WEBVTT 00:00:00.000 --> 00:00:07.440 I have in my very hands what is claimed to be the world's best processor for gamers and creators. 00:00:07.440 --> 00:00:18.480 This is the brand new AMD Ryzen 9950X3D, a 16-core, fairly expensive CPU that promises 00:00:18.480 --> 00:00:24.080 to make basically no compromises. That's right, you can have a CPU that has the horsepower for 00:00:24.080 --> 00:00:32.000 productivity without making compromises to gaming. At least that's what AMD says with this brand new 00:00:32.000 --> 00:00:36.960 chip in here. Ooh, look at that. Man, CPU unboxings are so exciting. Hey, look, we're done. All right, 00:00:36.960 --> 00:00:44.560 video over. For the uninitiated here, this is one of AMD's 3D vCash-enabled CPUs. So they basically 00:00:44.560 --> 00:00:52.880 take their normal CPUs, which to be fair, a 9950X, not a normal CPU, kind of the top dog, at least 00:00:52.880 --> 00:01:00.240 until today. And then they add a bunch of cash. And no, not like money. I mean, you need to add 00:01:00.240 --> 00:01:06.000 money. But in exchange, they give you a different type of cash, cash that lives on the CPU. It's 00:01:06.000 --> 00:01:11.120 kind of like the RAM in your system, where it holds the stuff you're actively working on for quick 00:01:11.120 --> 00:01:18.640 access, except it's way faster than RAM, like, inordinately faster than RAM. And so when you 00:01:18.640 --> 00:01:25.040 can have a good bit more than a normal CPU, certain applications and especially games get a huge 00:01:25.040 --> 00:01:30.240 performance increase. And that is the sell of this thing. And there's even some applications that 00:01:30.240 --> 00:01:36.160 aren't games that get a performance buff. Compared to the normal 9950X, which is the 16 core chip, 00:01:36.160 --> 00:01:44.480 this 9950X3D has double the L3 cash from 64 megabytes to 128 megabytes. And you might not 00:01:44.480 --> 00:01:53.120 think that that is that much, but that L3 cash is literally multiple, multiple times faster than 00:01:53.120 --> 00:01:59.120 your system memory. And if you can keep crucial information really close to the CPU, that means 00:01:59.120 --> 00:02:04.560 you can make things go really fast. Now this all sounds great in theory, but how does it look in 00:02:04.640 --> 00:02:14.240 practice? Let's get test bench. Oh my God. This is heavy. We've got an x870e AORUS master motherboard 00:02:14.240 --> 00:02:22.960 here with some 6000cl30 TridentZ Neo RAM. The AMD does still recommend 6000 mega transfers per second 00:02:22.960 --> 00:02:30.960 memory, just with low latency. Now compared to the previous generation, you know, the 7950X3D was 00:02:30.960 --> 00:02:39.200 a good CPU, but it did have its problems specifically. Now, if you look at the inside of these CPUs, 00:02:39.200 --> 00:02:45.760 it's not just one CPU like, well, most of the CPUs have been for many, many years. AMD uses 00:02:45.760 --> 00:02:50.240 something called chiplet technology, and they actually have basically two separate kind of like 00:02:50.240 --> 00:02:56.720 CPUs fused together on a PCB and they call them chiplets, except that only one of the chiplets, 00:02:56.720 --> 00:03:02.800 or eight cores of the 16 cores in the CPU gets extra L3 cache. The other eight just has the 00:03:02.800 --> 00:03:07.920 normal amount. So if you're playing something like a game and the game decides to try to use the cores 00:03:07.920 --> 00:03:11.280 that don't have the extra cache, well, you're going to have a performance hit. You're not getting 00:03:11.280 --> 00:03:16.480 that benefit. So on the last generation, it was sometimes a pretty noticeable issue where you 00:03:16.480 --> 00:03:21.200 were taking a big performance hit. So if you only play video games and you barely do any productivity 00:03:21.280 --> 00:03:27.200 stuff, where you can really take advantage of like 16 cores, it wasn't a good idea to buy 00:03:27.200 --> 00:03:32.080 one of those chips. But on this generation, thanks to some software stuff, which we'll see 00:03:32.080 --> 00:03:37.920 shortly, they in theory fixed that, which we'll show you in a sec. I just need to get a cooler on 00:03:37.920 --> 00:03:43.360 here with some PTM7950 phase change thermal pad from LTTstore.com. For graphics card, we have an 00:03:43.360 --> 00:03:49.680 RTX 4090. This isn't the latest and greatest from NVIDIA, but it is what AMD tested with in their 00:03:49.680 --> 00:03:55.120 press deck. So that is what we're using for now. While we wait for that to turn on, let's talk a 00:03:55.120 --> 00:04:01.760 little bit more about the specific specifications of this CPU compared to last generation. A lot of 00:04:01.760 --> 00:04:08.640 the specs on paper look the same. So most of this is kind of architectural. However, there are some 00:04:08.640 --> 00:04:15.680 big ones compared to the 7950X3D. The new version has the 3DV cache underneath the chip rather than 00:04:15.680 --> 00:04:21.200 on top and they use some fancy technology called through silicon BIOS to electrically connect it 00:04:21.200 --> 00:04:27.920 through the 3DV cache, which you might wonder why the big thing is now you can have those cores 00:04:27.920 --> 00:04:32.800 sitting on top running at whatever speed you want overclock to whatever you want, whatever power you 00:04:32.800 --> 00:04:38.160 want, and you don't have that insulating layer like stuck on top, preventing you from cooling the 00:04:38.160 --> 00:04:46.800 cores properly. So this new chip now has 170 watt TDP like the 9950X non 3D. The previous generation 00:04:46.800 --> 00:04:53.840 was limited at 120 watts. We've got the nice new TSMC 4nm node. We have a little bit more L1 cache 00:04:53.840 --> 00:05:01.280 than last generation, but other than that, it's kind of like a 9800X3D with eight more cores 00:05:01.360 --> 00:05:07.920 slapped right next to the ones with 3DV cache. And in theory, hopefully in games, it performs 00:05:07.920 --> 00:05:12.640 pretty much the same. There is technically higher max boost clock, which is interesting, 00:05:13.280 --> 00:05:18.560 but what I suspect we'll see is that the cores without the 3DV cache, the extra ones are going 00:05:18.560 --> 00:05:23.040 to be able to clock higher than the ones that do have it. And from what I understand, there's 00:05:23.040 --> 00:05:29.920 actually a setting in the BIOS now where you can pick your priority basically. So if you run applications 00:05:29.920 --> 00:05:35.200 that don't really use the 3DV cache and you could benefit from higher clock speed, you can 00:05:35.200 --> 00:05:39.280 actually tell it in the BIOS to prioritize using the cores with higher clock speed rather than the 00:05:39.280 --> 00:05:43.360 ones with Vcache, which is the opposite of the default behavior. However, if that's the case, 00:05:43.360 --> 00:05:50.240 you probably shouldn't buy this CPU. And look, wow, it's a computer. You can even fit our sponsor on 00:05:50.240 --> 00:05:56.400 here. And now that we're booted up, we can take a look and make sure our cache is there. Woo, 128 00:05:56.400 --> 00:06:04.960 megabytes of L3 cache. Let's take a look at how this all works. Now I got this cool piece of 00:06:04.960 --> 00:06:10.640 software called Park Control, which is supposed to let you tweak like core parking, which if you're 00:06:10.640 --> 00:06:16.480 not familiar, basically just means that core is being like kind of temporarily turned off so that 00:06:16.480 --> 00:06:21.280 the software on the computer and the operating system tries not to use them. And that is very 00:06:21.280 --> 00:06:26.080 important. Like I mentioned before, when we're in a game, we only want to be using the cores that 00:06:26.080 --> 00:06:31.040 have 3dv cache enabled since most games can't take advantage of more than like a or so cores 00:06:31.040 --> 00:06:34.720 anyways. And if you start dipping into the cores with less cache, usually you get a performance 00:06:34.720 --> 00:06:41.440 loss. So I'm gonna try I'm gonna try and load the built in solitaire. And let's see if we can see a 00:06:41.440 --> 00:06:48.480 look at that immediately. So if I'm focused on solitaire, you can see that half of the CPU cores 00:06:48.640 --> 00:06:56.320 just a 16 of 32. So that's the eight 3dv cache cores as well as their extra multi threads are 00:06:56.320 --> 00:07:01.520 enabled. And then the other half are completely parked or turned off so that the operating system 00:07:01.520 --> 00:07:05.840 doesn't use them. And if we go into task manager, this is taking a sweet time. Okay, you can see 00:07:05.840 --> 00:07:11.120 we've got like the first block of cores or some of them are being used at least and then the rest 00:07:11.120 --> 00:07:15.920 of them had a spike. And then now that the cores are parked, there's no activity going on there 00:07:15.920 --> 00:07:20.560 whatsoever. Now specifically, they've updated their provisioning package, as well as the 3dv 00:07:20.560 --> 00:07:25.840 cache optimizer, both features that are part of the AMD chipset driver. So if you have an existing 00:07:25.840 --> 00:07:32.080 7000 series x 3d chip like the 7950 x 3d with two CCDs, highly recommend updating the chipset 00:07:32.080 --> 00:07:36.240 driver, you're going to get pretty much probably some free performance in certain games. And then 00:07:36.240 --> 00:07:40.560 they also added a new thing, which actually I can see in task manager here, we can see that the 00:07:40.640 --> 00:07:45.520 cache performance optimizer service, but there's also now an application compatibility database 00:07:45.520 --> 00:07:50.960 service. Now what this application compatibility database service actually does, it uses a Windows 00:07:50.960 --> 00:07:56.880 compatibility toolkit called processor count lie, which as the name implies, it'll just lie to the 00:07:56.880 --> 00:08:02.320 app and say, Oh, I only have eight CPU cores and 16 threads, so that they only use the v cache. 00:08:02.320 --> 00:08:06.480 And it's for a specific set of games. And it's important to note, you need game bar installed 00:08:06.480 --> 00:08:10.640 if you want all of these optimizations to work because they rely on game bar 00:08:10.640 --> 00:08:15.120 to know whether or not a game is running, at least to some degree. And that's also helpful 00:08:15.120 --> 00:08:20.480 because if it doesn't detect it for some reason, you can manually tag something as a game with game 00:08:20.480 --> 00:08:26.080 bar. And then it should start to work, which is I want to try that actually. Oh, I have to go 00:08:26.080 --> 00:08:30.720 Windows G while it's the active window. Oh, I see. Okay, now and then I go, Hey, there we go. Okay, 00:08:30.720 --> 00:08:35.760 so remember, this is a game tech power up GPU Z. So if I click that, close this now, go back. 00:08:36.480 --> 00:08:43.120 Look at that. It works. You see now we've got 16 of 32 cores only, half of them are parked 00:08:43.120 --> 00:08:50.080 when we're using GPU Z. And if I click off of it, take it out of focus. Boom, it unparks the 00:08:50.080 --> 00:08:56.560 course. So the system works. Now I know that some people don't like game bar. But this way, 00:08:56.560 --> 00:09:01.440 it's like a thing that's already built into Windows that you can use to flag apps. And you 00:09:01.440 --> 00:09:05.920 don't have to run any like weird software. It's already built into the chipset driver. 00:09:05.920 --> 00:09:11.680 There's no like separate Intel performance optimizer app that you have to open and play with. 00:09:11.680 --> 00:09:17.120 It just pretty much just works. Let's look at the geo mean of all of the games we tested at low, 00:09:17.120 --> 00:09:23.920 which is only the three, it does still technically put the 9800 x 3d on top at low, but it's a close 00:09:23.920 --> 00:09:29.120 race. And this again is only across three games. I'm sure if you were to test across even more 00:09:29.120 --> 00:09:36.160 games, you would see probably a tighter variance. But overall, that is a really good showing. That 00:09:36.160 --> 00:09:40.560 makes me really excited about this chip. Wow, Rocket League can actually use some cores, 00:09:40.560 --> 00:09:47.520 even at 1080p max settings, the 9950 x and x 3d are on top neck and neck, then the 9800, 00:09:47.520 --> 00:09:53.200 and then Intel and last gen on the bottom. We've got Warhammer. Yeah, that seems pretty GPU limited 00:09:53.200 --> 00:09:59.520 at ultra. They're all the same same story as we saw in f 124. Oh, I almost forgot Stellaris. This 00:09:59.520 --> 00:10:05.440 is kind of like a top down RTS game. And this benchmark specifically simulates one year of 00:10:05.440 --> 00:10:11.680 game time. And while the numbers aren't crazy different, this game is super consistent in 00:10:11.680 --> 00:10:18.560 the benchmark with these numbers only varying on a given CPU, about one second, if that even so 00:10:18.560 --> 00:10:24.880 the 9950 x 3d and the 9800 x 3d are basically on par. But we do have a bit of speed up from Intel 00:10:24.880 --> 00:10:30.640 from the non 3d v cache variant and definitely a huge speed up from last gen. And that is good to 00:10:30.640 --> 00:10:37.920 see in all the games we tested pretty much the 9950 x 3d is performing as AMD said it's damn good 00:10:37.920 --> 00:10:43.280 for gaming. But let's look at productivity because that's what this chip is really for the content 00:10:43.360 --> 00:10:49.840 creator, the engineer, the whatever who also games and look at that Puget Bench in Photoshop, 00:10:49.840 --> 00:10:54.720 it's really not a big difference. It's a bit faster with a few more cores, but man, 00:10:54.720 --> 00:10:58.000 this is really close and in Premiere, you get a little bit of a bump. But overall, 00:10:58.000 --> 00:11:03.120 this new chip is still on top, even if the performance improvements are not huge in Adobe 00:11:03.120 --> 00:11:08.480 apps, we all know how well optimized those are in the blender monster benchmark. It's neck and neck 00:11:08.480 --> 00:11:14.080 with the non 3d v cache and faster than the rest. Good to see Cinebench. Man, Intel does really 00:11:14.080 --> 00:11:21.840 good in Cinebench. Either way, this is the only test we did, I think where Intel's 285k is actually 00:11:21.840 --> 00:11:27.280 on par with these AMD chips. And that is like really close. If we go to single core Intel wins 00:11:27.280 --> 00:11:32.400 out a tiny bit in handbrake. Oh, I lied. Intel actually does have another spot where they're 00:11:32.400 --> 00:11:40.000 winning here a little bit above everything else. But in h264, the 9950 x 3d and its non 3d v cache 00:11:40.000 --> 00:11:46.800 counterpart are on top with a good lead. So in terms of productivity, the 9950 x 3d is basically 00:11:46.800 --> 00:11:54.400 on top unless you're happened to be AV one encoding on CPU or running Cinebench. In the other 00:11:54.400 --> 00:11:59.840 applications, it was right up there. And in games, it seems to perform pretty much neck and neck with 00:11:59.840 --> 00:12:06.720 the 9800 x 3d. Except for a couple edge cases, there still seems to be. I'd be interested to see 00:12:06.720 --> 00:12:13.040 if city skylines does see further improvements with some future chipset driver updates, or if you 00:12:13.040 --> 00:12:18.160 might be able to remedy that with the gamebar trick I told you guys earlier. But what I want to try 00:12:18.160 --> 00:12:24.240 now is all of those tests were done with PBO off. And if you're not familiar, PBO is like AMD's 00:12:24.240 --> 00:12:29.200 automatic overclocking thing. You turn it on in the BIOS and it kind of removes a bunch of the power 00:12:29.200 --> 00:12:34.480 limits and stuff because this is a fully unlocked chip. I want to just do that and run Cinebench 00:12:34.480 --> 00:12:38.160 and see where we get. Oh, actually, before we do that, let's check out the power and thermals, 00:12:38.160 --> 00:12:45.760 because I'm really interested in that. The 9950 x 3d average CPU power during a Cinebench run 00:12:45.760 --> 00:12:53.840 around 200 watts. Our temperatures were around 75. It's clocking, you know, between 5.2 and 5.5 00:12:53.920 --> 00:12:59.920 ish gigahertz. Even though Intel did edge out like a tiny bit of a win, it's not a fair bit more 00:12:59.920 --> 00:13:06.000 power. You're talking at least 10% more power and it wasn't 10% more multi-threaded performance 00:13:06.000 --> 00:13:13.120 by any stretch of the imagination. And at max, it was more than 10%. You're talking 20, 25% 00:13:13.120 --> 00:13:19.520 more power. Damn, it almost seems like AMD made like a really good CPU here. This thing crushes in 00:13:19.520 --> 00:13:25.840 gaming, crushes in productivity, and it does it while drawing less power. Even like in a game, 00:13:25.840 --> 00:13:32.000 if we look at f1224 at 4k with ray tracing on, which is not going to be a huge CPU load, 00:13:32.000 --> 00:13:37.680 you know what the average of the 285k is a tiny bit lower in this very GPU bottlenecked test. I mean, 00:13:37.680 --> 00:13:43.840 the 9800 x 3d is like barely even drawing power if you're going for efficiency. Now, to be clear, 00:13:43.840 --> 00:13:48.800 these are two very limited situations. There are probably circumstances, depending on your 00:13:48.800 --> 00:13:53.840 application where one might be more efficient than the other, but overall seems like a very good 00:13:53.840 --> 00:14:01.920 showing for AMD. And those Intel chips are very efficient in certain applications. Honestly, 00:14:01.920 --> 00:14:07.920 not a ton higher on the wattage. I think we said the average before was like 200 and now it's 00:14:08.880 --> 00:14:14.800 to 26 ish. Based on the data I can see from the Labs testing with pbo on, it looks like we're 00:14:14.800 --> 00:14:23.920 getting about 100 megahertz more on average and our score is somehow lower. Compared to the motherboard 00:14:23.920 --> 00:14:31.440 default pbo not set to enabled but auto, we're doing about 20 25 watts more on average and getting 00:14:31.440 --> 00:14:39.600 somewhere between 70 to 120 more megahertz on average across the cores. But we did get a little 00:14:39.600 --> 00:14:45.920 bit lower score. So who knows? It it's definitely not something that's always going to make your 00:14:45.920 --> 00:14:50.000 stuff better. I kind of want to just play games on this thing now. But like overall, 00:14:50.000 --> 00:14:56.960 this CPU seems like a banger. What's the price? This new chip, the 9950 x 3d 16 core 3dv cache, 00:14:56.960 --> 00:15:03.760 seemingly monster is 699. It's the same price as last generation, a fair bit more expensive than 00:15:03.760 --> 00:15:09.920 the non 3d cache variant, and a lot more expensive than a 9800 x 3d. So if all you do is game, 00:15:09.920 --> 00:15:15.440 that's still going to be your best gaming option. But if you do other stuff, things that can use 00:15:15.440 --> 00:15:21.600 more cores, if you're running premier, if you're doing, I don't know, crunching pie on your computer, 00:15:22.320 --> 00:15:29.520 this chip, as far as we've seen today, can seemingly do both almost as good as a 9800 x 3d. 00:15:29.520 --> 00:15:35.840 In most of the situations we saw the 9800 x 3d and the 9950 x 3d were pretty much neck and neck. 00:15:35.840 --> 00:15:42.160 And you're not compromising on the productivity to get that gaming prowess. I mean, this thing is a 00:15:42.160 --> 00:15:51.040 is an absolute unit of a processor. And yeah, GG AMD. It's not cheap. But if you use your computer 00:15:51.040 --> 00:15:57.840 for work and getting stuff done and making money, this could be a very, very good option if you also 00:15:58.800 --> 00:16:05.440 And maybe check out the 5600 x 3d ShortCircuit I did a while ago. That was that is still a 00:16:05.440 --> 00:16:08.560 really solid chip. That's pretty cheap. Bye