WEBVTT 00:00:00.380 --> 00:00:06.080 AMD is back and NVIDIA is about to pay 00:00:04.080 --> 00:00:11.880 for their outrageous pricing practices the rdna3 announcement just concluded 00:00:09.480 --> 00:00:15.960 and while AMD gave us very little in the way of meaningful 00:00:13.679 --> 00:00:21.359 performance data I'm really optimistic anyway and here's why immediately after 00:00:19.020 --> 00:00:26.760 the stream ended our Labs team sprang into action building up test benches as 00:00:24.300 --> 00:00:32.579 close as possible to the ones that AMD used in their performance preview that 00:00:28.980 --> 00:00:35.160 allowed us to expand on AMD's rather 00:00:32.579 --> 00:00:41.100 vague slides that didn't include a single NVIDIA GPU and extrapolate some 00:00:38.100 --> 00:00:43.980 actually useful comparative performance 00:00:41.100 --> 00:00:48.360 data now the bad news is that NVIDIA might still be the performance King but 00:00:46.200 --> 00:00:54.660 the good news is that I can confidently say that AMD's new Radeon RX 7900 XTX 00:00:52.079 --> 00:01:01.980 should come within spitting distance of NVIDIA's RTX 4090 for 600 dollars less 00:01:00.360 --> 00:01:08.100 the only thing with better performance per dollar is this segue to our sponsor 00:01:05.000 --> 00:01:10.860 PDQ PDQ makes device management simple 00:01:08.100 --> 00:01:14.880 secure and pretty damn quick automate your patch management and software 00:01:12.299 --> 00:01:20.760 deployment processes with PDQ deploy and inventory start your free trial now at 00:01:16.979 --> 00:01:20.760 pdq.com LTT 00:01:28.280 --> 00:01:33.540 in AMD's together we advance underscore 00:01:31.560 --> 00:01:40.860 gaming live event they announced the two Flagship tier gpus that will lead their 00:01:35.939 --> 00:01:45.600 rdna 3 family the Radeon RX 7900 XT and 00:01:40.860 --> 00:01:48.659 7900 XTX with 20 and 24 gigabytes of 00:01:45.600 --> 00:01:51.000 gddr6 respectively which was my first 00:01:48.659 --> 00:01:55.520 surprise I mean why would they cheap out on the memory when they need to compete 00:01:52.979 --> 00:02:00.600 with NVIDIA cards that are using faster gddr6x two reasons one to save power 00:01:59.520 --> 00:02:04.500 which fair enough given the reputation their 00:02:02.520 --> 00:02:09.899 competition is built for unreasonable power consumption in thermals and two 00:02:07.380 --> 00:02:15.239 they just don't seem to need it the most useful slide that AMD provided was this 00:02:12.120 --> 00:02:18.660 one where we see the 7900 XTX performing 00:02:15.239 --> 00:02:22.980 between one and a half to 1.7 times the 00:02:18.660 --> 00:02:25.800 outgoing Flagship 6950xt at 4K now AMD 00:02:22.980 --> 00:02:29.940 was frustratingly vague about the exact settings that they used to achieve these 00:02:27.720 --> 00:02:35.640 results even when we probed them after the event but we managed to extract just 00:02:33.300 --> 00:02:41.220 enough information to create our own set of performance graphs by extrapolating 00:02:37.800 --> 00:02:45.060 from our own reference 6950xt 00:02:41.220 --> 00:02:48.360 let's start with cyberpunk 2077 AMD's 00:02:45.060 --> 00:02:52.080 previous best managed 46 FPS 00:02:48.360 --> 00:02:54.900 not great but multiply that by 1.7 and 00:02:52.080 --> 00:03:00.540 we've got a spicy 78 frames per second that's just a single frame per second 00:02:57.480 --> 00:03:02.580 behind the RTX 4090 what a generational 00:03:00.540 --> 00:03:09.420 Improvement in Call Of Duty Modern Warfare 2 our RX 6950xt managed a 00:03:06.480 --> 00:03:15.420 respectable 90 FPS when we multiply that by AMD's 1.5 times assuming of course 00:03:12.720 --> 00:03:21.959 they're not just outright lying we can expect the 7900 XTX to lag NVIDIA's 00:03:18.239 --> 00:03:25.500 Flagship by only six percent hot dang 00:03:21.959 --> 00:03:27.840 finally in Watchdogs Legion the 7900 XTX 00:03:25.500 --> 00:03:33.500 takes its largest L compared to the RTX 4090 but that still only has it lagging 00:03:31.319 --> 00:03:39.060 behind by about 10 not too shabby AMD 00:03:36.480 --> 00:03:44.760 of course performance is meaningless if it'll cost you two fingers and your 00:03:41.099 --> 00:03:47.940 right butt cheek NVIDIA and I have more 00:03:44.760 --> 00:03:50.519 good news the 7900 XTX is a thousand 00:03:47.940 --> 00:03:54.959 dollars while the 7900 XT comes in a hundred less at 900 which 00:03:53.159 --> 00:04:00.840 I mean I never thought I'd be happy about even a top-tier GPU being only a 00:03:58.620 --> 00:04:05.819 thousand dollars but I guess that really shows just how out to lunch NVIDIA has 00:04:03.900 --> 00:04:10.980 been with the pricing of the RTX 4000 series but wait a second didn't NVIDIA's 00:04:08.640 --> 00:04:16.440 CEO Jensen go out on stage and basically say the pricing is so high because the 00:04:13.620 --> 00:04:21.359 factors outside is control Moore's Law is dead he said 00:04:18.299 --> 00:04:24.360 well how did AMD pull it off then well 00:04:21.359 --> 00:04:26.540 first of all by not using the latest and 00:04:24.360 --> 00:04:33.120 greatest manufacturing process rdna 3 gpus at least the ones we've seen 00:04:29.699 --> 00:04:37.139 so far are made using a combination of 00:04:33.120 --> 00:04:40.080 tsmc's five nanometer and six nanometer 00:04:37.139 --> 00:04:46.800 processes making this both the very first chiplet design GPU and way cheaper 00:04:44.400 --> 00:04:51.960 to manufacture than the four nanometer monolithic dies that NVIDIA uses for RTX 00:04:49.680 --> 00:04:55.320 4000 but you'd think there'd be a downside to that right like don't hold 00:04:53.639 --> 00:04:59.220 their manufacturing nodes have worse power consumption 00:04:56.699 --> 00:05:06.900 believe it or not the total board power of the 7900 XT and XTX is just 300 and 00:05:03.860 --> 00:05:08.880 355 Watts respectively and it's 00:05:06.900 --> 00:05:14.280 delivered by two good old-fashioned eight-pin PCIe power connectors I never 00:05:11.639 --> 00:05:19.139 thought I'd be adding not a fire hazard as a selling point for a GPU but here we 00:05:17.220 --> 00:05:24.720 are this also means that if you're upgrading from something like a GTX 1080 00:05:21.360 --> 00:05:26.340 TI or an RTX 2080 you probably don't 00:05:24.720 --> 00:05:32.520 need to upgrade your system power supply and you also probably don't need to 00:05:29.220 --> 00:05:35.160 upgrade your case the reference 7900 XTX 00:05:32.520 --> 00:05:39.479 is a much more reasonable 287 millimeters wide and two and a half 00:05:37.259 --> 00:05:45.000 slots tall compared to the 304 millimeters and three and a half Slots 00:05:41.820 --> 00:05:46.919 of the RTX 4090 again I never thought a 00:05:45.000 --> 00:05:51.120 GPU physically fitting in your computer would be a major selling point 00:05:48.960 --> 00:05:56.400 but here we are how did they do this then well AMD also 00:05:54.419 --> 00:06:01.380 packed in some really cool power management Innovations for starters they 00:05:59.400 --> 00:06:07.199 decoupled the clock speed of the shaders which now run at up to 2.3 gigahertz and 00:06:04.440 --> 00:06:11.400 the front end which runs at 2.5 we've actually seen this before in Mobile 00:06:08.580 --> 00:06:15.060 products but never in a desktop GPU and the main benefit of this approach is 00:06:13.199 --> 00:06:19.979 that it allows them to prioritize the speed of one over the other depending on 00:06:17.699 --> 00:06:23.940 the workload for instance if the gain you're running doesn't hit the shaders 00:06:21.539 --> 00:06:29.639 super hard maybe those are fine just sipping power at 1.8 gigahertz while the 00:06:27.060 --> 00:06:35.340 front end runs at full speed AMD says this can reduce total power consumption 00:06:31.500 --> 00:06:37.259 by 25 percent which might not be 00:06:35.340 --> 00:06:41.639 something you care about if you have a beefy power supply and a well-cooled 00:06:38.940 --> 00:06:45.479 case already though fear not AMD says there will be aib partner cards with 00:06:43.800 --> 00:06:50.880 more Cooling and higher power consumption to enable even higher clock 00:06:47.580 --> 00:06:52.979 speeds or AMD says anyway so far we've 00:06:50.880 --> 00:06:56.220 only only found this ASUS tough model with three eight pin connectors which 00:06:54.840 --> 00:07:01.020 might indicate higher power consumption but nothing to say of the clock speeds 00:06:59.039 --> 00:07:06.300 now possibly the dumbest problem with the RTX 4090 aside from the burning up 00:07:04.259 --> 00:07:11.340 12 pin power adapters is that it only supports DisplayPort 1.4 which given how 00:07:09.660 --> 00:07:15.660 freaking fast it is could actually become a problem in the very near future 00:07:13.319 --> 00:07:19.979 as new displays become available AMD on the other hand managed to avoid 00:07:18.060 --> 00:07:25.020 this embarrassing situation by including DisplayPort 2.1 which can support 00:07:22.680 --> 00:07:34.680 ridiculous resolutions and frame rate combinations like 4K 480 Hertz or 8K 165 00:07:31.380 --> 00:07:36.960 Hertz now obviously whether you go team 00:07:34.680 --> 00:07:41.699 green or team red your GPU won't be able to keep up with that today but with the 00:07:39.840 --> 00:07:47.160 way that upscaling and frame generation technology is progressing DP 2.1 is some 00:07:44.940 --> 00:07:51.599 massive future proofing that could make the RTX 4090 unable to take full 00:07:49.979 --> 00:07:56.699 advantage of monitors that could be releasing as early as next year 00:07:53.940 --> 00:08:03.360 AMD has also included a new media engine that has 8K 60 av1 and code and decode 00:08:01.139 --> 00:08:06.900 support this is a huge Boon for streamers who are looking to get the 00:08:04.740 --> 00:08:11.699 best image quality going forward and they announced smart access video which 00:08:09.360 --> 00:08:18.240 can share the load of encode and decode operations between AMD gpus and CPUs 00:08:15.000 --> 00:08:20.639 apparently giving a 30 uplift in 4k 00:08:18.240 --> 00:08:25.379 multi-stream transcoding though we'll have to see how well this actually works 00:08:22.440 --> 00:08:29.759 when it lands an OBS in December I'm a little skeptical but In fairness to AMD 00:08:27.960 --> 00:08:34.200 the recent software upgrades that they made to their media engine for Radeon 00:08:31.319 --> 00:08:37.500 6000 did bring about image quality improvements to get them closer to 00:08:35.940 --> 00:08:42.479 NVIDIA zenbank than I would have thought possible so with this new media engine 00:08:39.779 --> 00:08:46.920 AMD might finally be a sensible choice for people who want to stream or record 00:08:44.520 --> 00:08:51.420 their gameplay another sensible choice is signing up for Floatplane.com for 00:08:49.380 --> 00:08:55.260 amazing behind the scenes and Extras like the shooting of our starforge 00:08:53.279 --> 00:08:59.760 systems review intro to be clear not everything that AMD 00:08:57.240 --> 00:09:05.339 announced today was mind-blowing in AMD's own slides they show a 1.5 to 1.6 00:09:03.480 --> 00:09:10.320 times increase in Ray tracing performance which 00:09:06.720 --> 00:09:14.519 will take you from literal slideshow to 00:09:10.320 --> 00:09:17.760 cinematic FPS at best and the RTX 4090 00:09:14.519 --> 00:09:19.800 walked all over AMD's best at nearly 00:09:17.760 --> 00:09:25.740 double the frame rate based on our cyberpunk test and these projected 00:09:22.320 --> 00:09:27.980 numbers also AMD built new AI 00:09:25.740 --> 00:09:34.019 accelerators that should allow for up to 2.7 times more AI throughput per compute 00:09:31.380 --> 00:09:38.519 unit but the way it's implemented makes this increase only matter for games at 00:09:37.019 --> 00:09:43.860 the moment it's not going to help you with general purpose Ai and actually 00:09:41.040 --> 00:09:49.680 come to think of it there was almost no mention whatsoever of non-gaming tasks 00:09:46.980 --> 00:09:54.360 and this this is a long time thorn in Radeon users sides some applications 00:09:52.260 --> 00:09:59.760 like blender or SolidWorks visualize only support NVIDIA Cuda making NVIDIA 00:09:57.300 --> 00:10:04.320 cards absolutely embarrass their AMD counterparts and it doesn't look like 00:10:02.100 --> 00:10:10.320 this is going to change anytime soon AMD also promised that their new fsr3 00:10:07.800 --> 00:10:15.420 upscaling would be up to a two times increase over fsr2 but it looks like at 00:10:14.220 --> 00:10:21.180 least one of the ways they're going to achieve this is with technology similar 00:10:17.880 --> 00:10:23.160 to NVIDIA's dlss 3 by generating new 00:10:21.180 --> 00:10:28.140 frames and well let's just say we're currently 00:10:25.860 --> 00:10:32.220 working on a video about dlss3 and its frame generation and 00:10:29.640 --> 00:10:37.920 it's a first gen product overall though it looks like the RX 7900 XT and XTX are 00:10:35.760 --> 00:10:41.160 going to be hacking fast gpus for the money and this might end up being the 00:10:39.779 --> 00:10:46.800 first generation of what feels like forever where AMD is positioned to grab 00:10:44.220 --> 00:10:51.660 a good chunk of market share from our green overlords which will hopefully 00:10:49.380 --> 00:10:56.940 push NVIDIA to compete harder on pricing and meanwhile give AMD's GPU division 00:10:54.600 --> 00:11:01.260 the cash that it needs to ramp up r d spending and come back even stronger 00:10:59.040 --> 00:11:05.220 next time I can't wait to give you guys the full review when these cards launch 00:11:02.820 --> 00:11:09.899 on December the 13th just like I can't wait to tell you about this segue to our 00:11:07.380 --> 00:11:13.980 sponsor micro center micro center carries a wide variety of laptops 00:11:12.480 --> 00:11:18.540 computer components monitors communication devices TVs networking you 00:11:16.260 --> 00:11:23.220 name it they've got it with over 30 000 items in stock they are 25 stores across 00:11:21.360 --> 00:11:26.820 the US provide amazing selection and competitive prices so you always leave 00:11:25.200 --> 00:11:30.540 with what you're looking for and we've done multiple video projects at Micro 00:11:28.800 --> 00:11:34.320 Center locations and have always been impressed by the staff's knowledge and 00:11:32.519 --> 00:11:37.980 the stock that's available Microcenter has great deals both in-store and online 00:11:35.940 --> 00:11:43.500 so click the link in the description to check them out today if you guys enjoyed 00:11:40.440 --> 00:11:45.660 this video go check out our RTX 4090 00:11:43.500 --> 00:11:50.459 undervolting video where the power consumption of the 4090 starts to look 00:11:47.880 --> 00:11:52.880 not that bad even if the price is still obscene