Crazy Efficient: AMD Threadripper 7980X & 7970X CPU Review & Benchmarks

[Music] this is one of our densest CPU benchmarks and reviews we've ever done and thread Ripper actually has some crazy efficiency it was enough to get the Whiteboard out we'll talk about that later today we're digging deep into the 79 70x and 7980x threader for parts for performance and later today we're going to be live streaming an overclocking session with a 96 core CPU with Bill and Amit to set some world records with liquid nitrogen we don't have a time yet so you'll have to keep an eye out for notifications you can hit the YouTube Bell unironically to catch it these are amd's new 64 core and 32 core parts for the TRX 50 motherboard and even if you're not in the market for them the performance is fun to look at and that's the biggest point of contention today it's the price also that AMD kind of abandoned thread Ripper last time so they'll have to earn that trust back it's been a while since AMD has launched an HDT thread Ripper part without the pro name attached to it buyers might also be hesitant to buy into tx50 or WR rx90 after the last last round the 3970x launched at $2,000 and is being replaced by the 79 70x at 2500 the 7980x comes in where the 3990x was previously except it's $5,000 from a technology standpoint though the efficiency is the most interesting story today power consumption thermals and efficiency actually and we're focusing our testing on HDT thread Ripper parts let's get started we brought you this video with a sale for 10% off all tools on store. gamers nexus.net this insane amount of testing and editing time is made possible by our audience for a limited time you can get 10% off of our projects and soldering mats which are heavyduty thick silicone mats for repairs or projects even things like model building you can also get that deal on our large anti-static mod mats for your next PC build perfect for a giant thread Ripper machine or you could buy one of our medium mod mats with enamel pin or a 10piece GPU tear down tool kit grab a good work service or tool for yourself and support our testing at the same time at store. Gamers access.net there's a massive amount of testing in this video today so we're going to keep it really focused in the intro and outro here they're going to be slim that's because we want to leave the air time available for everything else so uh for the testing we've introduced a number of new workstation tests these include spec workstation as a test Suite but also an extra focus on some of the existing tests that we have for workstations Intel has competing Parts kind of so Intel hasn't really made an HDT part as in a non Pro sort of Enthusiast grade x-class CPU in a long time the the 10980xe was maybe the most famous recent example but that line has basically been dead the xeons we don't have any testing for in this video today those would be possible competitors to the pr line of thread Ripper which we are not reviewing here we're looking at the HDT 64 core and 32 core parts today but we might add Zeon in the future if we do one of the thread reper Pro Parts the test platform we use for thread rep deviates a little bit from our standard test solution because it's just so different put that on the screen now this will also be in our article on Gamers access.net which will go live in a couple days specs for the new thread Ripper HDT and pro CPUs are on the screen as well if you want those in detail you can check the announcement post but as you skim these specs one important note these are not meant to be gaming CPUs gaming is a really small Focus for this review and I'm going to call you all out last time we reviewed a thread rer CPU was the 3960x and I went back and looked at the watch uh most watched parts of the video and everybody jumped to the gaming that's not what these CPS are meant for we're testing gaming to make sure they can play the games and that it's not awful as an experience that you don't have to go play games within bios and toggle things just to make it work but you do please do not buy these CPUs for a gaming PC it is a terrible waste of money if that's all you're gonna do with it uh and it'll be worse than if you bought something like a 7800x 3D or a 14900 K we just want to make sure everyone's putting their money in the best place for their needs like for example buying an entire computer for gaming for the cost of one thread R for CPU we'll start with the most interesting part which is power consumption and efficiency despite being in an absolute sense a relatively high power consumption these threader CPUs are actually some of the most efficient CPUs we've ever tested and in fact the 7980x with it 64 is the most efficient CPU we've ever run through our test Suite so it's a really interesting part of the discussion and one that we want to explore in depth today that we're going to be working through here for this testing we're measuring at the EPS 12vt cables before uh vrm efficiency losses so we're getting pure Hardware readings for the power consumption here's the chart for an all core blender workload where every single thread is pushed to 100% on these CPUs the 7980x pulled about 352 Watts all core with the 79 70x at 360 Watts that makes sense actually we found our 7980x to be a slightly better bin with lower voltage requirements when using Auto VC cor contributing to this small Gap the 750x pulled 250 watts in the same test so with 25% of the cores of the 7980x it pulled 71% of the power that's definitely an efficiency Advantage if you're actually leveraging those cores the prior HDT non-pro 3970x measured at 232 watts in the same test with Intel's old 10980xe at 191 Watts so for efficiency what we're doing is we're looking at a fixed unit of work so for a given frame to be rendered from an animation on the CPU uh what is the power required throughout that process then we get an energy consumption number at the end of it the efficiency is actually completely insane the 780x is the most efficient CPU we've ever tested the C9 50x with Eco Mode which isn't shown here could kind of compete with these numbers but that was eco mode while the power consumption here would make for an excellent headline the actual efficiency means a lot more to us it means that you'd use less energy in total to complete the same amount of work even as compared against CPS that consume technically less power and that includes the prior efficiency can the 5950 X so despite the 350 wat figure the 7980x is more efficient than the 300 watt CPUs buy a lot the 14900 K has 34.9 wat hour result whereas the 7980x it's at 12.9 the 7 97X isn't as efficient but still did pretty well with the 21.6 wat hour result remember that this will change based on the type of work as well of course and the utilization of those cores the 3970x of Y was also pretty efficient for its time actually at 20.1 wat hours we didn't make these charts back then so it's helpful to see it in this way but the 7980x genuinely impresses us for its power efficiency here here's why it's so efficient the 7980x specifically is at the best point in amd's volt frequency curve so it looks something like this and because of this you these CPUs with a 350 wat TDP that has amd's thread Ripper Parts as actually lower TDP than the server Parts before them so they have some room to play with here if we look at V Cor and the frequency for an all core workload in our testing the 7980x was actually shockingly low it tended to be about 0.84 to 0.845 Volts for V core so that puts us let's just call it 0.845 for frequency it was in the range of 3900 to 4,000 MHz so this means that AMD actually has a lot more room to boost if you wanted to throw away some efficiency you could do some overclocking or use PBO to make that tradeoff the 780x in our testing we found had a shockingly low uh approximately 3.5 watts per core for the sbi3 tfn Total Core Power through software not through uh clamp we found this to be around 223 or so Watts for the S so it was approximately 72 again through softare monitoring and some room in here for software error here's a table to make it easier than my handwriting the 64 cor CPU is operating at a lower frequency and as a result a much lower voltage also the Silicon quality is probably among the best on these higher core count CPUs the voltage though dropped from a range of 1.09 to 1.1 volts on the 7970 X to this 0.8 5ish result for the 7980x with software sensor loging we saw the 7 97X as cores were running any anywhere from 8 watts to almost 9 watts per core over double the 7980x per core power consumption that puts its svi 3 tfn core power around 270 Watts with s so at about 70 to 71 Watts let's talk briefly about thermal so in past years when we tested the original thread Ripper and following ones we found that full coverage versus a smaller contact patch actually did have a big impact on performance so we compared to relatively like for like noct air coolers back then for purposes of thermal testing here we're going to be looking at at what is possible with the cooler that AMD sent out with its review kits which is an acitech solution so it's got a smaller contact patch as you can see here and then they have an adapter plate which these have been around since the early thread Ripper days so we use that as well that gives us and we have a shot of it uh basically circular spread of thermal past right in the middle now we'll do separate testing in a followup for how much it really impacts things but for the discussion of thermals today it wasn't throttling and that's all that really matters for the benchmarks were running here it wasn't limiting the performance in any way here's the chart in an all core workload during a 30-minute blender render collected at steady state we found the 7980x is 8 ccds to run between 57.7 and 64° C at 21c ambient that's good overall we were able to keep the CPU in check with a 360 mm CLC at Max speeds and water blocks with full coverage would do great For Thread Ripper the 7 970x ran significantly warmer it's still fine but it is definitely higher at 80° maximum across the ccds and 742 for the minimum when averaged at steady state even though the CPU is technically lower end than the 64 core variant it's running hotter across the ccds let's take a look at the HDT layout the 7980x uses all eight of these ccds starting at CCD 0 and counting up to ccd7 the 7970 X populates CCD 0 CCD 1 2 and three according to AMD the reason we're seeing this thermal behavior is due to the density both of these CPUs measured 350 to 360 watts in our power testing that means that ISO power the 7 970x has all of its heat generated on the four innermost ccds in this diagram whereas the 7980x benefits from more surface area and lower heat and power density at the same power consumption the 7980x is spreading that power out over more cores and More Physical area this goes back to the per core power consumption we talked about earlier and that's why the 7 970x ran hotter than the 7980x in benchmarking for thermals now it's time to look at the production benchmarks blender serves as our first production test we use blender for tile based rendering so one tile is generated for each thread here's the chart the 7980x immediately impressed us with a 2.2 minute time to render a single frame of the GN intro animation logo that you watched for this video but we render it at a much higher quality and sample size for benchmarking the scaling here is clean the 7980x is 2.2 minute result haded about 1third of the render time of the 16 core 750x that's not perfectly linear but pretty close to it given the other differences the 79 70x required 3.6 minutes making the 7980x 39% faster for about 100% more money you're never getting value for these top-of-the-line Parts though compared to the 3970x is 5.2 minutes as its HDT non-pro predecessor the 7 970x required 31% less time to complete the render the 3970x was originally $2,000 so the increase is 25% more money for 31% Less render time the 14900 k for reference required 7.3 minutes giving the 750x a 12% lead the less relevant comparison by price but the pertinent one for today gives the 7970 x a 51% time requirement advantage and the 7980x a 70% Advantage as a final note we decided to just toggle the ECC modes we saw no difference for this test between the modes we've also historically noticed that this test isn't very memory dependent our next test is for code compile for this one we're compiling chromium on Windows a ton of you asked for us to bring this test back because we'd run it for years but recently we stopped running it we decided to bring it back in because so many people wanted to see it and For Thread Ripper it makes a lot of sense the 7980x completed this compile in the fastest time we've ever seen for our test Suite Maybe expectedly this is a highly threaded workload that can Leverage The cores the build time was 312 minutes here so it completed the compile in 27% less time than the 7970 X's 43 minute result there's definitely scaling but for the ultra high end the value doesn't linearly scale again it's not abnormal for a flagship part but it is something we really want to make sure everyone's aware of the 750x and x3d had times of 71 to 72 minutes and were roughly tied with the 14900 k at about about 71 minutes that means the 79 70x is doubling of core count over the 79 50x has it cut the time by about 40% it's close to linear as for the 3970x that one required 60 Minutes a clean hour which has the 7970 X about 28% improved finally ECC didn't seem to have an effect on the results for the 7980x when we toggled it on and off file decompression is next using szip as the test platform this test is measured in millions of instructions per second the 7980x posted insane numbers here clearing 500,000 mips the 504k result has it only about 1% ahead of the 797 X so there's no meaningful difference in this particular Benchmark we're hitting limitations of scaling on the 7980x which will look different actually with compression that said the 3970x illustrates that there is clear scaling with the 7970 X as a comparison at about 499,000 mips the 77x leads the 373k mips result of the 3970x by 34% the 7950 X's 276k result allowed the 77x to lead by about 81% once again jumping from the desktop 16 core to the HDT 32 core posts a result that's nearly linear that's pretty close to doubling considering the other differences within the chips it's possible that faster memory would benefit here it's also possible that we're just becoming bound elsewhere there's more scaling between these new parts in compression test in this one the 7980x performed 393,000 mips during compression or about 390k without ECC that's with an error it's less than 1% difference so they're about the same again the lead over the 770x is 12% here it's a larger Gap than decompression but still a lot less than we saw in blender and chromium the 750x performed 193,000 mips with x3d not really changing that much the 7 970x is about 82% ahead of it here with the 7980x with ECC at about 104% % ahead as for the 3970x it and the 3960x are within 1 and a half% of each other we went back and checked the results from the previous years that we ran this test and saw that it was the same there it looks like these two are encountering external limitations and so they perform about the same scaling of the 79 7x against these parts is clean with a 125% boost over the 3970x the 14900 K did well in this Benchmark running nearly at parody with the 750x our main takeaway here is that the modern high-end mainstream Parts like the 14900 K and 750x are extremely competitive for their price as compared to even higher end HDT Parts but when you need the performance the price might matter less to you and the parts get harder to make Adobe Photoshop using the Puget Suite is next in this one we get the reality check that more cores doesn't always equal more better Photoshop still likes frequency CPUs like the 750x and 14700 K leave this chart we need to rerun the 149 for this one though x3d also doesn't benefit this particular test the 79 70x scored 1449 points in aggregate allowing the 7 950x a lead of 11% over the 32 core part while costing a lot less money even the 7600 X did better than the 7970 X here the 7980x was worse still with the 1300 Point result Landing just barely above a 12600 k and a 5900 X but looking down the chart we see similar positioning with the 3960x and 3970x you shouldn't be buying these parts if your primary use case is Photoshop this shows some of the components that make up the extended aggregate score the GPU score here gives us an insight into more GPU Centric tasks within the Puget Photoshop test Suite shows us how much the CPU might be binding the gpu's capabilities the best performers have a GPU score of around 141 to 142 points this chart reveals one of two significantly weak aspects for the thread CPUs for Photoshop specifically it's limiting the GPU oriented Photoshop tasks looking at the general score it does much better the 7980x climbs from below the 5600x and filter performance that's still not worth buying for this kind of work but at least now we have a picture of where it was being held back this is our Adobe Premiere chart the 7980x is not on this chart because unfortunately it could not complete the test with the Puget Suite it does launch Premiere and it's capable of opening files it just repeatedly crashed during this test we're not sure if that's because of Puget su's execution or because of the CPU itself it could be the test rather than more of a native compatibility issue here though the 7 970x Brute Forces its way to the top at 918 points it leads the 7 950x by 8% that's not worth the price in nearly any instance but it does scale from a value perspective though that means the 750x achieves 93% of the performance for 22% of the price with the 4900 k at 92% of the performance of the 770x for 23% of the price the old 3970x Ran at 764 points giving the 79 70x an uplift of 18% we'd be more interested in that if it were a drop in upgrade but amd's abandonment of the previous threader per platform has cost it on the value proposition for smaller boost like this this could be justifiable if you could just put it in the same board but you can't so it's only valuable if you're buying this for some other work that better leverages it we wanted to break out a couple scores from the video editing testing to understand what was boosting the 79 70x the most that's partly because we're considering building a new system with it because we could use all the video production performance we can get obviously this shows the intr frame scoring the 77x has a strong and obvious lead here it's at 132 points which allows it to Gap the 7 950x by 11% and the 14900 K by 15% now in the sense of breaking out from from the average previous top of the chart it is a big jump in the sense of value it's not and you shouldn't buy this for only this work unless you were really severely performance limited and video production is part of how you make money threader also did well in raw score where the results previously got stuck clustered around the 14900 K 750x and 14 700k from 49 to 51 points the 7970 X's 57.7 Point result has it leading the 14900 K by 13% in working with raw footage the lead is similar against the 750x here it's not a huge jump again for doubling the cores and the price actually it's it's more than double let's go with over quadrupling the price over the son9 50x but for anyone who desperately needs the performance especially for professional paid work at least it gives an option it's just hard to justify that with the price jump and that's not even counting the motherboard next we want to look at the GPU effect score in video editing although these are GPU effects the CPU can govern the performance ceiling in this test the 14900 K and 750x were within error of each other and set the ceiling for the chart the 13900 K and 13700 K are also within error of each other and most of these results are so tightly packed that the differences between immediately adjacent results are often just one more pass away from swapping places the 797 X though dropped a few points at 97.4 for its score we see that GBU effects related performance remains a weaker aspect it's still achieving most of the maximum but it won't outperform the 14900 K or 750x in this specific type of test now we're going to get into spec workstation test so spec provides a number of test Suites that are prepackaged and they'll run things like computational fluid dynamics uh simulation software for um molecular movement and it also looks at biomedical and energy or oil all types of things Financial so running a suite that analyzes those different aspects that a threater per CPU might be good at uh and for this we use the default threads per process setting for all applicable tests and otherwise there are couple of these where we're not too familiar with the software itself some of them even have words we don't know because we're getting into biomedical but the numbers are there so if that means something to you you'll be able to look at the charts and extrapolate the meaning that hopefully helps guide you to the right CPU this set of Test shows the performance in various mathematical and computational fluid dynamics tools calculix is an open source tool that solves differential equations using finite element analysis WPC cfd is based on the open foam open source fluid Dynamic solver and rodinia cfd is described by spec workstation as quote an unstructured grid finite volume solver for 3D Oiler equations represent Inc compressible flow we know what numbers mean so here's the chart the 7980x and 7970 X were within 3% in the open foam type performance with the semi 8X there's a large lead and rodinia cfd at 62% ahead the 4900 K outperformed the 7 970x in this category as well and in calculu the 7980x pushed a large lead of 126% in aggregate over the 14900 K and 115% over the 750x we didn't run the 77x on this one because the Embargo is really close now Spec's Life Sciences application benchmarks include Sandia National lab's lamps for molecular property simulation an AMD for molecular Dynamic simulation and rodinia parts for Medical Imaging to track a heart particle potential thermal simulation and speckle reduction here the 7980x is noteworthy in all categories the 79 70x is likewise a high performer and interestingly last gen thread Ripper had a breakout in the ND calculations while the 750x beats the 14900 K in two of three tests the 7980x leads the 770x by about 30% in lamps and a larger 57% in what I'm going to start calling named spc.org provides a financial services Benchmark running Monte Carlo probability simulations for risk assessment black schs pricing models and binomial options pricing models frankly this is an area that we are not familiar with at all so we're just going to give you the numbers the 780x is massively ahead here we're not close to the space but assuming this extrapolates to real world use cases it have a lead of 60% over the 7 97X and 183% over the 7 950x let us know if you're part of this world and what it means to you next is Spec's energy Benchmark in this one the 7980x posts massive scaling in the convolution test which processes a random 100x 100 pixel filter on a 20K X 20K pixel image it also has a strong strong lead in fftw and srmp tests here now for the gaming charts quick reminder this is mostly to make sure games work and that you don't have to do weird things to make them run there's a higher run to run variance than typical with the High Core count threader Parts here this effects results averaging but not in a meaningful way for the gameplay experience ultimately all we're looking for is that they can still run games reasonably you should not expect them to do extremely well in balers Gate 3 the 79 70x ran at 90 FPS average with lows actually completely fine they're right in line with the neighbors so that's good the 90fps average result has it below the 12 900k and tied with a 7700x that's not bad considering historically threader has struggled in gaming scenarios it's far away from the best on the chart where the 7950 xpd has a 39% lead but it's not exhibiting bad frame pain Ing and it's more than playable you'd still at least be able to play balers Gate 3 on your thread River workstation that'll make wend from level one Tex happy the 79 adx ran at 92.2 FPS average so even the 64 core solution did fine at 1440p the scaling is about the same this test is only here to show that increasing the GPU load doesn't always immediately limit the CPU scaling but there are limitations to how well thread Ripper can do our next test will look at Stellaris for CPU simulation time this test measures in time rather than frame rate so it's a tangible change when we see swains of 10% or more in this chart the best results are all stuck around 30 seconds the 7980x required 42 seconds for simulation with the 770x taking 41 seconds the gap between these and the 795x is about 21% with the 14900 K and the 7800 x3d leading thread for 7,000 by upwards of 30% it's not bad performance just like last time but it is noticeable the older thread rer parts are down around 57 to 60 seconds which is significantly slower than Modern Hardware especially with frequent play of Stellaris in Final Fantasy 14 the 77x ran at 172 FPS average with lows spaced about the same as the flanking CPUs the 7980x was a little bit behind this at 159 to 162 FPS average Intel has lawn held the top portion of this chart there's a stark line you could draw between the best AMD and the best Intel CPUs here but on the AMD side the 750x leads the 7 970x by by 9% with its 188 FPS average result the 14900 K leads the 77x by 57% but if you're buying for workstation use first and foremost this at least does well enough in gaming to get by we didn't have to disable cores to play we didn't have to play any weird tricks or enable special features that's the most important part it needs to be easy to switch workloads 1440p was the same The Benchmark is completely scalable at the CPU level so no meaningful change here outside of runto run variants in cyberpunk 2077 with the Phantom Liberty expansion the 7 97X and 7980x were about tied at 134 to 136 FPS average the ceiling is 203 FPS average here set by the 7800 x3d and its x3d compatriots the 7800 x3d is about 50% ahead of the 770x with the 14900 K about 25% ahead frame time pacing again is no problem on the new threader CPUs in F1 2023 the maximum result in our list is the 7800 x3d at a staggering 511 FPS average the other x3d parts are alongside it so this game clearly benefits from cash which makes it interesting because thread CPS have that in abundance but they have other limitations that prevent this from mattering in f123 the 780x does okay it lands alongside the 7600 X and just ahead of the 14600 K but it can't make up for the frequency advantage on CPUs like a 14900 K and up and the 797 X was behind the 780x by about 25 to 30 FPS Landing it closer to the 14600 K and 13 600k so that's interesting just because they diverged a little more here than in the other games frame time pacing was fine the game is playable which is the bar we're setting our next chart to look at clock speeds and all core workloads we're reporting the average of all cores and the maximum single core at any given interval this was when testing blender in an allore and constant load the 780x plots first for this one we measured an average frequency starting at about 4,500 MHz but immediately dropping down with load the CPU bounced between 3.7 GHz to 4.35 GHz throughout the test for the most part the spiky line you're seeing is expected and plotting the maximum helps give perspective on that there's always at least one core boosting to around 47 91 MHz which is typically when the core is at the start or the end of processing its tile as for whether these numbers make sense amd's advertised maximum boost is not during an all core 100% load so you would see that number just before this test started here's the 7 970x plot in this test the all core frequency average was 4600 to 4700 MHz any single core's Peak was again 4791 as with the 79 adx there spikes above 5 GHz prior to the start of the test but the all core workload predictably drops the maximum speed which is expected wrapping all this up quickly to name a couple of the issues we ran into with testing Rainbow Six Siege didn't launch at all so that you still encounter that sometimes with games we encountered issues also with uh higher variant than typically in gaming so it wasn't experience ruining in the games we tested but we did notice a higher sort of run-to-run deviation additionally Premiere failed to render on the 7980x with version 23.6 via the Puget Suite it does open and it's possible it works in a different version of it or outside of The Suite but ours was not able to complete a render that aside the main takeaways are these The Thermals are lower in our test on the 64 core than the 32 core CPU which is interesting in and that's due to the arrangement of ccds and better distribution of heat the 7980x is also crazy energy efficient due to the low spot on the VF curve that AMD chose for it which means there's a lot of room for PBO or overclocking the value is blown way out as you would expect these are diminishing return CPUs these are the what else are you going to buy an equally expensive or maybe more expensive competitor so this is not something you buy for Value it's one of those things where if you actually need the performance hopefully you already know and what matters is that maybe you're making money with your computer having an extra bump even if it costs astronomical amounts more money will pay for itself eventually if you're working with the machine in either case the value is way way better at a $500 mainstream part from Intel or AMD it's just that those don't do what these do the 64 core at its price is effectively unattainable unless it really is a professional Endeavor and even then it's expensive also the 64 cord doesn't always scale well it depends on the application so it's important that you do research on the workload you're running and make sure it will scale that part's really the most important make sure you look at posts online reviews whatever try and find out if it'll scale like you hope it will with your application before you spend this kind of money uh and that's it for now if this review is completely packed we'd prefer to leave you with all the numbers for this one uh and just hopefully that helps you make a decision based on what you're buying or if you're watching this for ENT P now you get to see some crazy numbers from a CPU we wanted to maximize the air time for The Benchmark so check back for the live stream later today if you're watching when this goes up because we're going to be working on the 96 core and setting some World Records that's it for this one thanks for watching subscribe for more and go to store. Gamers access.net to support us directly and get 10% off of our tools and we'll see you all next time

Channel: Gamers Nexus

Views: 276,672

Keywords: gamersnexus, gamers nexus, computer hardware, cpu reviews, amd threadripper 7980x cpu review, amd threadripper 7970x cpu review, amd threadripper 7980x benchmarks, amd ryzen threadripper 7980x vs 7970x, amd 7980x vs 7970x, amd 3970x vs 7970x, asus trx50 sage

Id: yDEUOoWTzGw

Length: 32min 19sec (1939 seconds)

Published: Mon Nov 20 2023