Gadgetory

Marvin it's a large part of the reason that technical media should exist and part of our job is to filter out things like the adjectives and the marketing materials and filter out all the obfuscation and get to the objective truth of the matter whatever that may be Intel's initial marketing that contained the slide which indicated that the x-series CPUs would be better for three-way and four-way SLI or crossfire configurations for quote 12 K gaming ignoring the fact that 12 K gaming isn't really a thing and no one does it or at least very few people and other things about 12 K double cat to momentarily the orange bars on the slide that demarcate that the case q CPUs like the 7700 K are not capable of this same task of three or four-way SLI or multi GPU or more specifically and more importantly for this video the 7700 K and other case cues are not capable of 12 K gaming where the 7900 X might be today we're going to be looking at that and using that as a means or as a reason that we used to test PCIe Lane scaling across two cards in SLI talk about that in a moment and see how the impact looks for gaming performance before getting to that this coverage is brought to you by our patreon backers at patreon.com slash gamers Nexus you can join there to help us with this in-depth testing and these extra feature benchmarks which are made much easier by having the support of our patreon backers or if you'd rather you can go to store dog and Razak to start net and pick up a shirt we'll get to the testing in a moment but let's start with the 12 K gaming part of the slide this is the part that needs to point it out immediately first of all to define a few things 12 K gaming as Intel defines it on that slide is probably not what most people would expect when they hear something like 12k when you hear 8k we know that it's scaling 16 by 9 12 K and Intel's example is actually a bit different so let's break that down most commonly 4k means 3840 by 2160 which equals 8.3 million pixel then there's a 5k which is 50 120 by 2880 resolution and 14 point 7 million pixels or 14 point 7 megapixels we next go up to 8k resolution seventy six eighty by forty three twenty or thirty three point two million pixels thirty three megapixels so what is twelve K well based on this scaling we would expect twelve K to be something like eleven thousand five hundred and twenty pixels by something like sixty four eighty pixels using a 16 by 9 aspect ratio that puts us at seventy five million pixels or seventy five mega pixels that's more than two times the pixel count of 8k and that's a big deal if you could play twelve K games it would be pretty remarkable technically speaking twelve K is not officially defined I'm not presently aware of any displays in the consumer market which are twelve K ready but either way it's not what it sounds like when we look at Intel's version Intel doesn't mean seventy five million pixels Intel means twenty five million pixels at a resolution of eleven thousand five hundred twenty pixels of by 2160 or three at 4k displays next to each other what we would probably call triple Forte we emailed the companies to confirm this and that's what they said they said it is three 4k displays next to each other which they are calling or abbreviating twelve K that's a three times difference in pixel count from what the consumers expectation might be of twelve K and what the actual meaning is behind Intel's marketing but even then that's not really the part that matters regardless of some of Intel's embarrassing marketing for this launch like extreme mega tasking the part that matters is that they're claiming a higher resolution performs better on other CPUs even with PCIe Lane scaling this is still largely a marketing stunt here's why this is a GPU benchmark now just like when AMD claimed great 4k performance on Rison despite the fact that the GPU actually basically does all the work at that point and that it's completely irrelevant as a rallying point for CPU Intel is now piling on and claiming that triple 4k performance is because of the x-series CPUs that just came out in reality the only reason this would ever matter is because of the lane count difference which of course has been a major topic of discussion lately so then this is basically becoming a PCIe scalability test between multiple GPU configurations in this case 2-way SLI with 1080p is and we're going with two-way SLI because although Intel's slide and technical marketing team indicate a threeway or four-way SLI compatibility for multi-gpu the fact of the matter is that nvidia does not support 3-way or four-way SLI for gaming it's physically impossible to do it so with Pascal that's the that's the reality the only way you could do 3-way or four-way with a relevant card at this point would be to do something with either AMD or with the Maxwell series in which case even three 980ti s would not equate the power of two 1080 Ti so you really might as well just do this and as for an these cards you've basically got rx5 ATS right now Vega is not out and so from that perspective you'd have to pile on three or four or five ATS to come close to doing what two of these could do that's not the down talk to 580 but this really isn't what it's meant to compete with so what you end up with is something where there's a very specific use case where this test can make sense and then the rest of them that would actually draw all those PCIe lanes Intel's talking about just don't work anymore doesn't work with Pascal and there's no point in doing it with the rx5 ATS maybe with Vega but that doesn't exist today on the market so we can't quite test that just yet Intel does have a point that the lane count of some of the x-series CPUs is significantly higher than the lane count of the previous case cue CPUs but they also indicate on the slide that the 6 to 18 core parts are the only one that the statement applies to which means that the KB Lake X parts need not apply and that said the six core eight core parts as Paul has been saying quite frequently on his channel actually have twenty eight lanes sure that's still better than 16 on the 7700 K but that still doesn't get you to 2 by 16 cards anyway we emailed Intel about the fact that three-way and four-way SLI and Pascal is not possible for gaming and as if the company would like to respond for this video Intel's official statement to gamers Nexus is the following the 12 K gaming mentioned in the table from our product briefing info for press included an error referring to an older PC gaming GPU configuration the x-series platform enables true PCIe by 16 + by 16 configure nations whether or not it performs better is dependent on the software and the graphics vendor so having gotten through the marketing screw-up and Intel's statement to us what we're left with is actually a really cool testing opportunity for PCIe Lane scaling on two different CPUs we did this before with the same cpu where we used motherboard BIOS to basically toggle by eight verses by sixteen land but now we're testing it with the 7900 x-rays of the 7700 k specifically to verify intel's marketing claims and then hopefully as collateral to that you'll get a benefit in figuring out if doing something like 210 ATT is is actually worth it on either one of these platforms with regard to lane availability so with the 7900 a configuration we're doing a few things one we're running at stock in some of the tests and we're also doing a configuration where we've overclocked it to equate the 7700 k a stock cpu clock speed to 4.5 gigahertz and we've also disabled 12 of the threads leaving it with 8 that puts that basically a 7700 k so for those tests we're looking at more of a direct comparison core to core clock the clock of PCIe Lane bandwidth differences between the two platforms where one can peel off 32 lanes to the two GPUs and the other one is peeling off 16 in 8 and 8 or in the former 16 and 16 anyway let's get to the testing remember the scope is that Intel is saying a 7700 K isn't rated for triple 4k or what they erroneously called 12 K and it's saying that the x-series is better for it the implication is a better PCIe Lane performance as marked by the orange bars on their slide last time we did a PCIe SLI scaling test we saw less than 1% difference this time we're being a lot more judicious with a minimum run count of 5 test passes per test collecting a standard deviation as well and having spoken with Nvidia and sis about test scenarios that would most likely show a PCIe scaling difference so that we can really put this thing to the test rather than testing 30 some-odd games and really finding out that doesn't do anything it turns out even in the best possible light none of it matters and well there's one exception but we'll get to that now one problem here we don't have three 4k monitors to test and if we did we wouldn't have space with it but there's an important thing here with this type of testing all that really matters is the pixel count so three 4k monitors would give you a pixel count of twenty four million eight hundred eighty three thousand two hundred pixels and using our rarely used resolution that we found we are on 66 51 by 3741 so that gives us 24 million eight hundred eighty one thousand three hundred ninety one pixels you subtract that from the first and end up with an 1809 pixel difference really not bad that is so insignificant that we can basically call these resolutions the same and carry on with our testing and one more note here we're using DSR to achieve these resolutions in the past we've tested DSR scaling and its impact on performance or the performance hit to do the scaling and in that testing along with more recent testing for this we see a scaling of about one to three percent so there's really not a big difference in a way that would make this different in a meaningful fashion from if you actually had a monitor at this resolution or three 4k monitors at this pixel count so that's what we're working with we've got a 1 to 3% tag from DSR but it applies equally to all of the tests so basically doesn't exist for the purpose of this testing and other than that we're working with 66 51 by 37 41 now for the final testing information a couple more important things one multi-gpu is really hard to find modern games that actually scale well with i we've tried it in recent history and the options are basically sniper elite and maybe one or two other games really not that many sniper lis is kind of the seminal example so that's the first thing just because there might be scaling in the games we're using here doesn't mean you should expect multi-gpu to actually be worth it in anything you do IRL outside of a testing environment the next thing these things that in the past have tended to stress PCI interface limitations the most have been synthetic test that's fire strike with custom tune settings to really blast the GPUs now times by now superposition and then we can also look at something like ashes of the singularity which is for all intents and purposes synthetic and we've also got a couple of games in here as well that have been proven to generally show good SLI scaling or PCIe Lane testing in the past one of them Crysis 3 it's a bit old but it works really well for this purpose and then the other Sniper Elite 4 which has nearly 2 X K lean on multi-gpu that is the best we can get for getting both of these things to pump out data to their maximum potential basically I've got a couple other games in there as well but for full testing methodology check the link in the description below and then for tests where we've got the 7900 X with cores and threads disables going to clock boosted that's basically to eliminate the final variable which is this thing has more threads and a different clock so that's why that's done with fire strike configured it to fire strike ultra with a custom 6651 by 3741 resolution which we are going to call 25 mega pixel resolution from here on out because it's easier we're seeing the I 970 900 X CPU and dual 1080 TI f tws operate with graphics cores of 27.4 and 15 point seven four one and two compared to these 7700 K scores of 26.9 six and 15.3 for this data was collected over five test passes each producing a margin of variance of plus or minus 0.1 FPS not percent just 0.1 FPS resolution on these tests that's pretty accurate so we're looking at a performance difference in the range of 1 to 2% max we could somewhat confidently state that the extra eight lanes account for this but frankly the 8 thread 4.5 gigahertz version of the 7900 X helps us find out that that's not necessarily the case this runs framerate closer to the 7700 K and averages out to 26.5 3 FPS for graphics 1 1501 for graphics 2 once we've controlled 2 threads and speeds even that 1 to 2% window pretty much goes away is now within test to test variants the graphics core differences come down to 45 97 verses 44 97 if you care 100 point difference which is again accounted for by less than 1 FPS / graphics 1 and less than 0.5 FPS for graphics - but what if we made the test easier let's go down to firestrike ultra at 4k just because now by eliminating an awful load from the GPU to allow the CPU to start showing a difference we see scaling up a couple frames the difference is about three FPS on graphics one was the 7900 X on the lead and about three to four FPS on graphics - but again at this point we can no longer confidently state a difference caused by PCIe lanes disabling threads and normalizing for speed helps with that a bit we see the eight thread 700 X running a 75048 fps graphics one score and 48.1 fps graphics to score that brings us to about a two percent difference in graphics one at this point it looks like that could be an advantage of the extra lanes we're not fully confident because it is pretty close but it does look that way in this particular test time spy doesn't have a twenty five megapixel configuration that we could run so we ran to that 4k this one posts an FPS difference of about 2.3 FPS for gravity's won the sony 900 x leading and about 1.6 ft s for graphics 2 respectively the percent scaling yields an increase in performance of 2.9 percent and 2.8 percent for the 7900 x this difference again starts to hinge more on the cpus at 4k than on the lanes so let's increase the resolution and load on the GPUs again just to really stress the PCIe interface for using Unigine super position now and it doesn't have a 25 mega pixel setting either but it does have 8 K which is a higher pixel count of 33 megapixels versus 25 of triple 4k we're seeing an average FPS across 5 runs of 30.8 9 versus 39 5 technically favoring the 7700 K this difference is within our test variance with our margin of error calculating out to plus or minus 0.25 FPS again basically equal even the minimums output by the software are close and we don't normally like minimums but the maximums are also nearly equal and the score if you're curious comes out to 41 29.5 on the center 900 X vs. 4138 I'm a 7700 K within variance of test the test passes they are the same the PCIe Lane advantage isn't helping and the X series isn't magically better at this non tenable 8k resolution let's get back to that 25 mega pixel number again by using games with Crysis 3 which is old but still one of the best graphics benchmarking tools available for PCIe Lane scalability we're seeing a difference of about 1 FPS average our test variance here is 0.76 FPS which is most of one FPS so that means our CPUs are again effectively equal we cannot confidently declare that one is better than the other which means that we cannot confidently declare if there's a difference because of the PCIe lane and of course because everyone just looks at the charts to claim a victory going to say the 7500 K is faster that might be true if you were to analyze it really heavily and do a whole bunch more test passes but as it stands now we must abide by the confines of our numbers and those say that the two CPUs are equal in their respective configurations with ashes of the singularity at 25 mega pixels and extreme settings using 8 tap msaa and DirectX 12 we see that the game stresses the GPS to a point of running at around 40 fps honestly that's still fairly impressive performance overall given the resolution but then again it's a synthetic benchmark both regards to scaling performance it's not a huge difference that we are seeing more of one than we saw in the other titles this is a DirectX 12 bench so that could account for some of the differences the 7700 K runs at 39.9 fps averaged 13.5 1% blows 11.9 FPS 0 or some blows compared to the 4 core 8 thread 7900 X variant which starts at 42 FPS average we're looking at a performance increase with the extra lanes of about 5 to 6 percent the 7900 X without it's overclock does better in frame time variants consistent with our upcoming Sniper Elite results next sniper 4 is another special kind of game we had ashes of the singularity which has done built very heavily for benchmarking purposes and so means that it is optimized for Hardware Sniper Elite 4 is also optimized for hardware in our recent multi-gpu benchmarks we showed about a 100% scaling with crossfire 5 ATS and Sniper Elite only very few modern games managed this feat with this game we're seen scaling up about 100 percent even with DX 11 which was used because Nvidia's cards seemed to put more load on the PCIe bus when using the X 11 even with all of this going for it with all that utilization of both GPUs we're seeing Sniper Elite for roughly tied between tests the 7700 K runs at about 54 FPS average of 49 and 42 fps those the average was 54 FPS every single pass so that's consistent and the lows have a plus or minus error of about 1.7 FPS the 7900 x8 thread CPU ran about 53 to 54 FPS average with a range plus or minus 0.3 FPS and lows at 42 and 35 with the range of 0.5 and 1.7 respectively some of this difference can be chalked up to our increased voltage and the overclock with regard to the low performance the frame times which sometimes show a PHA and time variance change because you're overclocking and over voltage so things can happen the non overclocked 7900 X was more stable though at 55 FPS 48 fps and 42 FPS respectively averaged at 1% in point one percent finally total war Warhammer helps us get a wider spread of game variety but ultimately also failed to produce any meaningful differences the 7700 K runs an average FPS of around 36 here with lows at 21 and 20 the 7900 X manages 37 FPS average but it has too many other variables at play so we can lock it down increasing the clock speed in to say when 12 threads makes for a better test putting us at about 38 FPS average the clock speed matters more than cores in this game but it's clear that the CPU is still at play here we cannot confidently state an advantage from PCIe lanes alone in this test if there is one it's in the range of 3 to 5% or there abouts and would vanish the moments you decrease load on the GPUs and increase load on the CPUs so looking at all these there's really mostly no difference in terms of performance from PCIe lanes and scaling them but we did see 5 to 6% scaling and ashes of the singularity with DirectX 12 and 3 to 5 ish 5.5 with total war Warhammer there's no real scaling to speak of in Sniper none really in crisis and the synthetic spin seems much care either even if you're going to play triple 4k or do some kind of very high resolution monitor 8k maybe at the end of the day you're not going to torture it the way we are we're putting these things through ultra settings 8 1/2 MSAA which just seems wholly unnecessary for this kind of resolution and trying to put as much load on the PCIe bus as possible it is our intention to stress the PCIe bus what happens though if you back off of that well if you scale back and you lower the load on the GPU is you end up coming back to the CPU and loading it so at that point we're looking at something that would level out and completely nullify for the most part any actual PCIe lane differences that you might encounter regarding PCIe lanes for GPUs the concern seems a little blown out of proportion however there is still a concern for PCIe lanes in general on these CPUs because one it be a hell of a lot easier to understand what's going on what's supported between platforms and CPUs if they were either all the same or just had more lanes and two you still need to be cautious about what you're buying with regard to high-speed i/o devices how many you're using and whether or not you care about going through the DMI versus just talking to the cpu directly so all of that matters as well but it looks like with our testing the GPU performance isn't quite as big of a deal as some of the other stuff is with this platform and there are plenty of other problems the biggest thing here is that Intel is damaging its brand credibility by making claims like 12k gaming pointing at three and four Way SLI as a reason which is not really possible again and we end up looking at the same performance as the Saudis have 100k for the most part ultimately even with all of those caveats none of this matters because multi-gpu scaling is so rare anyway we I don't think have ever recommended it in all the years of working with SLI and crossfire I don't think either one has gotten her recommendation there are specific instances where you would do multi-gpu maybe like production where you're just going to run blender and task you're rendering to each card in which case it doesn't care about SLI or maybe you have a very specific use case for gaming and you've already purchased the best card in terms of raw performance on the market whether that's a Titan X or a 1080i or maybe in the future some Vega variant if you buy one of those maybe multi-gpu make sense because you have nowhere else to go above you but for the most part this is something that is a really limited use case the scaling does not look like it matters in a lot of instances but quoting our Nvidia industry contacts you'd really have to test 30-plus titles to get a full spread and truly understand what it looks like across everything on the market but for games that do actually scale with SLI and scale well its hit and miss with mostly miss especially on the synthetics which often do stress things the most for PCIe so that's all for this one if you'd like to help us with this type of testing is always good at patreon.com slash gamers Nexus or you can subscribe for more information thank you for watching I'll see you all next time you