The Brewing Problem with GPU Power Design _ Transients

**The Great GPU Power Supply Conundrum: Unraveling the Mystery**

In recent times, there has been a growing concern among computer enthusiasts and reviewers about the power supply situation for high-performance graphics cards. As manufacturers continue to push the boundaries of technology, we're seeing an increase in power consumption, which can lead to issues with transient current spikes and system shutdowns. In this article, we'll delve into the world of GPU power supplies, exploring the reasons behind these problems and what you can do to mitigate them.

One of the primary concerns is the upcoming 40 series of AMD GPUs, which promises significant performance boosts but also increased power consumption. While manufacturers are working to develop more efficient power supplies, it's essential to be aware of the potential risks associated with high-performance graphics cards. As we approach the 400-watt mark, even mid-range or lower-end power supplies might struggle to handle transient spikes, making it crucial to choose a reliable and capable power supply for your system.

**Transient Current Spikes: The Enemy of System Stability**

So, what exactly are these transient current spikes, and how do they affect our systems? Transient current spikes occur when the electrical current in the power supply surges beyond its normal operating range. This can happen due to various factors, such as high-power graphics cards, overclocking, or even a faulty power supply unit (PSU). The resulting spike can be so severe that it causes the system to shut down, potentially leading to data loss and other complications.

To understand this phenomenon better, we need to look at some of the testing results from our team. We created transient current spikes through benchmarking and gaming, and while we didn't find anything significant enough to alter our monthly power bill, we did encounter reproducible small form factor power shutdown scenarios. This tells us that even with well-designed power supplies, there's still a risk of system instability when dealing with high-power graphics cards.

**Motherboard Power Supply Compatibility: A Crucial Consideration**

Another critical aspect is the compatibility between motherboards and power supplies. Some motherboards seem more resilient to transient current spikes than others, but this doesn't mean that all users are immune to these issues. In fact, some power supply manufacturers warned us about certain motherboard lines being particularly sensitive to these problems.

For example, we encountered an issue with EVGA's GA line, which had a problem with OCP (Overcurrent Protection) triggering unwanted shutdowns. On the other hand, Seic was touchy with OCP as well, highlighting the importance of selecting a reliable power supply for your system. It's essential to note that not all small form factor power supplies are created equal and might not be suitable for high-performance graphics cards.

**The 30 Series: A Learning Curve**

While the 40 series is still on the horizon, we've had experience with the 30 series, which has its own set of challenges. We ran into some issues with small form factor power supplies, and it's possible that they might not be a good fit for certain high-performance GPUs like the 3080 or higher.

If you're experiencing shutdowns or system instability, it's essential to investigate further to determine if your issue lies with the power supply or the graphics card. Power supplies are generally easier to replace than GPUs, so don't hesitate to upgrade if necessary. However, before making a purchase decision, consider the specific power requirements of your system and choose a reliable power supply that can handle those demands.

**The Future of GPU Power Supplies: Preparing for the Worst**

As we move forward into the 40 series, it's essential to be prepared for potential issues with power consumption. Manufacturers are working hard to develop more efficient power supplies, but it's crucial to be aware of the risks associated with high-performance graphics cards.

We can't predict exactly what the future holds, but one thing is certain: as we approach the 400-watt mark, even mid-range or lower-end power supplies might struggle to handle transient spikes. It's essential to choose a reliable and capable power supply for your system, especially if you're planning to overclock or push your graphics card to its limits.

**Join the Conversation**

We want to hear from you! If you've experienced shutdowns or system instability, please share your story in the comments below. What motherboard, power supply, and GPU are you using? Have you encountered issues with transient current spikes or OCP problems? Your input will help us better understand these complex issues and provide valuable insights for our readers.

As always, we strive to create informative content that helps you learn something new and troubleshoot better when building your system. If you found this article helpful, please consider subscribing to our channel or supporting us on Patreon. Together, we can tackle the challenges of GPU power supplies and build faster, more reliable systems for years to come.

"WEBVTTKind: captionsLanguage: enthere's a Confluence of problems coming to a head between GPU makers GPU makers like Nvidia who are forced to pull more and more power to get competitive generational gains and power supply makers like these and the problem is that these two groups shift the blame between each other when a customer system does something like this and that's why today we're going to investigate transient power consumption spikes where the power consumption of a GPU for example might jump from 291 Watts over two times to 664 watts and that's just what today's gpus it's going to get worse with the RTX 40 series before that this in-depth video is brought to you by us and store. Gamers nexus.net we self-funded this huge project with your support from the GN store we just got a huge restock of our highquality red and black HUD Mouse mats which are desk-sized mousing services with custom Red Rubber Underside red stitching for anti Fray and a high resolution print we also just restocked our top selling product the black and blue wireframe Mouse mats which have CPU coolers Ram motherboards and more depicted on a mat that's built to last if you want even more functional products our 10piece toolkits for GPU disassembly and PC maintenance recently got a 7-year retroactive warranty and also have autographed variant visit store. gamas access.net to help us out on our next piece like this recent rumors have painted a concerning picture for the future of gpus or at least for the feature of the Nvidia RTX 40 series where now the discussion is moving away from average power consumption and instead towards transient power spikes or basically massive spikes that are largely unaccounted for that may only last up to 100 micros seconds and yet are enough to in the blink of an eye shut down the syst system in basically the same way as ocp would or overcurrent protection so that's the problem that we're looking at for the future of video cards and it's something that has to be accounted for today this is actually something that's been building for a long time we've seen evidence as far back as the 10 series and maybe even before that the difference is that today's gpus are being designed to nominally draw far more power and therefore the transient spikes are far more likely to knock a power supply off line and shut it down and so the fastest recap possible is that this issue looks something like this we can show all kinds of charts and oscilloscope shots to detail this issue and we will today but ultimately from the user experience end that's pretty much how it looks if it's the Ron mix of CPU GPU and motherboard then you might end up with a very difficult to troubleshoot problem that just trips power sometimes it's hard to pinpoint which exact component is causing that the power supply might be an easy fix but the motherboard 2 is involved so the system will just shut down and you'll lose anything unsaved or progress and whatever you're working on and the problem is that it is a mix of all these things the real issue is that it's not just as simple as it used to be for the most part you can take the average power consumption of the CPU and the GPU and figure out an appropriate power supply for those two now however you should also be baking an additional Headroom to your power supply selection for potential transient spikes especially at the high end with the RTX 30 series and probably the RTX 40 series as well in our lab testing here on this machine and just with real world test benches and computers we've seen that some motherboards handle this far better than others we've also seen that some video cards handle this better than others and some power supplies so it's an intermittent problem and that's frustrating for consumers without a lab full of equipment to just identify where it came from that's where we come in here's the issue we'll explain this chart in more detail a little bit later but we wanted to get this up front too this chart shows GPU power consumption under different scenarios in the game Doom Eternal the RTX 390 Master pulled 291 Watts under averaged power draw over an extended period And this is true with either a slower current clamp or with nvidia's power draw measurement tools like the PCAT analyzer but its blue and red bars paint another story the same card when tested with our oscilloscope instead of Which is far faster pulled 664 Watts at its peak transient Spike that's enough to put serious strain on power supplies that can't soak the temporary burst it's a 129% increase in power drop more than double from the Baseline average power consumption and that shows a major part of the story of these cards getting left out in at least our testing which we're working on addressing starting with today's video where now we and a lot of other reviewers mostly rely on average power consumption metrics that's enough to sort of estimate how much the computer will cost to run or at least the video card will cost to run it's enough to do a head-to-head comparison for overall power efficiency or performance per watt it's also enough to estimate the heat being produced by the device because ultimately the average overuse is what matters for all of those scenarios what it's not enough for is to know what the chances are of a power supply that's on the border of supporting this particular card to be able to stay up and not have an ocp drop so that existing testing methodology it has to change and that's what we've been working on with this for a couple of months now just improving behind the scenes because it cuts out the entire part of the story where you're looking at transient spikes that are probably the source of a lot of systems just shutting down out there where people are posting on forums having difficulty pinpointing the exact issue and no one wants the issue to be their video card they don't want to blame the video card because they had to fight hard for it and it was expensive so everybody wants the problem to be a cheaper component or an easier to Source One like a power supply or a motherboard you can get lucky with that but the GPU is sort of ultimately at the center of attention and this is why we continually overhaul our own testing processes to capture more of the picture of the component we're reviewing now back to this issue small form factur power supplies will be the most at risk of sinking under this demand as we saw in our NZXT H1 V2 testing with a highend CPU and highend GPU the small for fact power supply just couldn't take it small formfactor power supplies in particular struggle more with this because they have more limited real estate to actually build out the power supply and therefore reduced capability of dealing with these sudden spikes they're more likely to just shut down from ocp so that's the preview and now it's time to talk about how all of this works and what it means for you building computers and what it means for the industry talking about power supply manufacturers they're being told by Nvidia to bulk upon capacitance and better technology capabilities to deal with these transient spikes especially from the RTX 40 series where we're looking at Baseline power targets of around 450 Watts based on some of the leaks that are out there now for higher end cards and that's just the Baseline if you're spiking a couple times that you could see where it starts to introduce problems now power supply makers of course think this is nvidia's Fault Nvidia thinks it's the power supply manufacturer's issue to deal with and both of them have reasons to push the blame onto the other one power supply makers generally think it's unreasonable for NVIDIA to push all of the requirement of dealing with transient spikes onto them and think instead that Nvidia should be doing better at a GPU and a board level to supress those transient spikes but power supply makers also have other reasons to push back against Nvidia okay so here's what transients actually are we already mentioned that 30 series video cards can tax your power supply with Beyond two times the specified power load of the gpus on average and these quick periods of power consumption are often referred to as transients not transients but transient the term transient refers to a short-lived oscillation in a system caused by a sudden change of voltage or current or load in this case the transient is driven by the load which is the GPU and that manifests itself in the current we can measure this in the wires of a power supply cable by detecting the magnetic field in and around the wire and if you invest in high-end test clamps like we did you can isolate those spikes even though they last mere micros seconds in many scenarios the voltage from a power supply is carefully regulated so that the signal going to the components stays within the expected parameters which is usually plus or minus 5% since the power equation is p equals IV if the voltage is mostly fixed and the load demands more power then the power supply has to be able to vary its output quick current this is pretty basic stuff as for where transient comes back in we're referring to that as basically microsc to milliseconds Max for spikes in current one thing we've noticed in our own testing in the past and this also tracks with what Nvidia has told us it notices in its own testing is that the higher FPS but still GPU bound scenarios are more likely to cause undesirable effects like this one or just a crash like we earlier now it's not always that way but it's a correlation where you're more likely to find one of these problems in those GPU bound High frame rate scenarios whenever the GPU needs more power to operate more of its functional units simultaneously there's a sudden power Spike once that scen subsides the power needed to drive all the different contributing blocks of the GPU drops back to normal so this stuff we've just described that's an example of transient load but there's also transient response the voltages are most mostly fixed values at 3.3 5 and 12 volts so the temporarily spiked delivery comes in the form of larger amounts of current the current spikes that are delivered based on GPU demand are what we would call transient response when Nvidia started listing graphics card power with the 10 series The 1080 and the 108ti were rated at 180 and 250 Watts respectively at that point even with transients at two times the rated value a 600 wat power supply would still have Headroom to power the rest of the components by the time the 20 series Ser rolled in the 2080 power Target shifted up to 215 watt but the 2080 TI stayed locked in at 250 wat obviously different V bioses could do more but again 600 to 650 W power supplies would probably get you through even two times transients since the peak GPU power rating hadn't changed much but the 30 series and upcoming 40 series are different GPU power for the 3080 and the 380 TI are now listed at 320 and 350 watts and that's before counting for custom V biosis that's a 100 watt increase in average power consumption but a potential 200 watt increase in transient overhead requirements if you're looking at that 2x type Spike now the suggested power supplies are listed at 750 watts these days so if a 2X transient occurred at the 350 watt rating that only sort of leaves 50 watts of power for the rest of the system but not really that's not actually how it works a 750 W power supply might be able to power up to 975 watts of power for example for a short period of time before op trips or overpow protection we've shown this in our power supply benchmarks and reviews in the past transients live up to their name they come and go rapidly and can usually be mitigated by quality internal power supply components but that's the problem manufacturers who compete on price aren't interested in quality and as long as it looks like the blame is somewhere else they don't care if your system shuts down because of something happening as a result of the video card it's easier to just just pass the buck and say it's nvidia's fault and vice versa from Nvidia to the PSU maker now one of our sources at a power supply Manufacturing Company confirmed our findings and noted two to 2 and 1/2x transient spikes for existing 30 series cards they of course complained about this and said that it's coming back to power supply manufacturers when customers have problems getting the system to stay up but they also noted to us a significant concern about the RTX 40 series where the manufacturer stated that part of the reason there have been some delays in the past might be related to trying to deal with transient loads from the GPU in the 40 series because they're looking at upwards of 2 and A2 to 3x for the transient spikes if that happens and if the rumors that have been aligning to show a 400 watt to 450 watt GPU in the 4D series come to be true you're looking at potential transient spikes upwards of a kilowatt that's crazy now to be fair that might not come to pass the cards aren't out yet Nvidia has time to correct for this and try and deal with it if they can after receiving several viewer emails our research found many complaints online revolving around 750 to 850 wat power supplies when matched with 3080s 309s and 6900xt gpus from AMD the common theme was Graphics intensive applications power supplies around the Nvidia recommend 750 watts and unexpected shutdowns and restarts we even had our own lead from an EVGA 380 crashing during NZXT H1 V2 testing when this failure happened Patrick on our team doing the case testing sent the power supply over to Patrick Stone to do some bench testing with it and see if he could recreate the issue Stone then came back with a detailed report that was internal only and we'll put that on the screen so this allows you to see some of our internal validation processes that don't always get published publicly because we use them just to make sure we're doing things right in other parts of the operation but the lab report pretty much tells the story and the power supply shown in that clip that's an NZXT sff power supply that'll be relevant later not because of NZXT this time but because of small form factors so we'll come back to that so we need to talk numbers and then we'll talk about methods this time the testing we did over 40 different combinations of hardware and software in order to look for these types of spikes and potential failures and this took several weeks of research and data analysis and that was just for NVIDIA 30 series cards this also happens at least to some extent on AMD cards we only looked at that a little bit for now we'll look at that more if it starts become more of a problem with the 7,000 series uh and certainly we'll be looking at the 40 series but for now the focus today is on the 30 series because it took weeks and weeks just to do that alone okay so all of this needs testing and data our testing fall into four different stages for this project and we ended up using some special equipment that we bought as well so we employed some 10 to 100 amp clamping probes we also used an 800 khz sampling for the probes in the connection we recorded 3 million data points across 12 milliseconds so that's about 1.25 micros per data point and we'll talk more about that later as well and all of these equipment purchases were made possible by your support in our community on store. Camas nexus.net grabbing things like the modmat the toolkits the coaster packs and functional products from us as well as our shirts or on patreon.com Gamers Nexxus for bonus videos thank you for making it easy for us to buy unique and specific testing hardware for specific scenarios once we could consistently capture the transients our next goal was to determine what software could be used to trigger those so for this we were looking to reliably produce transients it was a slightly variable Target which made it difficult to dial in and we end up using a few different pieces of software a few games and some other stuff more synthetic like furmark or the heaven Benchmark for something a bit older and uh we also wanted to determine which gpus showed the worst transients so again we come back to the 40 different combined scenarios the first chart we'll look at comes from Doom Eternal testing we experimented with lower resolutions that produced over 500 FPS 4K resolution that bogged down the frame rates and we toggled rate tracing on and off in every scenario the highest transient spikes were consistently produced in 4k with rate tracing on so that's what we're showing in this chart the gigabyte 3090 stole the top spot for the highest transient spikes the 100 microc Peak average is 129% higher than the average power consumption shown in just firmar meaning the power over time plotted with something like nvidia's peacat or with a a normal current clamp that's slower while gaming in 4k this card can draw more than double its stable average power consumption so the doubling again won't matter for your power bill because it's only 100 micros lawn at a time at at most normally but it will matter to the power supply and the motherboard choice because that's where the failures happen the 3090 TI was only 14 wats lower than the gigabyte 390 and its peak average 100 microc measurement potentially due to better power design on the EVGA board or better firmware controls since we were focusing on the behavior of transients and their impact on system stability we ended up using the gigabyte 390 in our initial shutdown tests the 2 3080s behaved similarly with the Fe hitting a 4. % higher Peak average than the eagle cards in the same GPU family like 238s don't necessarily draw Power the same way just because they have the same GPU the board matters look at this 1080 TI it's been a little while since it was in the Limelight but it's back the reason we threw it in this mix is because we wanted you to see that huge transients aren't something new they've been around for at least three generations and definitely for longer but it matters more as Baseline power draw grows because we approach the power capacity people normally have Rainbow Six Siege also showed large spikes the 390 TI surpassed the 390 with a 2.6% higher 100 microsc Peak average the pair of 380s show up next with the eagle hitting 1.6% higher than the 380 Fe and the 100 microc measurement finally the 1080ti still shows significant transient spikes across 100 micros so again it's not a new problem it's just that Baseline is new so for example this at the 1080 TI number ends up about half of the 390 TI's Baseline average we did all of this testing against several power supplies as well and a couple of motherboards the answer to why one power supply might survive a transient Spike and why another shuts down lies with the internal components output filtering caps help to buffer against periods of heavy current draw as they act as reservoirs for electrical charge the higher the capacitance rating the more charge the capacitor can store and generally the better it can withstand a rapid discharge for the most part you don't need to be laser focused on these capacitors but it could be used as a determining Factor if two power supplies were otherwise comparatively almost completely the same and you were trying to decide which one to purchase of course it gets a little ridiculous if you're shopping by capacitor and there's a lot more to it than that so it's not a great solution anyway now small form facture power supplies are at more risk against transient spikes in terms of being knocked offline getting shut down from an ocp trip or something and the reason for that is there's just simply not as much room in a box as small for additional output capacitors and further still if the small form factur power supply is trying to be a budget small form factured power supply there's more chance we'll have trouble with those higher end gpus specifically and it could be an 850 watt or 750 wat power supply but if there's a significant Spike and it's cheap and it's small even though 850 is more than reasonable for the average power consumption of any GPU out in the consumer market today that will still be enough with a sufficient transient spike in that scenario to shut it down which is sort of what we're talking about today just to briefly go over the simple aspects of our testing we have three new current clamps that are going with our existing oscilloscope setup the oscilloscope already had standard 150 MHz 1X 10x probes and now we have the three new clamps that allow us to sample up to 800 khz which is awesome now these can be set to read as low as 10 amps or for higher load scenarios like what we're testing today they can go up to 100 amps the reason we bought three of them isn't just because some video cards have three pcie cables it's actually because as a standard with current clamps if you've used them you know this already the accuracy of the current clamp goes down or the range for error plus or minus the current reading uh widens as the current goes up or as it approaches the maximum range of the clamp so having three of them regardless of how many cables there are as long as they're used properly to allows us to dial in the current reading to be sort of more towards the middle of the spec of the clamp and allow better accuracy so as current flows through the wire it creates a proportional magnetic field perpendicular to the wire an iron alloy core in the clamp picks up on this magnetic field and concentrates it so that a connected hall effect sensor can detect it the hall effect sensor then generates an output signal that's Amplified and sent to the oscilloscope so that we can get an idea of how much current is Flowing without having to slice the wire open non-invasive probing is great for temporary test setups like this so we set our oscilloscope here to capture 3 million data points across 12 millisecond windows and because the clamps have an 800 khz limitation again that's about 1.25 uh microseconds per data point but that's actually still plenty good at this resolution we're going to be able to see the problems or the transient spikes on video cards and the real challenge we ran into actually was handling the data exporting it into some kind of software like Excel and trying to process it it was cumbersome so we had to build some uh some tools behind the scenes to deal with that as well another interesting aspect that we don't have airtime for on the main channel to go into today was getting the triggering right for each of these current clamps because uh sometimes the amount of current that's detected in a wire depending on the PCB design is higher for for Cable One versus 2 versus three and vice versa and this means that we have to manually check each cable for a trigger range to determine where it needs to detect and sort of capture data on the oscilloscope so we spent some time doing that the only other part of the oscilloscope setup was measuring the 12 volts output so we could get an accurate total power measurement we back probed the pcie output from the power supply to have a close to time synchronized Power number we wanted to know what was happening with the voltage when the current spiked and we hypothesized that the current spikes would coincide with the voltage sags and we wanted to capture the information to see if the sags were significant enough to matter in the final measurement we also attempted to take into account that the hall effect sensors had latency in their measurement so we time shifted the voltage results to compensate if we saw that the voltage had a variance of greater than 1% over the interval we actually generated don't even know like a dozen other charts with all kinds of different data and Analysis some cool stuff but we ultimately just pulled a few for this video cuz it sort of told the story more or less in its entirety and we were trying to keep this one a little more palatable for everyone but if there's enough interest maybe we'll run some extra data in a GNX video you can subscribe below so hopefully we've shed some light on what a transient is and how it works look realistically your computer is going through these all the time anyway it's normal power supplies are built to handle it to a certain extent and the components that sort of are responsible for these spikes are normally built in a way where they're not too exaggerated or extreme the thing we're trying to point out to you this isn't like a panic go by a new power supply situation and certainly if your computer's running fine now you have nothing to worry about now with this build but if you are having weird ocp trips if you're having the system click and shut down and you don't understand why and it's not short circuiting I mean like literally short circuit protection tripping then this might be why and we've gotten a lot of complaints from people talking about their system shutting down randomly and they ask us for help troubleshooting we're pretty sure it comes back to this so hopefully this will help a lot of you who've been asking about this but the real thing we wanted to get ahead of is the 40 series and whatever AMD is doing with its next series of gpus because power consumption is going up right now it tends to go in waves with the 400 500 series cards like fery from Nvidia or extremely high power consumption and running hot then it came back down for a while they got more efficient now we're on the upswing again for power draw so with transient spikes potentially being what manufacturers of power supplies are worried about to 2.5x or something for the 40 series if that comes to pass uh or certainly some of the cheaper boards it will be a problem then you're going to to run into more issues with power supply Choice one of the interesting things though talking with a few of the power supply manufacturers is that certain motherboards seem pretty resilient to this behavior and that's great but the problem is it gets ridiculous for anyone reviewer or consumer to try to put together this big complex Matrix of power supply board GPU that all play well together and it gets easier to just over Buy on the power supply which isn't great either because a little bit wasteful but it's more of a sure bet so in our testing we were able to consistently create transient current spikes through benchmarking and gaming we didn't find anything significant enough to alter the monthly power bill but we did lock in reproducible small form factor power shutdown scenarios and we learned some things that might help us understand how transients could affect future GPU and power supply purchases for the most part the 10 series wasn't a problem since overall power consumption was lower the 30 series seems to have had some problems with high transients but these appeared to be limited to specific lines of power supplies like seic being touchy with ocp for example and an EVGA GA line that had an initial problem that was quickly corrected in our testing we ran into some issues with small form factur power supplies and it's definitely possible that they might not be a good fit with any 380 or higher GPU unless they're at least 850 Watts if you have one that's working well though and it's below that with a 3080 or higher post a com let us know what it is cuz that would be helpful for everyone and that's what it comes down to this is really complex as a topic there are a lot of combinations of Hardware uh and just as a community we need to see what people are using so if you have issues with your system failing it'd be great in the the manner described here it'd be great to know what you're using for motherboard power supply and GPU and if you were just staying up but you're on a small form factor power supply 3080 plus uh and under 850 wats that'd be good to know too so anyway this isn't right now the end of the world with the 30 series it's already been out for a few years you'd know about it if you had a problem but hopefully this helps people who've had shutdowns they can't explain it should point you towards the power supply and the GPU frankly but power supply is easier to replace than GPU and then uh for the 40 series just prepare to buy up a little bit if this trend continues especially as we approach tdps of 400 watts or whatever Nvidia ends up cooking up because that's where you're going to start getting into territory where suddenly you're 700 750 wat power supply that's mid-range or lower end it might not be able to soak a transient Spike that's significant so just trying to set the stage get everyone ready for what's coming up thanks for watching this was fun research for us we have a lot more this we want to do in our testing lab and setup we've got the Sound Chamber up and running now some really cool data coming out as always our goal with this type of video is more to help people learn something and then troubleshoot better plan better when they're building let us know if we've achieved that in the comments below and subscribe for more as always you can go to store. Gamers access.net or patreon.com Gamers access to helps out directly thanks for watching we'll see you all next timethere's a Confluence of problems coming to a head between GPU makers GPU makers like Nvidia who are forced to pull more and more power to get competitive generational gains and power supply makers like these and the problem is that these two groups shift the blame between each other when a customer system does something like this and that's why today we're going to investigate transient power consumption spikes where the power consumption of a GPU for example might jump from 291 Watts over two times to 664 watts and that's just what today's gpus it's going to get worse with the RTX 40 series before that this in-depth video is brought to you by us and store. Gamers nexus.net we self-funded this huge project with your support from the GN store we just got a huge restock of our highquality red and black HUD Mouse mats which are desk-sized mousing services with custom Red Rubber Underside red stitching for anti Fray and a high resolution print we also just restocked our top selling product the black and blue wireframe Mouse mats which have CPU coolers Ram motherboards and more depicted on a mat that's built to last if you want even more functional products our 10piece toolkits for GPU disassembly and PC maintenance recently got a 7-year retroactive warranty and also have autographed variant visit store. gamas access.net to help us out on our next piece like this recent rumors have painted a concerning picture for the future of gpus or at least for the feature of the Nvidia RTX 40 series where now the discussion is moving away from average power consumption and instead towards transient power spikes or basically massive spikes that are largely unaccounted for that may only last up to 100 micros seconds and yet are enough to in the blink of an eye shut down the syst system in basically the same way as ocp would or overcurrent protection so that's the problem that we're looking at for the future of video cards and it's something that has to be accounted for today this is actually something that's been building for a long time we've seen evidence as far back as the 10 series and maybe even before that the difference is that today's gpus are being designed to nominally draw far more power and therefore the transient spikes are far more likely to knock a power supply off line and shut it down and so the fastest recap possible is that this issue looks something like this we can show all kinds of charts and oscilloscope shots to detail this issue and we will today but ultimately from the user experience end that's pretty much how it looks if it's the Ron mix of CPU GPU and motherboard then you might end up with a very difficult to troubleshoot problem that just trips power sometimes it's hard to pinpoint which exact component is causing that the power supply might be an easy fix but the motherboard 2 is involved so the system will just shut down and you'll lose anything unsaved or progress and whatever you're working on and the problem is that it is a mix of all these things the real issue is that it's not just as simple as it used to be for the most part you can take the average power consumption of the CPU and the GPU and figure out an appropriate power supply for those two now however you should also be baking an additional Headroom to your power supply selection for potential transient spikes especially at the high end with the RTX 30 series and probably the RTX 40 series as well in our lab testing here on this machine and just with real world test benches and computers we've seen that some motherboards handle this far better than others we've also seen that some video cards handle this better than others and some power supplies so it's an intermittent problem and that's frustrating for consumers without a lab full of equipment to just identify where it came from that's where we come in here's the issue we'll explain this chart in more detail a little bit later but we wanted to get this up front too this chart shows GPU power consumption under different scenarios in the game Doom Eternal the RTX 390 Master pulled 291 Watts under averaged power draw over an extended period And this is true with either a slower current clamp or with nvidia's power draw measurement tools like the PCAT analyzer but its blue and red bars paint another story the same card when tested with our oscilloscope instead of Which is far faster pulled 664 Watts at its peak transient Spike that's enough to put serious strain on power supplies that can't soak the temporary burst it's a 129% increase in power drop more than double from the Baseline average power consumption and that shows a major part of the story of these cards getting left out in at least our testing which we're working on addressing starting with today's video where now we and a lot of other reviewers mostly rely on average power consumption metrics that's enough to sort of estimate how much the computer will cost to run or at least the video card will cost to run it's enough to do a head-to-head comparison for overall power efficiency or performance per watt it's also enough to estimate the heat being produced by the device because ultimately the average overuse is what matters for all of those scenarios what it's not enough for is to know what the chances are of a power supply that's on the border of supporting this particular card to be able to stay up and not have an ocp drop so that existing testing methodology it has to change and that's what we've been working on with this for a couple of months now just improving behind the scenes because it cuts out the entire part of the story where you're looking at transient spikes that are probably the source of a lot of systems just shutting down out there where people are posting on forums having difficulty pinpointing the exact issue and no one wants the issue to be their video card they don't want to blame the video card because they had to fight hard for it and it was expensive so everybody wants the problem to be a cheaper component or an easier to Source One like a power supply or a motherboard you can get lucky with that but the GPU is sort of ultimately at the center of attention and this is why we continually overhaul our own testing processes to capture more of the picture of the component we're reviewing now back to this issue small form factur power supplies will be the most at risk of sinking under this demand as we saw in our NZXT H1 V2 testing with a highend CPU and highend GPU the small for fact power supply just couldn't take it small formfactor power supplies in particular struggle more with this because they have more limited real estate to actually build out the power supply and therefore reduced capability of dealing with these sudden spikes they're more likely to just shut down from ocp so that's the preview and now it's time to talk about how all of this works and what it means for you building computers and what it means for the industry talking about power supply manufacturers they're being told by Nvidia to bulk upon capacitance and better technology capabilities to deal with these transient spikes especially from the RTX 40 series where we're looking at Baseline power targets of around 450 Watts based on some of the leaks that are out there now for higher end cards and that's just the Baseline if you're spiking a couple times that you could see where it starts to introduce problems now power supply makers of course think this is nvidia's Fault Nvidia thinks it's the power supply manufacturer's issue to deal with and both of them have reasons to push the blame onto the other one power supply makers generally think it's unreasonable for NVIDIA to push all of the requirement of dealing with transient spikes onto them and think instead that Nvidia should be doing better at a GPU and a board level to supress those transient spikes but power supply makers also have other reasons to push back against Nvidia okay so here's what transients actually are we already mentioned that 30 series video cards can tax your power supply with Beyond two times the specified power load of the gpus on average and these quick periods of power consumption are often referred to as transients not transients but transient the term transient refers to a short-lived oscillation in a system caused by a sudden change of voltage or current or load in this case the transient is driven by the load which is the GPU and that manifests itself in the current we can measure this in the wires of a power supply cable by detecting the magnetic field in and around the wire and if you invest in high-end test clamps like we did you can isolate those spikes even though they last mere micros seconds in many scenarios the voltage from a power supply is carefully regulated so that the signal going to the components stays within the expected parameters which is usually plus or minus 5% since the power equation is p equals IV if the voltage is mostly fixed and the load demands more power then the power supply has to be able to vary its output quick current this is pretty basic stuff as for where transient comes back in we're referring to that as basically microsc to milliseconds Max for spikes in current one thing we've noticed in our own testing in the past and this also tracks with what Nvidia has told us it notices in its own testing is that the higher FPS but still GPU bound scenarios are more likely to cause undesirable effects like this one or just a crash like we earlier now it's not always that way but it's a correlation where you're more likely to find one of these problems in those GPU bound High frame rate scenarios whenever the GPU needs more power to operate more of its functional units simultaneously there's a sudden power Spike once that scen subsides the power needed to drive all the different contributing blocks of the GPU drops back to normal so this stuff we've just described that's an example of transient load but there's also transient response the voltages are most mostly fixed values at 3.3 5 and 12 volts so the temporarily spiked delivery comes in the form of larger amounts of current the current spikes that are delivered based on GPU demand are what we would call transient response when Nvidia started listing graphics card power with the 10 series The 1080 and the 108ti were rated at 180 and 250 Watts respectively at that point even with transients at two times the rated value a 600 wat power supply would still have Headroom to power the rest of the components by the time the 20 series Ser rolled in the 2080 power Target shifted up to 215 watt but the 2080 TI stayed locked in at 250 wat obviously different V bioses could do more but again 600 to 650 W power supplies would probably get you through even two times transients since the peak GPU power rating hadn't changed much but the 30 series and upcoming 40 series are different GPU power for the 3080 and the 380 TI are now listed at 320 and 350 watts and that's before counting for custom V biosis that's a 100 watt increase in average power consumption but a potential 200 watt increase in transient overhead requirements if you're looking at that 2x type Spike now the suggested power supplies are listed at 750 watts these days so if a 2X transient occurred at the 350 watt rating that only sort of leaves 50 watts of power for the rest of the system but not really that's not actually how it works a 750 W power supply might be able to power up to 975 watts of power for example for a short period of time before op trips or overpow protection we've shown this in our power supply benchmarks and reviews in the past transients live up to their name they come and go rapidly and can usually be mitigated by quality internal power supply components but that's the problem manufacturers who compete on price aren't interested in quality and as long as it looks like the blame is somewhere else they don't care if your system shuts down because of something happening as a result of the video card it's easier to just just pass the buck and say it's nvidia's fault and vice versa from Nvidia to the PSU maker now one of our sources at a power supply Manufacturing Company confirmed our findings and noted two to 2 and 1/2x transient spikes for existing 30 series cards they of course complained about this and said that it's coming back to power supply manufacturers when customers have problems getting the system to stay up but they also noted to us a significant concern about the RTX 40 series where the manufacturer stated that part of the reason there have been some delays in the past might be related to trying to deal with transient loads from the GPU in the 40 series because they're looking at upwards of 2 and A2 to 3x for the transient spikes if that happens and if the rumors that have been aligning to show a 400 watt to 450 watt GPU in the 4D series come to be true you're looking at potential transient spikes upwards of a kilowatt that's crazy now to be fair that might not come to pass the cards aren't out yet Nvidia has time to correct for this and try and deal with it if they can after receiving several viewer emails our research found many complaints online revolving around 750 to 850 wat power supplies when matched with 3080s 309s and 6900xt gpus from AMD the common theme was Graphics intensive applications power supplies around the Nvidia recommend 750 watts and unexpected shutdowns and restarts we even had our own lead from an EVGA 380 crashing during NZXT H1 V2 testing when this failure happened Patrick on our team doing the case testing sent the power supply over to Patrick Stone to do some bench testing with it and see if he could recreate the issue Stone then came back with a detailed report that was internal only and we'll put that on the screen so this allows you to see some of our internal validation processes that don't always get published publicly because we use them just to make sure we're doing things right in other parts of the operation but the lab report pretty much tells the story and the power supply shown in that clip that's an NZXT sff power supply that'll be relevant later not because of NZXT this time but because of small form factors so we'll come back to that so we need to talk numbers and then we'll talk about methods this time the testing we did over 40 different combinations of hardware and software in order to look for these types of spikes and potential failures and this took several weeks of research and data analysis and that was just for NVIDIA 30 series cards this also happens at least to some extent on AMD cards we only looked at that a little bit for now we'll look at that more if it starts become more of a problem with the 7,000 series uh and certainly we'll be looking at the 40 series but for now the focus today is on the 30 series because it took weeks and weeks just to do that alone okay so all of this needs testing and data our testing fall into four different stages for this project and we ended up using some special equipment that we bought as well so we employed some 10 to 100 amp clamping probes we also used an 800 khz sampling for the probes in the connection we recorded 3 million data points across 12 milliseconds so that's about 1.25 micros per data point and we'll talk more about that later as well and all of these equipment purchases were made possible by your support in our community on store. Camas nexus.net grabbing things like the modmat the toolkits the coaster packs and functional products from us as well as our shirts or on patreon.com Gamers Nexxus for bonus videos thank you for making it easy for us to buy unique and specific testing hardware for specific scenarios once we could consistently capture the transients our next goal was to determine what software could be used to trigger those so for this we were looking to reliably produce transients it was a slightly variable Target which made it difficult to dial in and we end up using a few different pieces of software a few games and some other stuff more synthetic like furmark or the heaven Benchmark for something a bit older and uh we also wanted to determine which gpus showed the worst transients so again we come back to the 40 different combined scenarios the first chart we'll look at comes from Doom Eternal testing we experimented with lower resolutions that produced over 500 FPS 4K resolution that bogged down the frame rates and we toggled rate tracing on and off in every scenario the highest transient spikes were consistently produced in 4k with rate tracing on so that's what we're showing in this chart the gigabyte 3090 stole the top spot for the highest transient spikes the 100 microc Peak average is 129% higher than the average power consumption shown in just firmar meaning the power over time plotted with something like nvidia's peacat or with a a normal current clamp that's slower while gaming in 4k this card can draw more than double its stable average power consumption so the doubling again won't matter for your power bill because it's only 100 micros lawn at a time at at most normally but it will matter to the power supply and the motherboard choice because that's where the failures happen the 3090 TI was only 14 wats lower than the gigabyte 390 and its peak average 100 microc measurement potentially due to better power design on the EVGA board or better firmware controls since we were focusing on the behavior of transients and their impact on system stability we ended up using the gigabyte 390 in our initial shutdown tests the 2 3080s behaved similarly with the Fe hitting a 4. % higher Peak average than the eagle cards in the same GPU family like 238s don't necessarily draw Power the same way just because they have the same GPU the board matters look at this 1080 TI it's been a little while since it was in the Limelight but it's back the reason we threw it in this mix is because we wanted you to see that huge transients aren't something new they've been around for at least three generations and definitely for longer but it matters more as Baseline power draw grows because we approach the power capacity people normally have Rainbow Six Siege also showed large spikes the 390 TI surpassed the 390 with a 2.6% higher 100 microsc Peak average the pair of 380s show up next with the eagle hitting 1.6% higher than the 380 Fe and the 100 microc measurement finally the 1080ti still shows significant transient spikes across 100 micros so again it's not a new problem it's just that Baseline is new so for example this at the 1080 TI number ends up about half of the 390 TI's Baseline average we did all of this testing against several power supplies as well and a couple of motherboards the answer to why one power supply might survive a transient Spike and why another shuts down lies with the internal components output filtering caps help to buffer against periods of heavy current draw as they act as reservoirs for electrical charge the higher the capacitance rating the more charge the capacitor can store and generally the better it can withstand a rapid discharge for the most part you don't need to be laser focused on these capacitors but it could be used as a determining Factor if two power supplies were otherwise comparatively almost completely the same and you were trying to decide which one to purchase of course it gets a little ridiculous if you're shopping by capacitor and there's a lot more to it than that so it's not a great solution anyway now small form facture power supplies are at more risk against transient spikes in terms of being knocked offline getting shut down from an ocp trip or something and the reason for that is there's just simply not as much room in a box as small for additional output capacitors and further still if the small form factur power supply is trying to be a budget small form factured power supply there's more chance we'll have trouble with those higher end gpus specifically and it could be an 850 watt or 750 wat power supply but if there's a significant Spike and it's cheap and it's small even though 850 is more than reasonable for the average power consumption of any GPU out in the consumer market today that will still be enough with a sufficient transient spike in that scenario to shut it down which is sort of what we're talking about today just to briefly go over the simple aspects of our testing we have three new current clamps that are going with our existing oscilloscope setup the oscilloscope already had standard 150 MHz 1X 10x probes and now we have the three new clamps that allow us to sample up to 800 khz which is awesome now these can be set to read as low as 10 amps or for higher load scenarios like what we're testing today they can go up to 100 amps the reason we bought three of them isn't just because some video cards have three pcie cables it's actually because as a standard with current clamps if you've used them you know this already the accuracy of the current clamp goes down or the range for error plus or minus the current reading uh widens as the current goes up or as it approaches the maximum range of the clamp so having three of them regardless of how many cables there are as long as they're used properly to allows us to dial in the current reading to be sort of more towards the middle of the spec of the clamp and allow better accuracy so as current flows through the wire it creates a proportional magnetic field perpendicular to the wire an iron alloy core in the clamp picks up on this magnetic field and concentrates it so that a connected hall effect sensor can detect it the hall effect sensor then generates an output signal that's Amplified and sent to the oscilloscope so that we can get an idea of how much current is Flowing without having to slice the wire open non-invasive probing is great for temporary test setups like this so we set our oscilloscope here to capture 3 million data points across 12 millisecond windows and because the clamps have an 800 khz limitation again that's about 1.25 uh microseconds per data point but that's actually still plenty good at this resolution we're going to be able to see the problems or the transient spikes on video cards and the real challenge we ran into actually was handling the data exporting it into some kind of software like Excel and trying to process it it was cumbersome so we had to build some uh some tools behind the scenes to deal with that as well another interesting aspect that we don't have airtime for on the main channel to go into today was getting the triggering right for each of these current clamps because uh sometimes the amount of current that's detected in a wire depending on the PCB design is higher for for Cable One versus 2 versus three and vice versa and this means that we have to manually check each cable for a trigger range to determine where it needs to detect and sort of capture data on the oscilloscope so we spent some time doing that the only other part of the oscilloscope setup was measuring the 12 volts output so we could get an accurate total power measurement we back probed the pcie output from the power supply to have a close to time synchronized Power number we wanted to know what was happening with the voltage when the current spiked and we hypothesized that the current spikes would coincide with the voltage sags and we wanted to capture the information to see if the sags were significant enough to matter in the final measurement we also attempted to take into account that the hall effect sensors had latency in their measurement so we time shifted the voltage results to compensate if we saw that the voltage had a variance of greater than 1% over the interval we actually generated don't even know like a dozen other charts with all kinds of different data and Analysis some cool stuff but we ultimately just pulled a few for this video cuz it sort of told the story more or less in its entirety and we were trying to keep this one a little more palatable for everyone but if there's enough interest maybe we'll run some extra data in a GNX video you can subscribe below so hopefully we've shed some light on what a transient is and how it works look realistically your computer is going through these all the time anyway it's normal power supplies are built to handle it to a certain extent and the components that sort of are responsible for these spikes are normally built in a way where they're not too exaggerated or extreme the thing we're trying to point out to you this isn't like a panic go by a new power supply situation and certainly if your computer's running fine now you have nothing to worry about now with this build but if you are having weird ocp trips if you're having the system click and shut down and you don't understand why and it's not short circuiting I mean like literally short circuit protection tripping then this might be why and we've gotten a lot of complaints from people talking about their system shutting down randomly and they ask us for help troubleshooting we're pretty sure it comes back to this so hopefully this will help a lot of you who've been asking about this but the real thing we wanted to get ahead of is the 40 series and whatever AMD is doing with its next series of gpus because power consumption is going up right now it tends to go in waves with the 400 500 series cards like fery from Nvidia or extremely high power consumption and running hot then it came back down for a while they got more efficient now we're on the upswing again for power draw so with transient spikes potentially being what manufacturers of power supplies are worried about to 2.5x or something for the 40 series if that comes to pass uh or certainly some of the cheaper boards it will be a problem then you're going to to run into more issues with power supply Choice one of the interesting things though talking with a few of the power supply manufacturers is that certain motherboards seem pretty resilient to this behavior and that's great but the problem is it gets ridiculous for anyone reviewer or consumer to try to put together this big complex Matrix of power supply board GPU that all play well together and it gets easier to just over Buy on the power supply which isn't great either because a little bit wasteful but it's more of a sure bet so in our testing we were able to consistently create transient current spikes through benchmarking and gaming we didn't find anything significant enough to alter the monthly power bill but we did lock in reproducible small form factor power shutdown scenarios and we learned some things that might help us understand how transients could affect future GPU and power supply purchases for the most part the 10 series wasn't a problem since overall power consumption was lower the 30 series seems to have had some problems with high transients but these appeared to be limited to specific lines of power supplies like seic being touchy with ocp for example and an EVGA GA line that had an initial problem that was quickly corrected in our testing we ran into some issues with small form factur power supplies and it's definitely possible that they might not be a good fit with any 380 or higher GPU unless they're at least 850 Watts if you have one that's working well though and it's below that with a 3080 or higher post a com let us know what it is cuz that would be helpful for everyone and that's what it comes down to this is really complex as a topic there are a lot of combinations of Hardware uh and just as a community we need to see what people are using so if you have issues with your system failing it'd be great in the the manner described here it'd be great to know what you're using for motherboard power supply and GPU and if you were just staying up but you're on a small form factor power supply 3080 plus uh and under 850 wats that'd be good to know too so anyway this isn't right now the end of the world with the 30 series it's already been out for a few years you'd know about it if you had a problem but hopefully this helps people who've had shutdowns they can't explain it should point you towards the power supply and the GPU frankly but power supply is easier to replace than GPU and then uh for the 40 series just prepare to buy up a little bit if this trend continues especially as we approach tdps of 400 watts or whatever Nvidia ends up cooking up because that's where you're going to start getting into territory where suddenly you're 700 750 wat power supply that's mid-range or lower end it might not be able to soak a transient Spike that's significant so just trying to set the stage get everyone ready for what's coming up thanks for watching this was fun research for us we have a lot more this we want to do in our testing lab and setup we've got the Sound Chamber up and running now some really cool data coming out as always our goal with this type of video is more to help people learn something and then troubleshoot better plan better when they're building let us know if we've achieved that in the comments below and subscribe for more as always you can go to store. Gamers access.net or patreon.com Gamers access to helps out directly thanks for watching we'll see you all next time\n"