187.5 FPS average makes basically everything the same the RTX 490 ran at the full cap here dropping almost zero frames you have to realize we've been running this Benchmark since 2015 now on gpus we've never seen a card to hold base basically the engine cap like this the lead over the 39 TI is only 38% so in that regard it's not worth it the value doesn't scale like in other games because we're artificially limited on the engine the CPU isn't capped either just to be clear it's only Rage that's capped and not mine but like the name of the game engine my rage is uncapped there's no cap No Cap okay no capping now we get into the really complicated testing which is Ray tracing and dlss or FSR XS test and those three sets of initialisms are why this has gotten so complex so not only do we have to test for just pure rasterization we also need to test for rate tracing but then because features like dlss are so prominent now we have to enable those and there's different ones for each vendor so this has become a massive Matrix of really messy tests to run we've done our best to distill it down into a few sets of games to give you a wide picture performance let's take a look we'll start with control which came back as one of the most played RT games when he pulled our community you all recently on YouTube Community posts cyberp Punk was one of the other most popular we'll test that too starting with 4K and without dlss the RTX 490 ran at 76 FPS average here that's with rton now the Lowe's are spaced out extremely well on all of these devices keeping strong rank thanks to the game engine and the build itself from the developers that is the 490 outperforms the 390 TI by about 60% here here and the 3090 by a staggering 86% the RTX 380 is next in the list with the 6950 XT at 31 FPS average which allows the 490 and Improvement of about 144% int didn't make the cut for this game but it's in some of our other tests at 1440p the 490 jumped to 148 FPS average roughly doubling from its 4K frame rate previously that has it 61% ahead of the RTX 390 TI about the same as the 4K result they're scaling in Step so the jump over the 6950 XT would scale and step as well except for amd's architectural Advantage where the Improvement is reduced to 127% from the 144 previously and now we're at 1080p the 490 had its lead reduced to 29% over the 390 TI a clear Mark of a bottleneck elsewhere in the system everything tested thus far is playable with these settings at 1080p even without the support of dlss or FSR and there's just no benefit to a 490 here that's meaningful at this resolution f 2022 is back now this time with rate Trace settings remember that this chart cannot be compared to the previous F1 charts those do not use RT settings we get emails about this every time we publish reviews there are RT charts and non-rt charts they can't be compared the rate tracing result at 4K has the 490 running an incredible Gap over the 390 TI forming a depressing Rift for pre-priced drop buyers the Improvement is 75% over the 3090 TI which came out this year the 390 is next and generationally against the non-ti the Improvement is 105% the closest AMD card is currently the 6950 XT at 31 FPS average Intel's Arc makes an appear here as well but it's nearly fallen off the chart already at 17 FPS 1440p the 490 leads the 390 TI at 70% the lead against the 280 TI is 200% and that card wasn't too dissimilar of a launch price at $1,200 when factoring in inflation over the last few years even been ignoring inflation we'd be at a 33% price increase for MSRP for the 4090 for a 70% RT performance jump and not that the 2080 TI was ever on a recommended list anyway things matter last at 1080 so the 490 is only 27% ahead of the 390 TI here there's a loss of scaling we have an increase in CPU workload as well with rt and so we're seeing a CPU bind cyberpunk 2077 is up next and this was one of the most played RT games in our community questionnaire that we posted recently we used dlss and FSR for this one and then is using FSR at the moment instead of xss cuz it's not really ready we're using the quality settings at equal resolution scaling so in other words at 4K it's scaling up for 1440p at 4K the RTX 490 ran at 79 FPS average with dlss 2 the existing one this is without D lss3 testing because that's basically an alpha or a beta depending on how you look at it and the cyberpunk team isn't really ready with it either so that has the 490 again at about 76% ahead of the 390 TI it's about 170% ahead of the 6950 XT so nvidia's hardware and dlss combination pushes it pretty far ahead Intel's a770 technically is on the chart but let's move on at 1440p it runs at 136 FPS average to the 390 TI 80 FPS that's with the ultra settings and dlss for both so the 390 TI ends up allowing the 490 a big lead again the 390 though the non-ti wasn't too far back from the ti at 1080p we're hitting a CPU bind again increased work on the CPU the 490 still holds a significant lead but it's getting diminished here finally here's Tomb Raider this one has Ray tracing enabled as well we'll just flash through the results on the screen you can pause if you're interested this review is long enough as is but the 490 saw about the same percent scaling of the others uh 75% over the 390 TI once again at least until scaling got caught at 1080 okay so wrapping up then first of all DL ss3 was a lot of marketing talk and there was a good amount of in that presentation the keynote because kind of like RT we're not really going to see dss3 truly hitting the market uh with this launch there's some stuff out there not a lot it's also a lot of Alpha and beta and it doesn't work perfectly so just let's just factor that out right now and ignore it much like with the 20 series launch you ignore RT in most cases we're seeing 40 to 75% gains over something like a 3090 TI in 4k depends on the game some of them are last like Total War Warhammer but in a lot of them we're seeing those higher values now at lower resolutions this is truly pointless this this is like you can be on the best CPU and if you're not playing at 4K or 1440 with a relatively high graphics quality game you are going to be CPU bound so that's where you need to know your use case if you are not really pushing Graphics quality and resolution this is a waste of money for you plain and simple just like the Titans always have been as well at 4K that's where you start to see the actual gains and obviously with features like R tracing so uh whether this is actually worth it we have not really talked about that yet this is an expensive card it is functionally in that Titan class where historically you go this isn't a gaming GPU and that's true if you look at something like a 3090 a 3090 TI we've never said buy this one we've always said grab the one lower down from it so in those examples the lower end stuff launched first so like a 3080 or uh a 980 a 1080 stuff like that came out before these monster flagships did like this one Nvidia switched it this time and this came out first which we think probably means if we had the data for a 4080 today we'd probably be taking the same stance as previously where we go save the money buy the cheaper non Titan non TI class 10 a TI notwithstanding that was a mistake they'll never make again uh and move with one of those so right now we're not really comfortable making a full value judgment until we have the 4080 so we're we're kind of in that let's I'll just wait and see be patient see what happens with the 480s and with rdna 3 cuz that's going to be really interesting with multi chip but Nvidia is in a strong strong spot with the architecture generationally it's just the value comes into question when you launch only the absolute Flagship and traditionally that's not what you do as a company like Nvidia as for the thermal design it is actually excellent it's weird to say like we said earlier Founders edition cards reference cards we have never been in favor of the thermal design on those Nvidia really uh did kind of fix this one it's massive as a result but then so are the partner models and that begins the interesting discussion of can Partners truly compete with a company that has infinite resources as one of nvidia's thermal Engineers recently said uh to do things like fan simulations when the partners are just sort of picking from the supply that's out there on the market so that's going to be a really interesting fight to see we won't actually the answer to that until we get partner models in power so the transients were not bad from what we saw it's possible there's something hiding there that we didn't discover please keep us informed if you see something from someone we'll look into it but for now we didn't see any transients that were concerning the power consumption is high it's 450 to 500 watts and you're going to need to accommodate transients of at least 30 to 40% uh for pretty long durations of 500 micros seconds up to 11 100 mic 1.1 milliseconds so that's abnormal that's kind of new is that duration of transients so that's it for this one we'll talk more about value once we can see more of the market shape up but right now keep in mind the 30 series which Nvidia is happy to sell you as well has come down in price as have the 6,000 series from AMD those are seriously worth considering at the moment uh even if you buy it new in box and paying that slight premium over something used if you don't feel comfortable to used market and we don't blame you thanks for watching as always you can subscribe for more go to store. Gamers access.net to help us out directly by grabbing things like our mod mats to build your PCS on or you can go to patreon.com / Gamers Nexus to help fund our reporting check back for more we'll see you all next time\n"