WEBVTT 00:00:00.000 --> 00:00:04.120 The way I see it, you clicked on this video for one of just three reasons. 00:00:04.120 --> 00:00:08.560 You either want to watch tech giants in a CPU dunking contest. 00:00:08.560 --> 00:00:13.040 You could legitimately be shopping for the fastest PC hardware on the planet. 00:00:13.040 --> 00:00:17.840 Congratulations, by the way. Or, most likely, you just wanna eat some popcorn 00:00:17.840 --> 00:00:21.920 and watch the fanboys squabble in the comments about an overpriced CPU. 00:00:21.920 --> 00:00:26.160 Well, guess what? Buckle in tight, cause you're gonna see all three today. 00:00:26.160 --> 00:00:29.840 The Ryzen 9950X3D2 Dual Edition 00:00:29.840 --> 00:00:33.440 is an eye-wateringly expensive, zero-compromise product, 00:00:33.440 --> 00:00:37.280 and I love it. It's the perfect purebred show horse 00:00:37.280 --> 00:00:42.840 to excite geeks like us by pushing the limits of modern technology, while also serving as a reminder 00:00:42.840 --> 00:00:46.600 that most people really don't need one. 00:00:46.600 --> 00:00:50.560 What everyone does need, though, is this segway to our sponsor. 00:00:51.560 --> 00:00:56.080 Hi, if you're wondering why this video is a day late, well, AMD didn't send us a chip to review, 00:00:56.080 --> 00:00:59.240 and they also didn't send a lot of people chips to review. 00:00:59.240 --> 00:01:03.080 Make of that information what you will. Anyway, here is our review. 00:01:21.000 --> 00:01:25.560 It's a 9950X, so that means you got 16 of AMD's latest 00:01:25.560 --> 00:01:28.760 Zen 5 architecture cores and 32 threads. 00:01:28.760 --> 00:01:33.120 It's 3D, which means it features a large stacked 3D vCash, 00:01:33.120 --> 00:01:37.400 which can offer a major performance benefit in certain workloads, notably gaming, 00:01:37.400 --> 00:01:41.320 and it's two, and also dual, 00:01:41.320 --> 00:01:47.800 both of which seem to mean the same thing, because compared to a regular 9950X3D non-two, non-dual, 00:01:47.800 --> 00:01:54.600 the big change that the dual edition brings is that AMD's super large 3D vCash now sits on both 00:01:54.600 --> 00:01:58.240 of the core complex dies or CCDs that are under the gray heat spreader 00:01:58.240 --> 00:02:05.360 rather than only on one of them. That gives the dual a whopping 192 megabytes of level 3 cache. 00:02:05.680 --> 00:02:09.240 That's enough to install Windows 95 on your CPU 00:02:09.240 --> 00:02:15.120 three times over. Doesn't really work like that. We just thought it was kind of a funny thing to point out. 00:02:15.120 --> 00:02:19.320 But hold on a second, if 3D vCash is so great, then why didn't AMD use it 00:02:19.320 --> 00:02:25.280 across all of their cores in the past? Well, because there are some trade-offs, the price for one, 00:02:25.280 --> 00:02:28.480 but also a considerable 30 watt increase 00:02:28.480 --> 00:02:34.480 to total design power that brings the dual to a sweaty 200 watts that in some situations, 00:02:34.480 --> 00:02:39.080 it even exceeded by a significant margin. And this extra power seems to have resulted 00:02:39.080 --> 00:02:42.380 in a small decrease in boost clock speeds. 00:02:43.880 --> 00:02:47.640 But hey, that should be more than made up for by the extra 3D vCash, right? 00:02:47.640 --> 00:02:52.280 Especially in gaming, right? Well, let's talk about that. Out of the gate, things are looking good 00:02:52.280 --> 00:02:55.720 in city skylines too. A game we chose more for its ability 00:02:55.720 --> 00:03:00.000 to bring CPUs to their knees than its community appeal. 00:03:00.000 --> 00:03:05.400 Here, the dual leads the pack with the other 9000 X3D chips in tow behind it. 00:03:05.400 --> 00:03:11.200 Makes sense? After all, they're supposed to be good for gaming, right? And we see the same deal in Cyberpunk 2077. 00:03:11.200 --> 00:03:15.760 Though it is worth noting here that while these high average FPS numbers here 00:03:15.800 --> 00:03:19.640 are very impressive, when it comes to the all important 1% lows, 00:03:19.640 --> 00:03:23.560 which represent the performance you can expect in the most demanding moments, 00:03:23.560 --> 00:03:27.800 some older and much less expensive chips 00:03:27.800 --> 00:03:30.920 are nipping dangerously close to their heels. 00:03:30.920 --> 00:03:34.240 In F124, it seems like any high end processor 00:03:34.240 --> 00:03:37.480 from the past couple generations is gonna do ya just fine. 00:03:37.480 --> 00:03:40.520 Shout out to the venerable 5800 X3D, by the way, 00:03:40.520 --> 00:03:45.200 which is rumored to be returning for a celebration of the AM4 Socket's 10th anniversary. 00:03:45.200 --> 00:03:49.880 Then in the last of us part, one, the 9950 X3D2 dual returns to the top position 00:03:49.880 --> 00:03:52.880 and is also about as fast as it gets in Counter-Strike 2, 00:03:52.880 --> 00:03:57.240 at least as far as 1% lows are concerned. Across our suite of gaming benchmarks, 00:03:57.240 --> 00:04:02.480 the Ryzen 9 9950 X3D2 dual edition is just plain ripping fast. 00:04:02.480 --> 00:04:06.480 And yet, still nearly impossible for me to recommend. 00:04:06.480 --> 00:04:11.520 Why? Well, for starters, we benchmark gaming CPUs at 1080p. 00:04:11.520 --> 00:04:16.120 This gives us a clear view of which chips will separate themselves from the pack 00:04:16.120 --> 00:04:22.480 in an intentionally CPU bottleneck situation. But if you're buying a $900 chip, 00:04:22.480 --> 00:04:25.560 you're almost guaranteed to be running at at least 1440p, 00:04:25.560 --> 00:04:31.120 if not 4K, where a GPU bottleneck is way more likely than a CPU one. 00:04:31.120 --> 00:04:36.500 The second reason is that even if you were an eSports professional trying to eek out every last frame, 00:04:36.500 --> 00:04:41.440 I still wouldn't sell you a dual because while it does have more 3DV cash goodness 00:04:41.440 --> 00:04:47.080 than anything we've ever seen before, which is cool. In the real world, it performs basically the same 00:04:47.080 --> 00:04:50.480 as any of the other 9000 series X3D chips. 00:04:50.480 --> 00:04:53.880 But why is that? Well, as some of you might know, 00:04:53.880 --> 00:04:58.240 high-end X3D chips have always come with kind of a weird trade-off. 00:04:58.240 --> 00:05:02.920 They had all these cores on them, but only some of the cores were best for gaming, 00:05:02.920 --> 00:05:06.120 the 3DV cash ones. So you always wanted to make sure 00:05:06.120 --> 00:05:11.280 that your games were running on those. To do this, AMD uses a technique called core parking, 00:05:11.280 --> 00:05:15.080 which unintuitively does not stop the core from doing any work, 00:05:15.080 --> 00:05:21.680 but rather it forces a given program to only use specific CPU cores and then ignore the rest. 00:05:21.680 --> 00:05:26.080 Over the past couple of years, AMD and Microsoft have worked on de-jancifying this behavior 00:05:26.080 --> 00:05:30.400 to the point where as long as you're running an up-to-date chipset, BIOS, and operating system, 00:05:30.400 --> 00:05:36.080 you should find that your two CCD CPU is using the right cores for any given job. 00:05:36.080 --> 00:05:40.400 That is assuming, of course, that Windows recognizes your games as a game. 00:05:40.400 --> 00:05:45.120 But we have 3DV cash on all the cores now. So why are we talking about parking anything anywhere? 00:05:45.120 --> 00:05:48.480 Vroom, vroom, right? Well, as it turns out, 00:05:48.480 --> 00:05:52.040 most games just plain don't need more than eight cores. 00:05:52.040 --> 00:05:58.760 And regardless of what type of cash your cores have, if they have to coordinate work across multiple CCDs, 00:05:58.760 --> 00:06:01.960 the data has to travel across the Infinity Fabric 00:06:01.960 --> 00:06:05.560 and the IO die, which is gonna introduce extra latency. 00:06:05.560 --> 00:06:09.880 So the solution, the same jank we were already using, 00:06:09.880 --> 00:06:13.480 core parking. And that is why the new dual edition performs 00:06:13.480 --> 00:06:18.440 basically exactly the same in gaming. So then, who is it useful for? 00:06:18.440 --> 00:06:22.160 Well, for starters, anyone who might run multiple workloads 00:06:22.160 --> 00:06:25.960 that benefit from extra cash, a great example would be someone who wants to stream 00:06:25.960 --> 00:06:30.600 or record their gameplay while they're gaming without burdening their GPU. 00:06:30.600 --> 00:06:35.160 A more niche benefit is that overclockers and system tuners could identify 00:06:35.160 --> 00:06:40.320 which of their CCDs runs the fastest and then put their most performance sensitive workloads 00:06:40.320 --> 00:06:43.400 on those cores. Or at least that would be an option 00:06:43.400 --> 00:06:48.160 if AMD and Windows allowed you to choose. From our testing, it seems that games are locked 00:06:48.160 --> 00:06:52.720 to CCD zero for the time being. Another focus for AMD's marketing for this chip 00:06:52.720 --> 00:06:56.640 is a claimed five to 10% increase in productivity. 00:06:56.640 --> 00:07:02.600 Let's take a look at that. And unsurprisingly, the Ryzen 9 9950X3D2 dual edition 00:07:02.720 --> 00:07:06.880 Core 2 dual extreme max plus the sequel is an absolute monster here as well. 00:07:06.880 --> 00:07:10.680 In seven zip, the dual has a clear lead over any other consumer CPU, 00:07:10.680 --> 00:07:15.340 both in compression and in decompression. And in Cinebench, it sits at that top of the chart 00:07:15.340 --> 00:07:19.040 or darn near it in both multicore and single core performance. 00:07:19.040 --> 00:07:26.920 The only problem is that its closest competitor is a, $550 cheaper at retail right now. 00:07:28.280 --> 00:07:33.340 But hey, those are kind of just benchmarks, right? And if you're in the market for a CPU this expensive, 00:07:33.340 --> 00:07:38.520 you probably want to get some real work done with it, right? We're in blender, the dual edition does manage 00:07:38.520 --> 00:07:43.040 and as advertised, 5% improvement over its less endowed twin 00:07:43.040 --> 00:07:48.240 and a considerable 15 ish percent over the 7950X3D. 00:07:48.240 --> 00:07:53.880 This is one of the first cases we've seen where the dual actually kind of stands out for its value. 00:07:53.880 --> 00:07:57.080 To be clear, it only beats the much more affordable 00:07:57.080 --> 00:08:02.240 270K plus by 10 seconds. But if you use your system to make money, 00:08:02.240 --> 00:08:05.400 every second counts and the dual is as fast as it gets 00:08:05.440 --> 00:08:09.800 without paying a considerable premium for a workstation platform motherboard. 00:08:09.800 --> 00:08:15.360 And it even continues AMD's bro move of supporting ECCU dims on select motherboards, 00:08:15.360 --> 00:08:20.160 making it a great semi pro offering. And that carries over to encoding as well, 00:08:20.160 --> 00:08:26.120 where the X3D2 is again at the top of the charts in H.264 and close in AV1. 00:08:26.120 --> 00:08:31.520 In Puget Bench's Photoshop benchmark, the dual is the first consumer x86 CPU we've tested 00:08:31.520 --> 00:08:34.920 that cracks 14,000 points, which is cool, 00:08:34.920 --> 00:08:38.920 but only a negligible increase over the rest of the Zen 5 contenders. 00:08:38.920 --> 00:08:44.000 And NVIDIA editing, she's a chart topper again, and more importantly, can offer an uplift 00:08:44.000 --> 00:08:48.360 that you might actually feel compared to an older generation 3DV cache chip 00:08:48.360 --> 00:08:51.720 like the 7950X3D. Now you wouldn't think that 00:08:51.720 --> 00:08:56.640 based on the overall score from Puget Bench, but we found that depending on the files you're working with, 00:08:56.640 --> 00:09:01.660 you might find performance enhancements that are much greater than the two ish percent on our chart. 00:09:01.660 --> 00:09:05.760 However, as with gaming, the situations where we see large improvements 00:09:05.760 --> 00:09:09.240 over the non-dual, are few and far between. 00:09:09.240 --> 00:09:12.420 As for Premiere Pro, it's kind of like a grade schooler going to the gym. 00:09:12.420 --> 00:09:17.480 The games are there, but they're minor. So yeah, in our creator focus testing, 00:09:17.480 --> 00:09:21.880 the dual doesn't offer much, but to be fair, there are lots of kinds of productivity 00:09:21.880 --> 00:09:27.020 and the extra level three cache can be huge in mathy and science C workloads. 00:09:27.020 --> 00:09:31.880 Check out this result from Veronica's review of the original 9950X3D. 00:09:31.920 --> 00:09:37.160 Since we're a day late, we have the rare second mover advantage and we can actually just show you some of Veronica's results. 00:09:37.160 --> 00:09:41.720 And you can see that there's some workloads where the chip benefits a lot from the extra V cache. 00:09:41.720 --> 00:09:47.800 You might be thinking, that's not very many workloads and yeah, yeah. 00:09:47.800 --> 00:09:50.840 Anyways, if you do want to learn more, we highly recommend going over to Veronica's, 00:09:50.840 --> 00:09:54.720 looking at their Linux and scientific benchmarking. It's pretty cool. 00:09:54.720 --> 00:09:58.080 There's going to be a link in the description. But I will say there's a solid chance 00:09:58.080 --> 00:10:01.880 that if you are de-gritting radio signals from space, 00:10:01.880 --> 00:10:06.440 this could be your new powerhouse. Speaking of power, at 200 Watts, 00:10:06.440 --> 00:10:10.720 this is the highest TDP that AMD has ever set on a consumer chip. 00:10:10.720 --> 00:10:17.360 And who does it ever deliver on its promise? In Cinebench, this thing pulls up to 260 Watts 00:10:17.360 --> 00:10:23.280 across the total CPU package. That is like Intel 1400K levels of power draw. 00:10:23.280 --> 00:10:27.000 But thankfully, it doesn't have the same power density 00:10:27.040 --> 00:10:30.040 and IHS problems as the 1400K. 00:10:30.040 --> 00:10:35.200 With that said, if you were thinking of upgrading from a Ryzen 5950X to this new CPU, 00:10:35.200 --> 00:10:38.400 you should make sure that your power supply can handle the extra power draw. 00:10:38.400 --> 00:10:41.760 And oh yeah, probably also your CPU cooler. 00:10:41.760 --> 00:10:46.560 That's a tech tip. The good news in all of this is that the extra power draw only rears its head 00:10:46.560 --> 00:10:49.640 in heavy all-core CPU workloads. 00:10:49.640 --> 00:10:54.560 In F124, for example, we found this chip drew about the same power as its siblings, 00:10:54.560 --> 00:11:00.720 meaning that you can enjoy manageable temperatures while just pretending to race in the Las Vegas heat. 00:11:00.720 --> 00:11:04.360 In conclusion, the Ryzen 9950X3D2 Dual Edition 00:11:04.360 --> 00:11:07.800 puts a punctuation point on AMD's last three years 00:11:07.800 --> 00:11:12.880 of CPU domination. But while I love the new fastest one, 00:11:12.880 --> 00:11:19.240 as much as anyone does, it's hard to recommend a chip that demands such a price for such small gains 00:11:19.240 --> 00:11:22.360 and it feels a little bit like this is an unnecessary 00:11:22.360 --> 00:11:26.840 victory lap and I would rather AMD focus their development efforts on things their community 00:11:26.840 --> 00:11:33.000 really wants, like oh, I don't know, maybe bringing FSR4 to our DNA3 GPUs 00:11:33.000 --> 00:11:37.620 or hey, how about finally bringing Zen 5 to budget buyers? 00:11:37.620 --> 00:11:42.920 Where is Ryzen 3 AMD? Also, where is my sponsor? 00:11:42.920 --> 00:11:48.400 If you guys enjoyed this video, why not check out the one we did recently on how X3D chips in particular 00:11:48.400 --> 00:11:52.480 don't really need the fastest memory for gaming which could help save you some money 00:11:52.480 --> 00:11:54.320 if you're putting together a new system.