**AMD's Radeon RX 6900 XT Review**

The wait is over for AMD's latest flagship graphics card, the Radeon RX 6900 XT. We've been testing this powerhouse of a GPU to see if it lives up to its lofty promises and delivers on its core strengths.

In our tests, we found that the Radeon RX 6900 XT is indeed a beast of a card, capable of delivering impressive performance in a wide range of games and applications. With its high clock speeds and large number of cores, this GPU can handle even the most demanding workloads with ease.

However, one thing that stands out about the Radeon RX 6900 XT is its ability to overclock like a champ. In our testing, we were able to push the card's performance to new heights by tweaking its settings and clock speeds. This level of flexibility makes the RX 6900 XT an attractive option for enthusiasts who want to squeeze every last bit of performance out of their GPU.

But what about power consumption? We tested the Radeon RX 6900 XT in a variety of workloads, including Total War: Three Kingdoms, which is a demanding game that pushes even the most powerful GPUs to their limits. In our tests, we found that the card's power draw was relatively moderate, with an average of around 200 watts under load.

Interestingly, when we overclocked the Radeon RX 6900 XT, its power consumption increased significantly, but only linearly up until a certain point. Beyond that, it began to drop off in terms of performance gains for every additional watt consumed. This is a familiar trend in the world of GPU testing, and it suggests that there may be limits to how much this card can be pushed before it begins to slow down.

Despite these limitations, we were impressed by the Radeon RX 6900 XT's ability to deliver consistently high frame rates across a wide range of resolutions. In particular, at 4K resolution, this GPU was able to outperform its rival, the NVIDIA GeForce RTX 3080, in many games. However, it's worth noting that the RTX 3080 does have an edge over the RX 6900 XT when it comes to real-time ray tracing and other advanced features.

**Comparison to Other GPUs**

So how does the Radeon RX 6900 XT compare to other high-end GPUs on the market? In our testing, we found that it was able to deliver strong performance in games like Assassin's Creed Odyssey and Cyberpunk 2077. However, when it came to more specialized applications like Blender, the GPU fell short of its rival, the NVIDIA GeForce RTX 3080.

In terms of value for money, the Radeon RX 6900 XT is certainly a compelling option. With its strong performance and reasonable power consumption, this GPU offers great bang for your buck. However, it's worth noting that the price point for this card may be higher than some users are willing to pay.

**Conclusion**

Overall, we were impressed by the Radeon RX 6900 XT's performance and features. While it may not have the same level of power as its rival, the NVIDIA GeForce RTX 3080, it offers a compelling alternative for those who want a high-end GPU without breaking the bank. With its strong overclocking capabilities and consistent frame rates across a wide range of resolutions, this GPU is certainly worth considering for anyone in the market for a new graphics card.

**Upcoming AMD News**

AMD's upcoming Radeon RX 6000 series will feature significant improvements over its predecessors, including faster clock speeds and more cores. We'll be keeping a close eye on these developments as they become available, so stay tuned for our in-depth coverage of the latest AMD news.

**Overclocking the Radeon RX 6900 XT**

We were able to overclock the Radeon RX 6900 XT to impressive levels, pushing its performance beyond what we thought was possible. However, this level of overclocking is dependent on a variety of factors, including the specific GPU model and the settings used in our testing.

The most impressive part of the RX 6900 XT's cooling system was its mechanical mount, which proved to be one of the best we've seen in terms of pressure testing. This design helps to keep the GPU running cool even at high temperatures, making it an attractive option for overclockers and enthusiasts who want a reliable build.

**Value vs. Performance**

One thing that stands out about the Radeon RX 6900 XT is its ability to deliver strong performance without breaking the bank. While it may not be as powerful as some of its rivals, this GPU offers great value for money, especially when compared to NVIDIA's GeForce RTX 3080 series.

However, if you're looking for a specific application or use case where the RX 6900 XT can shine, then it may be worth considering. For example, in games like Assassin's Creed Odyssey and Cyberpunk 2077, this GPU offers strong performance at high resolutions. Additionally, in applications like Blender, the RX 6900 XT can deliver impressive results when paired with a compatible CPU.

**Final Verdict**

Overall, we're impressed by the Radeon RX 6900 XT's performance, features, and value for money. While it may not be perfect, this GPU offers a compelling alternative to NVIDIA's GeForce RTX 3080 series, especially for those who want a high-end GPU without breaking the bank.

"WEBVTTKind: captionsLanguage: entoday we're reviewing the nvidia rtx 3070 founders edition card that just hit the market this has a 500 msrp there are partner models coming out shortly as well partners of course being people like evga aces gigabyte so forth those will be out in about two days from the time of this review going up that's when on videos allowing the release of partner model cards and we'll be reviewing those separately but today we're focusing on the founders edition card and testing will include thermals power frequency overclocking gaming mechanical assembly so like pressure testing stuff like that but we're also going to have a separate teardown that focuses just explicitly on taking the card apart looking at assembly quality ease of access if you need to repair something and cooler design from a mechanical assembly side of things so check back in a few hours on this channel for the tear down video because that'll be going up the same day as this one but for this we're going to be focusing on performance versus 3080 fe for a point of comparison amd has rdna 2 coming out soon but we can't test that today obviously because we don't have it so you'll have to check back during our dna2 announcements to see if we have any additional opinions on this card once that one is formally announced which is tomorrow at the time of this going up before that this video is brought to you by us and our gamers nexus wireframe mousemat aside from being the best way to directly support our lawn form investigative reporting you can also get a custom made high quality mouse mat made with a high detail 3d design that we created to show off heat sinks coolers video cards and more the mouse mat uses a stitched blue border for added longevity a blue rubber underside for unique flair and a microfiber cloth for smooth tracking the mat is 36 inches by 12 inches and fits a keyboard and mouse easily to back order your mouse mat and ensure you get one in the next run go to store.gamersnexus.net and backorder yours while reducing our reliance on advertisers or click the link in the description below it's kind of weird to review a video card right now at least one like this where at this point nvidia is sort of just one upping itself while amd is preparing its next gpu launches that means it's a little bit difficult to do the usual comparisons because a lot of the time in the charts we spend a good deal of effort analyzing the performance differences between the closest competitor from amd versus nvidia and vice versa in this scenario though it's really just nvidia versus 20 series versus 10 series nvidia versus the 3080 to see what kind of performance difference there is there so it's a little different but uh it's still worth testing obviously this card at 500 comes in about 200 cheaper than the floor price of the 3080s for the most part partner models you should be attacking on anywhere from 20 to 80 depending on how high end you're going where the the highest of the high end mostly affords you additional features like overclocking support or perhaps water cooling water blocks hybrid cars stuff like that 4k is also going to be a point of interest here over the years there have been many cards that were fine for 4k gaming the 1080 ti was in fact one of them but obviously the goal posts move as the gaming industry evolves and the graphics options become more complex so 4k has been kind of a difficult target to hit for video cards over the years this one does overall fine with approaching 4k gaming as you'll see in our test today but it's not the first one to have done it and the 3080 has gotten a lot more attention as the true world's first like fully 4k ready always 4k playable type of card this however does threaten that a little bit let's look at the specs the rtx 3070 runs 5888 cuda cores versus 10496 on the 3090 or 8704 on the 3080. despite the 3080 being a 47ish percent increase in cuda cores there's obviously a lot more to it than that in terms of determining gaming performance we won't see linear scaling and we never do frequency also comes into play which we'll show soon as does memory the biggest change to the 3070 is that it's moving to gddr6 from g6x so it's bumping back down to memory that's easier to cool but also significantly slower at 16 gigabits per second stock versus 19 gigabits per second on the 3080 which also has 10 gigabytes of memory as another benefit as a result memory bandwidth is down to 512 gigabytes per second on the 3070 from 760 gigabytes per second previously on the 3080 the bus width has changed from 320 bit to 256 bit and another major update aside from the rt cores is in rops it's going from 88 robs to 64. now this should be much more noticeable in 4k than say 1080p test as the resolution goes up the 3080 with the rob's benefit memory bandwidth benefit should start to pull away from the 3070 at say 4k resolution we've got a block diagram we can show of the 70 die layout as well and in our tear down or we can drop in a b-roll clip here we've gotten die shots of it where it is a physically different piece of silicon the mounting mechanism is actually very interesting we'll talk about that in the tear down but it's a different piece of silicon it's smaller and so it's still amp here but it's a cut down version of amp here anyway we've got a ton of testing to go through today so let's get into it we'll open up with something different that you won't see in the other 3070 fe reviews just to keep things fresh our teardown of this card going live a few hours after this video shows just how tightly put together the 3070 fe card really is there aren't that many screws this time which is good but they're evenly spaced with only four that secure the cooler to the pcb the remainder of them secure the unibody shroud to the back plate we initially had concerns about this regarding how much pressure is applied but in this pseudo-color image produced using a nist traceable pressure sensing piece of equipment that gn paid for thanks to its patreon backers and store.gamersnexus.net supporters we're able to see just how much that unibody shroud matters although it isn't leveraged directly as a piece of the cooler but it could be because it's aluminum it does secure to the back plate in a way that it's drastically changing the pressure application to gpu silicon in this pseudo color image without the shroud we can see a major gap in coverage towards one side of the die while the unshrouded 3070 fe has the most even contact we've yet seen with this new testing equipment this is basically perfect or very close to it we'll put some of our previous measurements on screen as well like the other founders edition cards and one of the partner models as shown here the 3070 fe with the shroud has even contact and has good tension across the entire pcb at least as far as gpu silicon is concerned that's off to a good start but pressure alone doesn't make a good cooler the pressure map looks good but it isn't a height map this chart shows the depth from a known zero point on a special needle for measuring depth on things like coolers more consistent is best so we want to see a smaller box with measurements clustered tightly together the 30-70 feet does fine it's not as consistent and point-to-point measurement as the 2080ti strix or the gigabyte eagle but we also didn't find any significant outliers in the measurement that blew out the scale unlike the gigabyte eagle the 3070 fe seems fine overall in this department so off to the other tests we'll start with the frequency validation plot this looks at temperature and frequency over time which is helpful in illustrating behavior of gpu boosting tables the rtx 3070 founder's edition card started at about 1965 megahertz when it was under 60 degrees celsius at the very beginning of the test after this it fell slowly to 1890 megahertz to 1920 megahertz went at steady state thermals of about 72 degrees celsius in this workload as a reminder this is with full auto settings so it represents out of the box usage and out of the box thermals the v-bios temperature target seems to be about 72 in this 3dmark frozen frame workload which is representative of gaming scenarios our power virus testing later will plot higher frequency is relatively stable but far from the flatline behavior of water cooled cards that have eliminated thermals as a consideration or cards which have eliminated power from consideration by way of mods as expected we're losing about 45 megahertz to temperature as this card warms up to steady state this is where board partner models will often show their value for perspective this plot will show the frequency of some other cards that we've tested recently we've intentionally zoomed the scale so that it's easier to read where each one falls you'll first see that the 30 70 fe line plots exactly where it did previously as the data hasn't changed the scale's just been zoomed in the rtx 3080 eagle by gigabyte plots around 2000 megahertz and this helps us better understand one of the many aspects of why the 3080 will outperform the 3070. there's clearly hardware level differences like fp32 capabilities memory memory bandwidth but frequency also counts for performance deltas the 3080fe is much closer to the 3070 fe at about 1925 to 1950 megahertz the asus 3080 tough cards also stick around the 2000 megahertz marker and the rtx 3090 ftw3 runs about 1940 to 1950 megahertz the highest end nvidia cards like the 3090 typically run slower than the slim's down counterparts like the 3080s here that's the nature of balancing stability against more hardware level physical advantages like more fp32 or other large die gpu benefits time to move on to thermals we'll start with 40 dba noise normalized thermals normalizing for noise helps us understand the efficiency of a cooler when it's disallowed from blasting its fan speeds to compensate for cooler size or design we haven't tested other 3070s yet obviously because we're not allowed to right now but we can at least show behavior compared to the other rtx 30 cards that we've previously tested obviously these should ideally be power normalized as well for pure cooler comparison but since we're comparing across gpu skus it doesn't really matter here's the gpu chart for noise normalized thermals we'll look at memory next the rtx 3070 fe at 227 watts power for total board power as validated with physical interposers plots at 65 degrees celsius in a 21 degree ambient environment this is with the card set to 40 dba at a 20 inch distance in a room with a noise floor of 26 db that has it about 8 to 9 degrees warmer than the 3080 tough by asus which held 56 degrees celsius at 340 watts without even looking at other rtx 3070 cards this allows us to demonstrate how partner models can excel the tough is handling 120 watts more power in this power virus workload anyway and yet manages better results now of course a smaller die would also mean less surface area and a higher thermal density but the 3070 fe won't be the best cooler for the rgx 3070 series either that said it's significantly better than what we've seen in the past for reference coolers from nvidia and nvidia has absolutely improved in its cooler design and thermal performance on the fe cards credit where it's deserved but there's more ground to gain on partners yet this chart shows the noise normalized performance for board components as tested with thermocouples that we manually placed all over the board for the 3070 fe the bottom memory module measured at 53 degrees on the flip chip pga side of the silicon which is well within spec and partially thanks to the drop from gddr6x just to gdr6 the top module that we measured ran about 60 degrees celsius when noise normalized but keep in mind that we're running the fans much faster than auto here we'll look at that next the vrm moss measurement ran about 72 degrees celsius on the i o side of the board which is getting into warm territory but still plenty far from the 100c plus capabilities of the mosfets this chart shows the same test except with the card auto controlled in this one the memory runs at 61 degrees and 69 degrees celsius on the rtx 3070 fe with the internal silicon number between 5c and 10c higher than that exterior measurement that we've taken that's still fine but it's moving to the warm end of the scale the mosfet measurement is more concerning however at 82 degrees externally it's potentially in the low 90s for the internal silicon although we can't get the actual mosfet measurements because nvidia has decided to not expose them through software anywhere we're approaching territory where an extra 10 degrees internal case ambient temperature could mean those vrm components are running borderline acceptable at least for long term use this is the nature of condensing the pcb down as tiny as nvidia did coupled with the component choices for the vrm and with the decision to run these fans so quietly it's expounded we think nvidia should sacrifice some of the noise for an extra 2 dba worth of fan speed which would alleviate a lot of the long-term concerns with the cards left to self-regulate their performance behaviors the rtx 3070 fe held around 33 dba making it one of the quietest out-of-box cards on our rtx 30 benches when they're all left to self-regulate it's also among the lowest power consumption thus far so this all makes sense the 3070 holds around 75 degrees celsius when left to fully control its own fan speeds in this power virus workload so vbio seems to have established this temperature target and a fan curve that holds around 1480 rpm auto in this benchmark remember that there are driver level changes in behavior between things like firmark and 3dmark so that's why we'll see some additional differences in behavior between each workload the card runs 5 degrees warmer than the 40dba ftw3 at 400 watts but because the noise isn't normalized here you'd have to look at our previous chart for a real efficiency comparison this chart shows a weak point for the 3070 founders edition it's showing the v bios programming managed against the gaming workload applied which is useful for analyzing hysteresis and responsiveness interestingly the 3070 fe has a much more abrupt fan behavior than we've seen in most other carts rather than ramping gradually and at a steady pace the 3070 fe first ramps to about 1000 rpm under initial load on a lag by the way of about 30 seconds from initial load start and then it tries to hold this rpm while load continues to ramp it holds here for another 40 to 45 seconds and then ramps sharply to 1300 rpm attempts to hold and then ramps sharply again to roughly where it rests in the 1480 rpm range the end result is that the fan's initial spikes are more noticeable to a human and therefore more annoying if rather than attempting to hold that 1000 rpm first it had instead continued a gradual ramp the noise change would have been less noticeable from a human factor's perspective nvidia could improve on its fan programming for v-bios on the 30-70 fe namely by changing the lag for ramp this chart shows the noise levels as plotted against fan rpm measured with the same methodology discussed previously the rtx 3070 fe has a fairly straight increase in noise with each five percent increment in pwm signal the maximum noise level stops at about 56.9 dba at 20 inches although it'll never reach this without manual input from the user software allows this card to be controlled in one percent increments all the way down to zero percent which is mostly unheard of and that's for good reason because it's broken that's below what pwm can actually support and so the card has a hard stop at 30 and 1000 rpm for actual change although it can be technically adjusted in software down to 20 percent one percent or anywhere between it'll just still spin at thirty percent speed normally you see a hard cut off at thirty percent where software will not allow you to drag it down lower uh but this is a mix of software bugs and things that will be ironed out later time to get into the game benchmarks for the 30-70 as a quick reminder you can always calculate improvement against whatever gpu you're using by looking at the aggregate for old benchmarks where you might see the 970 versus the 1070 and say the later years of the 1070. by doing that you can look at the delta the average percentage improvement between whatever gpu you have and something that is on these charts by referencing older data and then you can figure out the delta between your gpu and what is on these charts so if yours isn't here you're still able to calculate where it should end up falling we'll start our game benchmarks with red dead redemption 2 at 1080p keep in mind that many tests become somewhat bound at the top end and don't show as dramatic scaling as you'd see at 1440 or 4k in this scenario tested first with dx12 and high-end custom settings the nvidia rtx 3070 fe stock card ran at 123 fps average with its oc providing just 3.2 percent uplift in average fps and fortunately no loss in the lows stock to stock the 3070 fe leads the 2080 super by 12.4 percent with 5700 xt and 2070 super placement led by 23 the 1080 ti ftw 3 was led by the 30 70 by 37 percent this definitely isn't just another 1080 ti in the form of a 30 70. it's actually improved although maybe nvidia will do that for the 3060 ti or whatever that's called the 3080 fe leads the 3070 fe by 24 at 1080p and that's when both are stuck let's move on to 1440p we can start by comparing those previous percentages at 1080p the 3070 fe led the 2080 super by 12.4 percent but now it's leading by 19 the lead over the 1080 ti of tw3 expands from 37 to 45 percent while the 380 fes lead of 24 at 1080p stays about the same maintaining 24 here as well the 30 70 fe plants itself just under the 2080 ti and just ahead of the 2080 super oc compared to the 1070 from a few years ago the gains are about 128 percent with one percent and 0.1 lows improving significantly as well tested at 4k the 3070 fe still maintains above 60 fps average with lows reasonably spaced behind and without any major stutters we didn't run into any memory limitation issues here as expected because we're hitting other limits first comparing the previous percentages the 380 fe now jumps to 34 percent the head up from a previous 24 stretching its legs a bit more with the memory bandwidth and drops advantages the 3070 fe holds an 18 lead over the 2080 super staying about the same distance as in the 1440p results the 5700 xt for an amd reference is led by the 30 70 fe by 34 percent as is the 2070 super and the 1080 ti with an overclock they're all about the same placement that's up from the 1080p results as well and so modern changes in memory subsystems and the architecture have more notably benefited these higher resolution tests again the rops are another notable contributor here if you're curious about vulkan performance this chart shows dx12 versus vulcan scaling in red dead 2. this helps to establish how much the api matters on one generation versus another we've previously found that vulkan benefits the rtx 30 series by about 10 percent for the cards out so far and that mostly persists with the 3070 at 9.2 percent ahead of its dx12 results some of the baseline performance may change because of differences in what's being rendered between the two apis but the end result is that the 30 series handles the switch over better than the 10 series cards before it the 20 series isn't too different in the 2080 super nti class and the 5700 xt for reference was about 1.4 percent better in vulcan and so the choice should likely hinge more on whichever one is the least buggy at any given time and that comes down to the game's developers rainbow six siege is next we'll start at 4k this time just because rainbow six siege is so relatively low in gpu load that it's more realistically usable the 3070 fe ran a 144 fps average talk result with lows at 123 and 118. that puts it tied with the 2080 ti strix when stock but allows the strix oc a lead of 12 percent our 3070 feoc although stable elsewhere proved problematic here and posted one percent worse results than stock for that reason we've removed the data from the chart to reduce confusion we're not able to get a worthwhile oc dialed in for this game within the review time window unfortunately and let's have to try in a few different settings and combinations the 3080 beats the 3070 at 179 fps average or 24 stock the stock the 3070 leads the 1070 by 137 it leads the 1080 ftw by also about double and the 1070 ti would be alongside the 1080 here the 2060 ko is also led by about double in the frame rate and improved in frame time consistency by moving to the 30 70. the 2070 super is led by just over 40 percent for averages and that's a significant generational gain when considering the pricing and considering that the super was a refresh here's a frame time plot to better look at the frame to frame intervals for each rendered frame we're talking a lot about percentages but that's all calculated against an average this is the closest we can get to an empirical look at gaming performance for each frame because we can see each frame's render time requirement in the on present metric lower is better and more consistent is best in this test the 3070 stock runs about 7 milliseconds frame to frame intervals on average and is typically plus or minus one millisecond from one frame to the next the largest gap we see is three milliseconds in general and as long as the frames don't deviate or experience an excursion from the mean of more than eight to twelve milliseconds frame to frame at least then no one will ever notice that difference below eight milliseconds especially here we're more talking one millisecond frame to frame is not perceptible to a human the 3080 fe stock card ran around five milliseconds with slightly wider deviation but also not one that ever be noticed at 1440p the rtx 3080's lead over the 3070 is reduced to 19 from 24 as the bandwidth benefit is reduced here the 3070 stock fe leads the 2080 ti and the 2080 super oc although to be fair the 2080 super oc did a lot for this game just like the 2080 tioc did it's unfortunate that our 3070 overclock couldn't hold as well in this game as others but if we had more time within the review window we could have eventually tuned it to work the improvement over the 1070 is reduced to 120 percent from 137 percent again posting higher scaling as resolution increases and we're not cpu bound here so it's bandwidth and rops and architectural changes like fp32 capabilities it's sort of odd though to review a gpu in general right now just because there's no direct competitor from amd nvidia isn't having any trouble outmatching its last gen products so it's almost a little bit more on the boring side than it might be otherwise until at least amd gets our dna 2 out at 1080p the 3070 fe manages 385 fps average so it's showing playable scaling all the way from 1080 to 4k the 3080s lead is reduced to 14 while the 30 70's lead over the 1070 is reduced to 94 nvidia's investment in bandwidth for this generation shows it's still a significant improvement obviously and if you're unhappy with your 1070s performance today it's a big enough job to be worthwhile that said the price also jaunts an additional 50 over original msi rp for the 1070 despite being cheaper than last generation 20 series hardware three kingdoms at ultra is our most abusive gpu benchmark you could play this game at a much wider range of gpus if you're willing to drop down to high settings but it helps to have a benchmark title more brutal towards cards to fully stress the scaling at 4k the 39d fe can't even hold 60fps although obviously a reduction in settings would make it possible the 3080 holds 47 here with lows just above 30. that has its average 42 percent ahead of the 30-70 fe which is a massive lead when compared to other tests that's thanks to the additional strain produced by these settings 3070 overclock gains us 4.2 percent in this title and the 30-70 stock and oc end up flanked by the 2080ti and 2080 super the lead over the 1070 is 115 percent here so it's not further exaggerated than the previous titles despite the extra exaggeration at the top end of the chart the lead over the 5700 xt on the 3070 is a 45 in this test we're going to start going faster through these charts now that there's a clear pattern formed 1440p is interesting though for three kingdoms there's a clear advantage in the 2080ti 3080 and 30 90 and one percent and 0.1 lows interestingly this advantage is also afforded to the 5700 xt and radeon 7. although the lows and these aren't superior to the 3070 they are disproportionately better than their immediate flanks in the 1080 ti and 1080. the 3070 is led by 39 percent in the 380fe and the 3070 oc improves by 3.9 percent over stock 3070. at 1080p the 370fv stock gains by 3.7 to its oc results or 36 percent to the 380 fe that's not much different than 1440p scaling horizon zero dawn at 4k quality settings had the 370fv at 67 fps average so it's still plenty capable of playing 4k at higher settings you don't need a 3080 for that and you never did a few years ago we remarked that the 1080 ti was enough for many of the times games to play 4k with moderate settings and in 2018 we said the same of many of the 20 series cards the goal posts move as games advance clearly but the marketing around the 3080 as the first true 4k card was always silly and clearly still is the 3080 leads the 3070 by about 30 percent here the 30 70 oc gives us an uplift of a couple percent once again and the 3070 fe leads the 1080 ti by 36 percent the 5700 xt by 60 and the 1080 by 79 percent at 1440p the 3070 fe leads the 2080 ti strix by a couple fps but it's led by the 2080 ioc by also a couple fps the 3080s lead over the 3070 here is 23 while the 370s lead over the 2070 super is 37 the same goes for the 1080 ti since nvidia decided in the 20 series that it wanted to make a lot more 1080pis with a different name although we haven't seen that emerge yet for the 30 series in shadow of the tomb raider the 30 70 fe at 4k is able to run 66 fps with high settings so this maintains that 4k is plenty accessible on the new 500 card the stack is as follows the 3080 leads the 30-70 at 92 fps versus 66 or about 40 the 30-70 leads the 20 super stock by 9.7 percent 2070 super stock by 23 and 5700 xc by 28 and the 1080 and therefore 1070 ti by 76 percent at 1440p the stack remains the same in terms of hierarchy the 3070 fe holds 117 fps average here so the 2080 ti does keep a lead in this benchmark it's still close especially given the price of the former flagship card but overall the positioning is roughly the same as we've seen elsewhere we have a few other games but they're not worth talking through as we've already established the pattern we'll flash through a few of those charts here you can pause them if you'd like any more detail on a specific game up next will be the rtx games with a mix of non-dlss and dlss 2.0 coverage for both angles of the argument rtx has seen major uplift in the specific generation as compared to the 20 series and we explained all the reasons why in our amp here architecture deep dive if you'd like to learn more about that we'll link the architecture piece in the description below and we'd encourage you to check it out we'll start with quake 2 rtx this is interesting because it's entirely path traced using vulcan and because it's not a hybrid render approach it ends up exaggerating the generational rt core improvements as well at 4k the 3070 fe clearly struggles and is functionally equal to the 2080 ti the 3080 leads by a higher than average 44 at 1440p the 3070 is in playable territory at 52 fps average again tied with the 2080 ti and allowing the 3080 to hold a 43 lead 1080p posts the same hierarchy the 2080ti and 3070 are about the same and the 3070 leads the 2080 super by about 26 percent we use dlss in control because our user base said that it prefers dlss in this game and it's the lss 2.0 so it's better than the vaseline smear that was offered by dlss-1 at 4k and with dlss on so not quite true 4k as a reminder the 3070 runs 47 fps average just behind the 20 dti and allowing the 3080 a 37 lead at 65 fps at 1440p with the lss the 3070 manages 85 fps average positionally the same as before and with the 30 80 31 percent ahead we won't bother running the chart but fps improves to 118 at 1080p if you're interested in that minecraft rtx at 1080p had the rtx 3070 just ahead of the 2080 ti strix and the 3080 31 percent ahead of the 3070 but if we switch to 1440p results the 3070 fe was at 102 fps average ahead of the 2080 ti strix by about four percent power testing is last we're doing this using a power interposer setup for total card power not for system power if you want to figure out how big of a power supply you need you can basically add this number to cpu review numbers for a good idea of the worst case scenario total load although it's commonly one or the other component that's loaded for games obviously then had a buffer after that for the power supply size in fermark the 3070 fe card pulled about 227 watts stock or 247 when overclocked the 3080s for reference were anywhere from 324 watts on the fe to 400 watts on partner models although we won't plot all of them here closest proximity power numbers come from the 1080 ti ftw3 loosely anyway at 283 watts and the 2080 fe at 290 watts the 2070 pulled about 174 watts in this test where 200 when overclocked so the 3070 lands between some of these in total war 3 kingdoms just for a gaming workload we monitored about 174 watts on the 2070 stock card 200 watts on the 2070 oc 222 on the 3070 and about 321 on the 3080. this is actually nearly perfectly linear scaling in this specific test at least the 3080 is gaining about one percent performance for roughly every one percent increase in power consumption it maxes out at around 44 percent more power for about 42 percent higher average fps again in this specific title it won't always be this clean but it's cool to see the scaling at least persist here once you go beyond the 3080 power numbers though it all starts to fall apart for power efficiency versus performance gains so there is a ceiling to how long this will linearly scale closing out then again it's a bit weird to review a card like the 3070 right now just because there's no direct competitor from andy the card in a purely objective sense is good and that it's not doing anything horrible the drivers haven't given us any issues the overclocking was unimpressive definitely on this fe model at least and we'll be looking at partner models later but overclocking is also so much a function of silicon quality and unit to unit variants that it's hard to determine from one sample what it's going to look like on the whole but ours was not a very good overclocker which is unfortunate because our 3080 fcw3 for example our first one was one of the best overclockers we've ever had but sometimes you get duds that doesn't mean it'll reflect on all of them but we do have partner models to look at so overall we didn't run into anything experience ruining in terms of value while versus the 20 series nvidia certainly may have painted a pretty good picture for itself because it's bringing down its own pricing for a significantly better or similar class of hardware depending on if you're looking at rt performance or just rasterization performance and rasterization it's minimally better overall for the same price if not better overall for a lower price depending on which unit you're looking at between the skus that are out now for rt the performance goes up significantly more if you care about rtx performance and dxr or maybe you use something like optics in blender those are all things that would be benefited greatly by the ampere series and this brings the price floor down now rdna 2 is right around the corner but andy's uh update that's coming out tomorrow at time of this video going live is going to be an announcement not a launch as far as we understand now so there shouldn't be product availability immediately but that announcement should at least give us an initial idea for when the cards are going to come out at that point in our amd news piece we'll talk about if our thoughts on this have changed at all or how the rd a2 news has influenced the the market the mechanical mount for this cooler is actually the most impressive we've seen thus far in our pressure testing with our scanner that we use so that was interesting but gaming performance overall has you in the range of 20 to 44 percent higher than a 3080 it varies depending on the game and the resolution at 4k the 3080 pulls away on average from the 30 70 a little bit more then i'd say 1080p and then also in rt performance in fully path trace scenarios the 3080 does have a little bit more of an advantage over the 30 70 that becomes a slightly disproportionate versus rasterization performance so 1080 ti to a 30 70 to us doesn't particularly seem maybe worth it unless you are unhappy with your 1080 ti performance now but otherwise that's still a great card 1070 is about a 2x or greater increase depending on the game so that could be a worthwhile change if you're getting tired of your 1070 you're not happy with the performance anymore you want to increase your monitor resolution something like that that would be a worthwhile change 970 holders on this is actually a pretty good upgrade opportunity at this point uh although the pricing has changed from back when the 970 was launched so that's it for the review obviously if you're happy with the video card you have now there's no reason you have to upgrade just because something new came out hold on to it make the cards that you have last as long as you can and then upgrade when it really truly makes sense because you're unhappy with the performance you have uh so subscribe for more as always you can go to store.camerasexus.net or patreon.comgamersnexus helps out directly and we'll see you all next time thetoday we're reviewing the nvidia rtx 3070 founders edition card that just hit the market this has a 500 msrp there are partner models coming out shortly as well partners of course being people like evga aces gigabyte so forth those will be out in about two days from the time of this review going up that's when on videos allowing the release of partner model cards and we'll be reviewing those separately but today we're focusing on the founders edition card and testing will include thermals power frequency overclocking gaming mechanical assembly so like pressure testing stuff like that but we're also going to have a separate teardown that focuses just explicitly on taking the card apart looking at assembly quality ease of access if you need to repair something and cooler design from a mechanical assembly side of things so check back in a few hours on this channel for the tear down video because that'll be going up the same day as this one but for this we're going to be focusing on performance versus 3080 fe for a point of comparison amd has rdna 2 coming out soon but we can't test that today obviously because we don't have it so you'll have to check back during our dna2 announcements to see if we have any additional opinions on this card once that one is formally announced which is tomorrow at the time of this going up before that this video is brought to you by us and our gamers nexus wireframe mousemat aside from being the best way to directly support our lawn form investigative reporting you can also get a custom made high quality mouse mat made with a high detail 3d design that we created to show off heat sinks coolers video cards and more the mouse mat uses a stitched blue border for added longevity a blue rubber underside for unique flair and a microfiber cloth for smooth tracking the mat is 36 inches by 12 inches and fits a keyboard and mouse easily to back order your mouse mat and ensure you get one in the next run go to store.gamersnexus.net and backorder yours while reducing our reliance on advertisers or click the link in the description below it's kind of weird to review a video card right now at least one like this where at this point nvidia is sort of just one upping itself while amd is preparing its next gpu launches that means it's a little bit difficult to do the usual comparisons because a lot of the time in the charts we spend a good deal of effort analyzing the performance differences between the closest competitor from amd versus nvidia and vice versa in this scenario though it's really just nvidia versus 20 series versus 10 series nvidia versus the 3080 to see what kind of performance difference there is there so it's a little different but uh it's still worth testing obviously this card at 500 comes in about 200 cheaper than the floor price of the 3080s for the most part partner models you should be attacking on anywhere from 20 to 80 depending on how high end you're going where the the highest of the high end mostly affords you additional features like overclocking support or perhaps water cooling water blocks hybrid cars stuff like that 4k is also going to be a point of interest here over the years there have been many cards that were fine for 4k gaming the 1080 ti was in fact one of them but obviously the goal posts move as the gaming industry evolves and the graphics options become more complex so 4k has been kind of a difficult target to hit for video cards over the years this one does overall fine with approaching 4k gaming as you'll see in our test today but it's not the first one to have done it and the 3080 has gotten a lot more attention as the true world's first like fully 4k ready always 4k playable type of card this however does threaten that a little bit let's look at the specs the rtx 3070 runs 5888 cuda cores versus 10496 on the 3090 or 8704 on the 3080. despite the 3080 being a 47ish percent increase in cuda cores there's obviously a lot more to it than that in terms of determining gaming performance we won't see linear scaling and we never do frequency also comes into play which we'll show soon as does memory the biggest change to the 3070 is that it's moving to gddr6 from g6x so it's bumping back down to memory that's easier to cool but also significantly slower at 16 gigabits per second stock versus 19 gigabits per second on the 3080 which also has 10 gigabytes of memory as another benefit as a result memory bandwidth is down to 512 gigabytes per second on the 3070 from 760 gigabytes per second previously on the 3080 the bus width has changed from 320 bit to 256 bit and another major update aside from the rt cores is in rops it's going from 88 robs to 64. now this should be much more noticeable in 4k than say 1080p test as the resolution goes up the 3080 with the rob's benefit memory bandwidth benefit should start to pull away from the 3070 at say 4k resolution we've got a block diagram we can show of the 70 die layout as well and in our tear down or we can drop in a b-roll clip here we've gotten die shots of it where it is a physically different piece of silicon the mounting mechanism is actually very interesting we'll talk about that in the tear down but it's a different piece of silicon it's smaller and so it's still amp here but it's a cut down version of amp here anyway we've got a ton of testing to go through today so let's get into it we'll open up with something different that you won't see in the other 3070 fe reviews just to keep things fresh our teardown of this card going live a few hours after this video shows just how tightly put together the 3070 fe card really is there aren't that many screws this time which is good but they're evenly spaced with only four that secure the cooler to the pcb the remainder of them secure the unibody shroud to the back plate we initially had concerns about this regarding how much pressure is applied but in this pseudo-color image produced using a nist traceable pressure sensing piece of equipment that gn paid for thanks to its patreon backers and store.gamersnexus.net supporters we're able to see just how much that unibody shroud matters although it isn't leveraged directly as a piece of the cooler but it could be because it's aluminum it does secure to the back plate in a way that it's drastically changing the pressure application to gpu silicon in this pseudo color image without the shroud we can see a major gap in coverage towards one side of the die while the unshrouded 3070 fe has the most even contact we've yet seen with this new testing equipment this is basically perfect or very close to it we'll put some of our previous measurements on screen as well like the other founders edition cards and one of the partner models as shown here the 3070 fe with the shroud has even contact and has good tension across the entire pcb at least as far as gpu silicon is concerned that's off to a good start but pressure alone doesn't make a good cooler the pressure map looks good but it isn't a height map this chart shows the depth from a known zero point on a special needle for measuring depth on things like coolers more consistent is best so we want to see a smaller box with measurements clustered tightly together the 30-70 feet does fine it's not as consistent and point-to-point measurement as the 2080ti strix or the gigabyte eagle but we also didn't find any significant outliers in the measurement that blew out the scale unlike the gigabyte eagle the 3070 fe seems fine overall in this department so off to the other tests we'll start with the frequency validation plot this looks at temperature and frequency over time which is helpful in illustrating behavior of gpu boosting tables the rtx 3070 founder's edition card started at about 1965 megahertz when it was under 60 degrees celsius at the very beginning of the test after this it fell slowly to 1890 megahertz to 1920 megahertz went at steady state thermals of about 72 degrees celsius in this workload as a reminder this is with full auto settings so it represents out of the box usage and out of the box thermals the v-bios temperature target seems to be about 72 in this 3dmark frozen frame workload which is representative of gaming scenarios our power virus testing later will plot higher frequency is relatively stable but far from the flatline behavior of water cooled cards that have eliminated thermals as a consideration or cards which have eliminated power from consideration by way of mods as expected we're losing about 45 megahertz to temperature as this card warms up to steady state this is where board partner models will often show their value for perspective this plot will show the frequency of some other cards that we've tested recently we've intentionally zoomed the scale so that it's easier to read where each one falls you'll first see that the 30 70 fe line plots exactly where it did previously as the data hasn't changed the scale's just been zoomed in the rtx 3080 eagle by gigabyte plots around 2000 megahertz and this helps us better understand one of the many aspects of why the 3080 will outperform the 3070. there's clearly hardware level differences like fp32 capabilities memory memory bandwidth but frequency also counts for performance deltas the 3080fe is much closer to the 3070 fe at about 1925 to 1950 megahertz the asus 3080 tough cards also stick around the 2000 megahertz marker and the rtx 3090 ftw3 runs about 1940 to 1950 megahertz the highest end nvidia cards like the 3090 typically run slower than the slim's down counterparts like the 3080s here that's the nature of balancing stability against more hardware level physical advantages like more fp32 or other large die gpu benefits time to move on to thermals we'll start with 40 dba noise normalized thermals normalizing for noise helps us understand the efficiency of a cooler when it's disallowed from blasting its fan speeds to compensate for cooler size or design we haven't tested other 3070s yet obviously because we're not allowed to right now but we can at least show behavior compared to the other rtx 30 cards that we've previously tested obviously these should ideally be power normalized as well for pure cooler comparison but since we're comparing across gpu skus it doesn't really matter here's the gpu chart for noise normalized thermals we'll look at memory next the rtx 3070 fe at 227 watts power for total board power as validated with physical interposers plots at 65 degrees celsius in a 21 degree ambient environment this is with the card set to 40 dba at a 20 inch distance in a room with a noise floor of 26 db that has it about 8 to 9 degrees warmer than the 3080 tough by asus which held 56 degrees celsius at 340 watts without even looking at other rtx 3070 cards this allows us to demonstrate how partner models can excel the tough is handling 120 watts more power in this power virus workload anyway and yet manages better results now of course a smaller die would also mean less surface area and a higher thermal density but the 3070 fe won't be the best cooler for the rgx 3070 series either that said it's significantly better than what we've seen in the past for reference coolers from nvidia and nvidia has absolutely improved in its cooler design and thermal performance on the fe cards credit where it's deserved but there's more ground to gain on partners yet this chart shows the noise normalized performance for board components as tested with thermocouples that we manually placed all over the board for the 3070 fe the bottom memory module measured at 53 degrees on the flip chip pga side of the silicon which is well within spec and partially thanks to the drop from gddr6x just to gdr6 the top module that we measured ran about 60 degrees celsius when noise normalized but keep in mind that we're running the fans much faster than auto here we'll look at that next the vrm moss measurement ran about 72 degrees celsius on the i o side of the board which is getting into warm territory but still plenty far from the 100c plus capabilities of the mosfets this chart shows the same test except with the card auto controlled in this one the memory runs at 61 degrees and 69 degrees celsius on the rtx 3070 fe with the internal silicon number between 5c and 10c higher than that exterior measurement that we've taken that's still fine but it's moving to the warm end of the scale the mosfet measurement is more concerning however at 82 degrees externally it's potentially in the low 90s for the internal silicon although we can't get the actual mosfet measurements because nvidia has decided to not expose them through software anywhere we're approaching territory where an extra 10 degrees internal case ambient temperature could mean those vrm components are running borderline acceptable at least for long term use this is the nature of condensing the pcb down as tiny as nvidia did coupled with the component choices for the vrm and with the decision to run these fans so quietly it's expounded we think nvidia should sacrifice some of the noise for an extra 2 dba worth of fan speed which would alleviate a lot of the long-term concerns with the cards left to self-regulate their performance behaviors the rtx 3070 fe held around 33 dba making it one of the quietest out-of-box cards on our rtx 30 benches when they're all left to self-regulate it's also among the lowest power consumption thus far so this all makes sense the 3070 holds around 75 degrees celsius when left to fully control its own fan speeds in this power virus workload so vbio seems to have established this temperature target and a fan curve that holds around 1480 rpm auto in this benchmark remember that there are driver level changes in behavior between things like firmark and 3dmark so that's why we'll see some additional differences in behavior between each workload the card runs 5 degrees warmer than the 40dba ftw3 at 400 watts but because the noise isn't normalized here you'd have to look at our previous chart for a real efficiency comparison this chart shows a weak point for the 3070 founders edition it's showing the v bios programming managed against the gaming workload applied which is useful for analyzing hysteresis and responsiveness interestingly the 3070 fe has a much more abrupt fan behavior than we've seen in most other carts rather than ramping gradually and at a steady pace the 3070 fe first ramps to about 1000 rpm under initial load on a lag by the way of about 30 seconds from initial load start and then it tries to hold this rpm while load continues to ramp it holds here for another 40 to 45 seconds and then ramps sharply to 1300 rpm attempts to hold and then ramps sharply again to roughly where it rests in the 1480 rpm range the end result is that the fan's initial spikes are more noticeable to a human and therefore more annoying if rather than attempting to hold that 1000 rpm first it had instead continued a gradual ramp the noise change would have been less noticeable from a human factor's perspective nvidia could improve on its fan programming for v-bios on the 30-70 fe namely by changing the lag for ramp this chart shows the noise levels as plotted against fan rpm measured with the same methodology discussed previously the rtx 3070 fe has a fairly straight increase in noise with each five percent increment in pwm signal the maximum noise level stops at about 56.9 dba at 20 inches although it'll never reach this without manual input from the user software allows this card to be controlled in one percent increments all the way down to zero percent which is mostly unheard of and that's for good reason because it's broken that's below what pwm can actually support and so the card has a hard stop at 30 and 1000 rpm for actual change although it can be technically adjusted in software down to 20 percent one percent or anywhere between it'll just still spin at thirty percent speed normally you see a hard cut off at thirty percent where software will not allow you to drag it down lower uh but this is a mix of software bugs and things that will be ironed out later time to get into the game benchmarks for the 30-70 as a quick reminder you can always calculate improvement against whatever gpu you're using by looking at the aggregate for old benchmarks where you might see the 970 versus the 1070 and say the later years of the 1070. by doing that you can look at the delta the average percentage improvement between whatever gpu you have and something that is on these charts by referencing older data and then you can figure out the delta between your gpu and what is on these charts so if yours isn't here you're still able to calculate where it should end up falling we'll start our game benchmarks with red dead redemption 2 at 1080p keep in mind that many tests become somewhat bound at the top end and don't show as dramatic scaling as you'd see at 1440 or 4k in this scenario tested first with dx12 and high-end custom settings the nvidia rtx 3070 fe stock card ran at 123 fps average with its oc providing just 3.2 percent uplift in average fps and fortunately no loss in the lows stock to stock the 3070 fe leads the 2080 super by 12.4 percent with 5700 xt and 2070 super placement led by 23 the 1080 ti ftw 3 was led by the 30 70 by 37 percent this definitely isn't just another 1080 ti in the form of a 30 70. it's actually improved although maybe nvidia will do that for the 3060 ti or whatever that's called the 3080 fe leads the 3070 fe by 24 at 1080p and that's when both are stuck let's move on to 1440p we can start by comparing those previous percentages at 1080p the 3070 fe led the 2080 super by 12.4 percent but now it's leading by 19 the lead over the 1080 ti of tw3 expands from 37 to 45 percent while the 380 fes lead of 24 at 1080p stays about the same maintaining 24 here as well the 30 70 fe plants itself just under the 2080 ti and just ahead of the 2080 super oc compared to the 1070 from a few years ago the gains are about 128 percent with one percent and 0.1 lows improving significantly as well tested at 4k the 3070 fe still maintains above 60 fps average with lows reasonably spaced behind and without any major stutters we didn't run into any memory limitation issues here as expected because we're hitting other limits first comparing the previous percentages the 380 fe now jumps to 34 percent the head up from a previous 24 stretching its legs a bit more with the memory bandwidth and drops advantages the 3070 fe holds an 18 lead over the 2080 super staying about the same distance as in the 1440p results the 5700 xt for an amd reference is led by the 30 70 fe by 34 percent as is the 2070 super and the 1080 ti with an overclock they're all about the same placement that's up from the 1080p results as well and so modern changes in memory subsystems and the architecture have more notably benefited these higher resolution tests again the rops are another notable contributor here if you're curious about vulkan performance this chart shows dx12 versus vulcan scaling in red dead 2. this helps to establish how much the api matters on one generation versus another we've previously found that vulkan benefits the rtx 30 series by about 10 percent for the cards out so far and that mostly persists with the 3070 at 9.2 percent ahead of its dx12 results some of the baseline performance may change because of differences in what's being rendered between the two apis but the end result is that the 30 series handles the switch over better than the 10 series cards before it the 20 series isn't too different in the 2080 super nti class and the 5700 xt for reference was about 1.4 percent better in vulcan and so the choice should likely hinge more on whichever one is the least buggy at any given time and that comes down to the game's developers rainbow six siege is next we'll start at 4k this time just because rainbow six siege is so relatively low in gpu load that it's more realistically usable the 3070 fe ran a 144 fps average talk result with lows at 123 and 118. that puts it tied with the 2080 ti strix when stock but allows the strix oc a lead of 12 percent our 3070 feoc although stable elsewhere proved problematic here and posted one percent worse results than stock for that reason we've removed the data from the chart to reduce confusion we're not able to get a worthwhile oc dialed in for this game within the review time window unfortunately and let's have to try in a few different settings and combinations the 3080 beats the 3070 at 179 fps average or 24 stock the stock the 3070 leads the 1070 by 137 it leads the 1080 ftw by also about double and the 1070 ti would be alongside the 1080 here the 2060 ko is also led by about double in the frame rate and improved in frame time consistency by moving to the 30 70. the 2070 super is led by just over 40 percent for averages and that's a significant generational gain when considering the pricing and considering that the super was a refresh here's a frame time plot to better look at the frame to frame intervals for each rendered frame we're talking a lot about percentages but that's all calculated against an average this is the closest we can get to an empirical look at gaming performance for each frame because we can see each frame's render time requirement in the on present metric lower is better and more consistent is best in this test the 3070 stock runs about 7 milliseconds frame to frame intervals on average and is typically plus or minus one millisecond from one frame to the next the largest gap we see is three milliseconds in general and as long as the frames don't deviate or experience an excursion from the mean of more than eight to twelve milliseconds frame to frame at least then no one will ever notice that difference below eight milliseconds especially here we're more talking one millisecond frame to frame is not perceptible to a human the 3080 fe stock card ran around five milliseconds with slightly wider deviation but also not one that ever be noticed at 1440p the rtx 3080's lead over the 3070 is reduced to 19 from 24 as the bandwidth benefit is reduced here the 3070 stock fe leads the 2080 ti and the 2080 super oc although to be fair the 2080 super oc did a lot for this game just like the 2080 tioc did it's unfortunate that our 3070 overclock couldn't hold as well in this game as others but if we had more time within the review window we could have eventually tuned it to work the improvement over the 1070 is reduced to 120 percent from 137 percent again posting higher scaling as resolution increases and we're not cpu bound here so it's bandwidth and rops and architectural changes like fp32 capabilities it's sort of odd though to review a gpu in general right now just because there's no direct competitor from amd nvidia isn't having any trouble outmatching its last gen products so it's almost a little bit more on the boring side than it might be otherwise until at least amd gets our dna 2 out at 1080p the 3070 fe manages 385 fps average so it's showing playable scaling all the way from 1080 to 4k the 3080s lead is reduced to 14 while the 30 70's lead over the 1070 is reduced to 94 nvidia's investment in bandwidth for this generation shows it's still a significant improvement obviously and if you're unhappy with your 1070s performance today it's a big enough job to be worthwhile that said the price also jaunts an additional 50 over original msi rp for the 1070 despite being cheaper than last generation 20 series hardware three kingdoms at ultra is our most abusive gpu benchmark you could play this game at a much wider range of gpus if you're willing to drop down to high settings but it helps to have a benchmark title more brutal towards cards to fully stress the scaling at 4k the 39d fe can't even hold 60fps although obviously a reduction in settings would make it possible the 3080 holds 47 here with lows just above 30. that has its average 42 percent ahead of the 30-70 fe which is a massive lead when compared to other tests that's thanks to the additional strain produced by these settings 3070 overclock gains us 4.2 percent in this title and the 30-70 stock and oc end up flanked by the 2080ti and 2080 super the lead over the 1070 is 115 percent here so it's not further exaggerated than the previous titles despite the extra exaggeration at the top end of the chart the lead over the 5700 xt on the 3070 is a 45 in this test we're going to start going faster through these charts now that there's a clear pattern formed 1440p is interesting though for three kingdoms there's a clear advantage in the 2080ti 3080 and 30 90 and one percent and 0.1 lows interestingly this advantage is also afforded to the 5700 xt and radeon 7. although the lows and these aren't superior to the 3070 they are disproportionately better than their immediate flanks in the 1080 ti and 1080. the 3070 is led by 39 percent in the 380fe and the 3070 oc improves by 3.9 percent over stock 3070. at 1080p the 370fv stock gains by 3.7 to its oc results or 36 percent to the 380 fe that's not much different than 1440p scaling horizon zero dawn at 4k quality settings had the 370fv at 67 fps average so it's still plenty capable of playing 4k at higher settings you don't need a 3080 for that and you never did a few years ago we remarked that the 1080 ti was enough for many of the times games to play 4k with moderate settings and in 2018 we said the same of many of the 20 series cards the goal posts move as games advance clearly but the marketing around the 3080 as the first true 4k card was always silly and clearly still is the 3080 leads the 3070 by about 30 percent here the 30 70 oc gives us an uplift of a couple percent once again and the 3070 fe leads the 1080 ti by 36 percent the 5700 xt by 60 and the 1080 by 79 percent at 1440p the 3070 fe leads the 2080 ti strix by a couple fps but it's led by the 2080 ioc by also a couple fps the 3080s lead over the 3070 here is 23 while the 370s lead over the 2070 super is 37 the same goes for the 1080 ti since nvidia decided in the 20 series that it wanted to make a lot more 1080pis with a different name although we haven't seen that emerge yet for the 30 series in shadow of the tomb raider the 30 70 fe at 4k is able to run 66 fps with high settings so this maintains that 4k is plenty accessible on the new 500 card the stack is as follows the 3080 leads the 30-70 at 92 fps versus 66 or about 40 the 30-70 leads the 20 super stock by 9.7 percent 2070 super stock by 23 and 5700 xc by 28 and the 1080 and therefore 1070 ti by 76 percent at 1440p the stack remains the same in terms of hierarchy the 3070 fe holds 117 fps average here so the 2080 ti does keep a lead in this benchmark it's still close especially given the price of the former flagship card but overall the positioning is roughly the same as we've seen elsewhere we have a few other games but they're not worth talking through as we've already established the pattern we'll flash through a few of those charts here you can pause them if you'd like any more detail on a specific game up next will be the rtx games with a mix of non-dlss and dlss 2.0 coverage for both angles of the argument rtx has seen major uplift in the specific generation as compared to the 20 series and we explained all the reasons why in our amp here architecture deep dive if you'd like to learn more about that we'll link the architecture piece in the description below and we'd encourage you to check it out we'll start with quake 2 rtx this is interesting because it's entirely path traced using vulcan and because it's not a hybrid render approach it ends up exaggerating the generational rt core improvements as well at 4k the 3070 fe clearly struggles and is functionally equal to the 2080 ti the 3080 leads by a higher than average 44 at 1440p the 3070 is in playable territory at 52 fps average again tied with the 2080 ti and allowing the 3080 to hold a 43 lead 1080p posts the same hierarchy the 2080ti and 3070 are about the same and the 3070 leads the 2080 super by about 26 percent we use dlss in control because our user base said that it prefers dlss in this game and it's the lss 2.0 so it's better than the vaseline smear that was offered by dlss-1 at 4k and with dlss on so not quite true 4k as a reminder the 3070 runs 47 fps average just behind the 20 dti and allowing the 3080 a 37 lead at 65 fps at 1440p with the lss the 3070 manages 85 fps average positionally the same as before and with the 30 80 31 percent ahead we won't bother running the chart but fps improves to 118 at 1080p if you're interested in that minecraft rtx at 1080p had the rtx 3070 just ahead of the 2080 ti strix and the 3080 31 percent ahead of the 3070 but if we switch to 1440p results the 3070 fe was at 102 fps average ahead of the 2080 ti strix by about four percent power testing is last we're doing this using a power interposer setup for total card power not for system power if you want to figure out how big of a power supply you need you can basically add this number to cpu review numbers for a good idea of the worst case scenario total load although it's commonly one or the other component that's loaded for games obviously then had a buffer after that for the power supply size in fermark the 3070 fe card pulled about 227 watts stock or 247 when overclocked the 3080s for reference were anywhere from 324 watts on the fe to 400 watts on partner models although we won't plot all of them here closest proximity power numbers come from the 1080 ti ftw3 loosely anyway at 283 watts and the 2080 fe at 290 watts the 2070 pulled about 174 watts in this test where 200 when overclocked so the 3070 lands between some of these in total war 3 kingdoms just for a gaming workload we monitored about 174 watts on the 2070 stock card 200 watts on the 2070 oc 222 on the 3070 and about 321 on the 3080. this is actually nearly perfectly linear scaling in this specific test at least the 3080 is gaining about one percent performance for roughly every one percent increase in power consumption it maxes out at around 44 percent more power for about 42 percent higher average fps again in this specific title it won't always be this clean but it's cool to see the scaling at least persist here once you go beyond the 3080 power numbers though it all starts to fall apart for power efficiency versus performance gains so there is a ceiling to how long this will linearly scale closing out then again it's a bit weird to review a card like the 3070 right now just because there's no direct competitor from andy the card in a purely objective sense is good and that it's not doing anything horrible the drivers haven't given us any issues the overclocking was unimpressive definitely on this fe model at least and we'll be looking at partner models later but overclocking is also so much a function of silicon quality and unit to unit variants that it's hard to determine from one sample what it's going to look like on the whole but ours was not a very good overclocker which is unfortunate because our 3080 fcw3 for example our first one was one of the best overclockers we've ever had but sometimes you get duds that doesn't mean it'll reflect on all of them but we do have partner models to look at so overall we didn't run into anything experience ruining in terms of value while versus the 20 series nvidia certainly may have painted a pretty good picture for itself because it's bringing down its own pricing for a significantly better or similar class of hardware depending on if you're looking at rt performance or just rasterization performance and rasterization it's minimally better overall for the same price if not better overall for a lower price depending on which unit you're looking at between the skus that are out now for rt the performance goes up significantly more if you care about rtx performance and dxr or maybe you use something like optics in blender those are all things that would be benefited greatly by the ampere series and this brings the price floor down now rdna 2 is right around the corner but andy's uh update that's coming out tomorrow at time of this video going live is going to be an announcement not a launch as far as we understand now so there shouldn't be product availability immediately but that announcement should at least give us an initial idea for when the cards are going to come out at that point in our amd news piece we'll talk about if our thoughts on this have changed at all or how the rd a2 news has influenced the the market the mechanical mount for this cooler is actually the most impressive we've seen thus far in our pressure testing with our scanner that we use so that was interesting but gaming performance overall has you in the range of 20 to 44 percent higher than a 3080 it varies depending on the game and the resolution at 4k the 3080 pulls away on average from the 30 70 a little bit more then i'd say 1080p and then also in rt performance in fully path trace scenarios the 3080 does have a little bit more of an advantage over the 30 70 that becomes a slightly disproportionate versus rasterization performance so 1080 ti to a 30 70 to us doesn't particularly seem maybe worth it unless you are unhappy with your 1080 ti performance now but otherwise that's still a great card 1070 is about a 2x or greater increase depending on the game so that could be a worthwhile change if you're getting tired of your 1070 you're not happy with the performance anymore you want to increase your monitor resolution something like that that would be a worthwhile change 970 holders on this is actually a pretty good upgrade opportunity at this point uh although the pricing has changed from back when the 970 was launched so that's it for the review obviously if you're happy with the video card you have now there's no reason you have to upgrade just because something new came out hold on to it make the cards that you have last as long as you can and then upgrade when it really truly makes sense because you're unhappy with the performance you have uh so subscribe for more as always you can go to store.camerasexus.net or patreon.comgamersnexus helps out directly and we'll see you all next time the\n"