WEBVTT 00:00:00.160 --> 00:00:09.840 the most powerful gaming CPU on the planet is it enough to play modern games 00:00:05.759 --> 00:00:12.719 on a massive 8k OLED tv like this z1088 00:00:09.840 --> 00:00:19.520 from lg and what graphics card could possibly be worthy to find out we have 00:00:15.920 --> 00:00:23.359 the two absolute fastest top-of-the-line 00:00:19.520 --> 00:00:25.359 gpus from both AMD and NVIDIA 00:00:23.359 --> 00:00:30.320 representing team red and weighing in at a cool 00:00:26.439 --> 00:00:34.239 1500 us dollars is the rog strix Radeon 00:00:30.320 --> 00:00:36.960 rx 6900 xt liquid cooled edition and for 00:00:34.239 --> 00:00:43.600 team green we've got the 2 000 00:00:38.559 --> 00:00:46.320 MSI NVIDIA rtx 3090 supreme x 00:00:43.600 --> 00:00:52.239 either one of them alone costs more than an xbox series s series x and 00:00:48.640 --> 00:00:55.199 playstation 5 combined and is capable of 00:00:52.239 --> 00:00:59.840 consuming in excess of 700 watts of power under heavy load at stock speeds 00:00:58.480 --> 00:01:04.640 and today we'll be overclocking both of them to 00:01:02.079 --> 00:01:09.760 their absolute limits fortunately our sponsor seasonic has got 00:01:07.200 --> 00:01:15.119 us covered with their tx 1000 80 plus titanium power supply it's efficient 00:01:12.400 --> 00:01:20.600 quiet fully modular and comes with a 12 year warranty let's make this poor thing 00:01:17.040 --> 00:01:20.600 hurt shall we 00:01:27.119 --> 00:01:33.680 the MSI rtx 3090 supreme x is a powerful 00:01:30.720 --> 00:01:38.720 card to be sure and at least on paper it is well ahead of its competitor the 00:01:35.439 --> 00:01:41.759 Radeon 6900xt thanks in part to its 00:01:38.720 --> 00:01:43.200 super fast GDDR 6x memory 00:01:41.759 --> 00:01:49.280 but here's the thing AMD pulled off a pretty major surprise 00:01:46.240 --> 00:01:51.439 when the rx 6900xt managed to hold its 00:01:49.280 --> 00:01:57.360 own against this monster in traditional rendering at 4k thanks in large part to 00:01:54.640 --> 00:02:01.759 its even faster but much smaller infinity cache also 00:02:00.000 --> 00:02:06.399 the built-in water cooling on this version of the card can't hurt its 00:02:03.680 --> 00:02:10.800 chances actually both of these are among the best available in their class 00:02:08.319 --> 00:02:16.560 featuring overkill cooling a factory overclock triple eight pin power 00:02:13.520 --> 00:02:18.239 connectors oh and of course RGB for that 00:02:16.560 --> 00:02:22.560 extra FPS and the Radeon actually has another 00:02:20.239 --> 00:02:27.599 trick up its sleeve that NVIDIA doesn't have support for yet 00:02:24.879 --> 00:02:32.239 smart access memory also known as resizable bar support allows the CPU in 00:02:30.480 --> 00:02:38.560 our system because we're using a ryzen 5900 xt to access all of the GPU's 00:02:36.080 --> 00:02:43.760 memory at once rather than in the traditional 256 megabyte chunks this 00:02:41.680 --> 00:02:50.080 would ordinarily speed up memory intensive games sometimes by a lot but 00:02:46.959 --> 00:02:51.760 whether that advantage translates to 8k 00:02:50.080 --> 00:02:56.640 remains to be seen just like my sick limited edition foil 00:02:54.160 --> 00:02:59.440 GPU shirt from ltdstore.com like these gpus we can only make a 00:02:58.160 --> 00:03:03.040 limited number of these by the way and we are expecting them to sell out fast 00:03:01.200 --> 00:03:07.920 so go go go anyway meantime 00:03:04.640 --> 00:03:10.959 our GPU bench is ready to freaking go 00:03:07.920 --> 00:03:13.760 here with the aforementioned ryzen 5950x 00:03:10.959 --> 00:03:19.360 16 core CPU a fresh install of Windows and of course our seasonic prime tx 1000 00:03:16.640 --> 00:03:22.080 80 plus titanium power supply again that's to be sure that we're not 00:03:20.480 --> 00:03:26.560 bottlenecking ourselves in any meaningful way we're also going to be 00:03:23.760 --> 00:03:31.519 using these tools from NVIDIA right here to monitor our power consumption in real 00:03:29.440 --> 00:03:35.040 time as we run our tests so that's pretty fun there's no way to please 00:03:33.040 --> 00:03:38.799 everyone these days hey they want you to test in a closed case so that you don't 00:03:36.959 --> 00:03:43.760 give a thermal advantage to some card designs over others but 00:03:41.120 --> 00:03:47.680 the thing about a closed case is that it's like you can't access the i o when 00:03:45.760 --> 00:03:51.519 you have the power measurement stuff hooked up and everything before we 00:03:50.000 --> 00:03:55.280 actually overclock the cards we're doing some baseline testing at their stock 00:03:53.440 --> 00:04:00.959 speeds to see how they compete at 8k 00:03:57.439 --> 00:04:03.280 like AMD technically has the vram for it 00:04:00.959 --> 00:04:07.760 way more than the rtx 3080 but it's slower so like 00:04:06.000 --> 00:04:11.760 that's tough hey yeah it's like almost half the speed yeah it seems like it'd 00:04:09.439 --> 00:04:16.479 be fine for you know working on large projects like as a prosumer card but for 00:04:14.560 --> 00:04:19.040 actual gaming then again we will try to overclock it 00:04:17.840 --> 00:04:26.400 later that's actually not terrible it's like 50 plus FPS and 00:04:23.759 --> 00:04:31.600 remember this is a g-sync capable display so 00:04:29.680 --> 00:04:36.080 it is not currently enabled oh did you not enable it it's best to disable it 00:04:33.600 --> 00:04:40.320 for uh benchmarking that's true still in terms of the gaming experience 55 FPS 00:04:38.560 --> 00:04:45.759 with adaptive sync not bad certainly like on a console you wouldn't 00:04:43.680 --> 00:04:48.800 complain about that nope on a pc i wouldn't complain about that yeah maybe 00:04:47.199 --> 00:04:53.040 not in this game in this game i'd like to have a little bit more oh we're 00:04:50.479 --> 00:04:56.800 definitely getting some judder though you can really see it in those panning 00:04:54.479 --> 00:04:59.840 shots yeah with the g-sync that wouldn't be such a big deal though yeah i want to 00:04:58.160 --> 00:05:05.120 get out into a more open area but i'm getting like 75 80 FPS in here man this 00:05:02.880 --> 00:05:09.360 game is so well optimized it could probably run on like a toaster if you 00:05:07.600 --> 00:05:14.000 really needed to how are we doing for power consumption Anthony looks like 00:05:10.960 --> 00:05:16.800 we've peaked at uh 461 watts and we're 00:05:14.000 --> 00:05:20.639 averaging about 420. nice something to note here though is 00:05:18.639 --> 00:05:24.160 that while the game is never dropping below 60 FPS 00:05:22.479 --> 00:05:29.039 Anthony do you see that judder and those micro stutters oh yeah yeah it's like 00:05:26.560 --> 00:05:31.039 hitching every i don't know like every time 00:05:29.840 --> 00:05:34.800 maybe yeah yeah so 00:05:32.960 --> 00:05:38.800 hopefully overclocking will help us a little bit with that let's see 00:05:36.479 --> 00:05:43.360 oh it's hot oh it's hot yeah yeah yeah yeah always yeah that 00:05:40.639 --> 00:05:48.479 backplate is used actually for cooling yeah no kidding jeez 00:05:46.240 --> 00:05:52.960 well it's handling it actually looks like NVIDIA might not have the only 8k 00:05:51.039 --> 00:05:57.520 gaming card on the market what does the frame rate display show uh 00:05:55.039 --> 00:06:00.400 it shows 55. oh wow that's okay 00:05:58.320 --> 00:06:04.880 very similar and you know what i need to get out into a more open area first but 00:06:02.479 --> 00:06:09.360 it actually doesn't feel quite as juddery how's our power consumption 00:06:07.199 --> 00:06:15.919 power consumption has peaked so far at 384 and we're in around 311 average 00:06:13.840 --> 00:06:20.000 and to be clear that's just the GPU the numbers we're talking about there not 00:06:17.759 --> 00:06:22.400 the entire system so yeah like 100 watts less for 00:06:21.360 --> 00:06:26.080 similar questions performance 00:06:24.400 --> 00:06:31.520 it's a lower FPS reported by the game 00:06:28.639 --> 00:06:37.520 but it's actually a little bit smoother feeling in terms of the frame delivery 00:06:34.400 --> 00:06:40.160 which is not what i actually expected 00:06:37.520 --> 00:06:42.960 uh freesync is anon right i could check that 00:06:41.039 --> 00:06:46.720 ah i don't know i see tearing i don't think so 00:06:44.479 --> 00:06:50.160 okay so back to the same thing with the pillars look how much smoother that is i 00:06:48.800 --> 00:06:53.120 think we're gonna need to run some more testing and get the peeps some some real 00:06:51.919 --> 00:06:58.960 numbers here the rest of our games went about as you'd expect with AMD trailing NVIDIA 00:06:56.720 --> 00:07:02.960 significantly at about three quarters of the performance to shadow the tomb 00:07:00.240 --> 00:07:07.440 raider and it's even worse with ray tracing enabled and that doesn't even 00:07:04.880 --> 00:07:10.960 consider dlss NVIDIA's deep learning super sampling that runs the game at a 00:07:09.440 --> 00:07:15.840 lower resolution then uses machine learning to upscale it 00:07:13.919 --> 00:07:20.160 it's a great feature and in supported games honestly NVIDIA is just 00:07:18.880 --> 00:07:24.000 untouchable red dead redemption 2 has a similar 00:07:22.000 --> 00:07:28.720 story in minimum frame rates but the average closes the gap somewhat while 00:07:26.160 --> 00:07:32.400 forza horizon brings us much closer at roughly a 10 performance difference 00:07:31.360 --> 00:07:39.199 finally ha would you look at that 00:07:35.039 --> 00:07:40.720 cs go has AMD winning at stock speeds 00:07:39.199 --> 00:07:44.960 very interesting this simpler title lends credence to the 00:07:42.639 --> 00:07:50.639 notion that AMD's raw raster performance may be higher than NVIDIA's regardless 00:07:47.599 --> 00:07:53.280 of memory throughput and f1 2020 gives 00:07:50.639 --> 00:07:56.000 us another win for team red dang i was worried this wasn't going to 00:07:54.639 --> 00:07:59.919 be much of a like ultimate showdown but 00:07:58.479 --> 00:08:05.919 turns out we've got a battle on our heads now let's get to overclocking NVIDIA's 00:08:04.080 --> 00:08:11.039 newer versions of GeForce experience provide a way to automatically do 00:08:08.080 --> 00:08:16.160 overclock scanning and set power targets unfortunately adjusting the voltage or 00:08:13.680 --> 00:08:22.000 power target made the oc scan crash the system and MSI afterburner similarly 00:08:19.280 --> 00:08:25.840 either crashed or settled on an unstable overclock whenever performing an oc scan 00:08:24.319 --> 00:08:29.360 for a while there we thought our card might already just be at peak 00:08:27.599 --> 00:08:35.039 performance but manually overclocking we were able to 00:08:31.280 --> 00:08:36.959 bump the core by about 100 megahertz and 00:08:35.039 --> 00:08:42.080 800 megahertz on the memory before we started losing stability and performance 00:08:39.599 --> 00:08:46.640 to down clocking and error correction in order to do that though of course we 00:08:43.519 --> 00:08:46.640 have to raise our power limit 00:08:47.360 --> 00:08:53.839 and apply as for your core voltage honestly 00:08:51.040 --> 00:08:58.560 increasing it might work against you as much as for you because you'll be more 00:08:55.680 --> 00:09:02.240 likely to hit your power limit so Anthony says he actually got about the 00:09:00.160 --> 00:09:06.000 same results with it cranked as with it just to at nothing so 00:09:04.800 --> 00:09:10.160 whatever you know what you only live once all right crank it yolo yolo 00:09:08.800 --> 00:09:15.440 something to watch out for here is that those gddr6x memory chips get extremely 00:09:13.040 --> 00:09:20.560 hot even under normal circumstances and they are going to get even hotter now so 00:09:17.920 --> 00:09:24.560 the back plate of this card is necessary for keeping them cool so far this 00:09:22.640 --> 00:09:30.320 actually doesn't really seem any better we're getting about 00:09:26.560 --> 00:09:32.320 70 65 to 75 FPS which is 00:09:30.320 --> 00:09:35.519 pretty much what we saw before i'm also feeling a bit more stuttering 00:09:33.839 --> 00:09:40.000 than i did on AMD that's something that based on the measurements we took ahead of time we 00:09:38.320 --> 00:09:43.680 were not expecting yeah there's definitely a judder yeah 00:09:42.560 --> 00:09:49.440 see that did it look worse than the Radeon card to you i feel like the Radeon 00:09:47.680 --> 00:09:52.880 didn't have as severe stuttering 00:09:51.200 --> 00:09:58.240 at least based on what i'm seeing here and what's in my memory uh maybe i'm 00:09:55.440 --> 00:10:02.160 just crazy but it looks worse here that does not look like 70 frames per 00:10:00.880 --> 00:10:07.200 second period yeah it looks kind of like 45 or 00:10:04.240 --> 00:10:10.240 50. yeah if that and not a smooth 50. yeah it almost reminds me like the 00:10:08.480 --> 00:10:15.040 microstarter you'd get from multi GPU yeah frame rate high consistency low how are 00:10:13.600 --> 00:10:19.279 we doing for power 500 peak 00:10:18.079 --> 00:10:26.959 and the average looks like it's floating in around 00:10:22.200 --> 00:10:28.800 440 450 so we're up a lot higher yep 00:10:26.959 --> 00:10:34.320 that's higher all right half our power supply dedicated just to 00:10:30.800 --> 00:10:36.079 the GPU and this tool does not measure 00:10:34.320 --> 00:10:39.200 fine enough time increments to see these like micro spikes that can take place 00:10:37.760 --> 00:10:43.120 from time to time now we'll throw the AMD card back in and i'll try not to 00:10:41.279 --> 00:10:46.720 punk myself here and touch the back plate 00:10:44.959 --> 00:10:51.120 we expect it to be even hotter than last time AMD's Radeon software offers 00:10:48.959 --> 00:10:57.760 overclocking controls directly setting the card to 2600 megahertz seems to be 00:10:54.160 --> 00:11:00.160 about as high as we can go out of the 00:10:57.760 --> 00:11:04.079 gate any higher and it'll throttle to this speed anyway 00:11:01.760 --> 00:11:08.800 just add full voltage and a power target and then the memory we were actually 00:11:05.600 --> 00:11:10.959 able to set as high as possible and it's 00:11:08.800 --> 00:11:15.040 possible we could have even gone farther if we had a bit more control over memory 00:11:12.959 --> 00:11:18.880 frequency okay so let's try cranking the power limit for one thing 00:11:16.880 --> 00:11:23.200 uh main frequency you can leave max frequency 2600 00:11:21.440 --> 00:11:27.120 and RAM just you can go advanced control just to see 00:11:24.720 --> 00:11:31.600 how much that is uh yeah 2150. okay 00:11:29.440 --> 00:11:36.079 apply changes okay it was not my imagination last time this is noticeably 00:11:34.320 --> 00:11:40.079 smoother than the NVIDIA experience although if the numbers are anything to 00:11:37.920 --> 00:11:44.160 go by it's a bit of an outlier what's also a better experience compared 00:11:42.399 --> 00:11:48.240 to NVIDIA though is the overclocking headroom on this card 00:11:45.680 --> 00:11:53.519 at the low end we still only got about uh five percent improvement but on the 00:11:50.320 --> 00:11:56.560 high end we were looking at up to 15 16 00:11:53.519 --> 00:12:00.160 faster performance so that's 00:11:56.560 --> 00:12:02.720 pretty cool this might be my imagination 00:12:00.160 --> 00:12:07.200 but it seems like it is far smoother than before um i think it is smoother i 00:12:05.680 --> 00:12:14.320 don't know if i would describe it as far smoother it's definitely smoother though 00:12:10.880 --> 00:12:16.240 we're getting 60 to 70 mid FPS we were 00:12:14.320 --> 00:12:19.680 getting like 50 to 60 before weren't we actually that's true isn't it 00:12:18.240 --> 00:12:24.480 okay yeah and it is noticeably smoother to operate 00:12:22.240 --> 00:12:29.680 i'm surprised AMD comes out with a pretty clear win at least in do maternal 00:12:27.360 --> 00:12:32.959 although you know what Anthony i am seeing more stuttering now that we've 00:12:31.600 --> 00:12:36.720 overclocked it or judder rather 00:12:34.880 --> 00:12:40.079 see that how it'll go smoothly for a little bit and then it'll like 00:12:38.480 --> 00:12:44.800 like the NVIDIA card how's our power consumption looks like 00:12:42.320 --> 00:12:51.680 415 watts maximum uh we're averaging around 360 362 watts 00:12:49.920 --> 00:12:56.240 as far as you know not too shabby overall neither card had 00:12:54.720 --> 00:13:00.000 terribly impressive overclocking headroom but that's not a huge surprise 00:12:58.240 --> 00:13:04.800 considering that both of them already had factory overclocks applied 00:13:02.399 --> 00:13:08.880 for our results we did eek out a frame or two here and there on both cards in 00:13:07.200 --> 00:13:12.160 shadow of the tomb raider but ray tracing remains a pain point for AMD 00:13:10.720 --> 00:13:17.440 with minimums that are still in the single digits red dead redemption 2 and 00:13:14.720 --> 00:13:22.240 forza horizon meanwhile put AMD a couple of percentage points closer to NVIDIA 00:13:19.120 --> 00:13:25.519 than stock which in forza puts the 6900 00:13:22.240 --> 00:13:28.399 xt within the 10 mark and as we might 00:13:25.519 --> 00:13:33.360 expect cs go improved by a significant margin maintaining AMD spot well ahead 00:13:30.959 --> 00:13:36.560 of NVIDIA but this time in every measurable way 00:13:34.720 --> 00:13:41.120 the tldr then is this AMD's best puts up a valiant effort for 00:13:39.120 --> 00:13:46.000 the 500 price difference against the 3090 sometimes coming close or even 00:13:44.000 --> 00:13:50.959 beating it at 8k but only sometimes it still falls short 00:13:48.800 --> 00:13:54.320 in more demanding titles even with its greater overclocking headroom which only 00:13:52.880 --> 00:13:59.199 serves to further highlight the importance of technologies like dlss and 00:13:56.959 --> 00:14:03.680 how much not having it is holding AMD back right now and not just in ray 00:14:01.279 --> 00:14:08.560 tracing either outside of doom eternal which surprisingly was quite a bit 00:14:05.519 --> 00:14:10.800 smoother on AMD gddr6 seems to be 00:14:08.560 --> 00:14:17.519 helping NVIDIA out tremendously because remember ak is four times the number of 00:14:14.160 --> 00:14:19.360 pixels as 4k just the frame buffer alone 00:14:17.519 --> 00:14:25.040 that is each frame displayed on the screen is nearly 00:14:21.079 --> 00:14:28.320 128 megabytes which is exactly enough to 00:14:25.040 --> 00:14:30.560 hobble the 6000 series 128 megabytes of 00:14:28.320 --> 00:14:35.600 ridiculously fast infinity cache leaving them with memory that's just over half 00:14:32.880 --> 00:14:40.720 as fast as ampere's for everything else with that said if AMD had a dlss analog 00:14:38.399 --> 00:14:44.800 to reduce this memory footprint AMD might have been more competitive more 00:14:42.480 --> 00:14:49.360 often so it seems reasonable to suggest that if AMD had equipped the 6900xt with 00:14:47.800 --> 00:14:53.680 gddr6x things might look quite a bit different 00:14:51.920 --> 00:14:57.600 which makes me wonder how much of a bottleneck it would have been for NVIDIA 00:14:55.440 --> 00:15:00.800 if they had stuck with pci express gen 3 this time around 00:14:58.959 --> 00:15:03.040 get subscribed for our video because we are going to be exploring that by the 00:15:02.079 --> 00:15:08.399 way so then is 8k gaming a lie 00:15:06.480 --> 00:15:13.920 well it's complicated while i participated in some of NVIDIA's 00:15:10.560 --> 00:15:15.199 early promotion for 8k gaming on the rtx 00:15:13.920 --> 00:15:21.040 3090 and it does work in the same way that a 00:15:18.000 --> 00:15:22.880 primitive hammer can pound in a nail 00:15:21.040 --> 00:15:27.519 it's clear that it's not practical for every game today and as more visually 00:15:26.000 --> 00:15:32.560 elaborate titles trickle out over the next few years new hammers will be much 00:15:30.560 --> 00:15:36.720 more appropriate tools to handle them at this resolution but this was still an 00:15:34.959 --> 00:15:42.160 interesting exercise for a couple of reasons first because of how demanding 00:15:39.519 --> 00:15:45.680 it is gaming at 8k can give us a glimpse into how a card's performance might 00:15:43.920 --> 00:15:50.639 evolve over time AMD famously squeezed several years of 00:15:48.399 --> 00:15:55.199 life and performance improvements out of its Radeon 7970 00:15:52.880 --> 00:15:59.279 and it looks like compared to NVIDIA they may have a bit more headroom here 00:15:57.040 --> 00:16:02.480 again oops AMD cards over there and second while you might roll your 00:16:01.360 --> 00:16:08.240 eyes and say who cares nobody has an 8k monitor 00:16:05.360 --> 00:16:12.240 the same was true for 4k a short while ago and even if you only use a 4k 00:16:11.279 --> 00:16:16.800 display super sampling or running at a higher 00:16:14.399 --> 00:16:20.720 resolution than native then down scaling it is still the best way to reduce 00:16:18.800 --> 00:16:26.959 shimmering and other artifacts while preserving 100 of the detail in an image 00:16:23.680 --> 00:16:28.880 and 4x super sampling at 4k means 00:16:26.959 --> 00:16:33.600 rendering at 8k fun fact to close this out dlss NVIDIA's 00:16:31.920 --> 00:16:37.199 killer app that allowed them to market this card as an 8k gaming solution with 00:16:36.000 --> 00:16:41.759 a straight face is currently used to render at a lower 00:16:39.440 --> 00:16:46.959 resolution and then blow it up to your display's native res but that wasn't how 00:16:44.480 --> 00:16:51.680 it was first introduced it was a form of anti-aliasing hence the super sampling 00:16:49.519 --> 00:16:54.480 in the name and it's very good at this task 00:16:52.639 --> 00:16:59.199 so good in fact that its second iteration it now legitimately enables 00:16:57.040 --> 00:17:03.279 high impact features like ray tracing to run at a lower resolution while 00:17:00.800 --> 00:17:07.760 preserving most of the sharpness and detail of native rendering this point is 00:17:05.600 --> 00:17:12.000 especially painful for AMD whose ray accelerators already significantly lag 00:17:09.919 --> 00:17:17.760 behind NVIDIA's rt cores anyway the main takeaway is that neither 00:17:14.400 --> 00:17:21.120 AMD nor NVIDIA is even close to creating 00:17:17.760 --> 00:17:22.240 a GPU that can natively run 8k 00:17:21.120 --> 00:17:26.640 maxed out aaa games and that upscaling tricks are 00:17:25.199 --> 00:17:31.360 here to stay and by the way one more takeaway is that 00:17:29.200 --> 00:17:37.760 unless your screen is this big 8k has a much smaller impact on your 00:17:33.840 --> 00:17:39.200 gaming experience than say HDR or 00:17:37.760 --> 00:17:42.559 running at a higher refresh rate for smoother animations and that last one 00:17:41.440 --> 00:17:49.039 you can actually check out our whole video about that here on the subject of smoothless smoothness 00:17:47.280 --> 00:17:53.919 big shout out to sonic for keeping our test benches running smoothly as can be 00:17:51.919 --> 00:17:57.679 they have observed these gpus both of them actually spiking north of a 00:17:55.600 --> 00:18:02.559 thousand watts for very very short durations which can wreak havoc on over 00:18:00.160 --> 00:18:05.679 current protections and our tx1000 handled them like a champ you can check 00:18:04.559 --> 00:18:12.080 it out at the link in the video description and of course they've got lots of other power supplies to fit any 00:18:09.600 --> 00:18:12.080 budget