Ryzen 9000 X3D at 4K: Full Breakdown & Transcript

[00:05] gains delivered by new hardware launches have been diminishing, there's perhaps one key technological innovation that's been genuinely transformative. 3D Vcashe in AMD's Ryzen lineup of processors. The original Ryzen 7 5800 X3D is one of the

[00:20] all-time greats in my opinion, while the latest Ryzen 9000X3D products continue to deliver massive gaming performance uplifts. This new technology is now featured in Alienware's Area 51 flagship gaming PC.

[00:36] And in this video brought to you by Digital Foundry and Alienware, I want to go in depth on how much X3D chips can make a difference even with 4K gaming, even with the most powerful gaming GPU on the planet. Now, at this point, the

[00:51] status of X3D in the gaming space is beyond question, usually exemplified by 1080p benchmarks, but paired with an RTX 5090 running high-end 4K workloads, including raid tracing and path tracing. Well, to what extent does X3D still make

[01:07] a difference compared to AMD's standard processors is an interesting topic, been wanting to talk about for some time. The balance of a gaming PC, the time. The balance of a gaming PC, the relationship between CPU and GPU. But I

[01:21] don't think it's at all controversial at this point to say that pairing a Ryzen this point to say that pairing a Ryzen 9000X3D chip with RTX 5090 is the default for a top-end gaming PC. Which brings us to the Alienware Area 51.

[01:34] First of all, a brand new chassis with excellent cooling handling both X3D processors and the 600 W monster that is the RTX5090 with a plum, all powered with a 1500 W power supply. The main board is based on the latest X87E

[01:50] chipset with overclocking support available. And our unit came with 64 gigs of 6400 mega transfers per second DDR5. premium design, premium components, and with our specific spec getting both the 9950X3D

[02:05] and the RTX5090, we have the highest end components for both gaming performance and productivity workloads. Looking to just focus on gaming? Well, the 9850X3D option may have half the cores of the 9950X3D and

[02:20] not the best choice for multi-core tasks like video editing, for example, but performance will be mostly the same in the vast majority of gaming scenarios. First of all, just a quick recap of how transformative X3D is viewed through the

[02:35] lens of standard benchmarks before we move on to custom 4K testing. So the general rule of thumb in benching is to isolate the tested component by removing limitations on all others. By running a high-end graphics card at a low

[02:50] resolution, the limit to frame rates is always the CPU, giving the best possible overview of maximum CPU throughput. So, tracking back to our original launch tracking back to our original launch Ryzen 9000X3D benchmarks using an RTX

[03:04] Ryzen 9000X3D benchmarks using an RTX 4090, the 9950X3D is a remarkable 32% faster than the 9950X here in Dragon's Dogma 2, despite having the same core count and configuration. It's that unique vc, an extra 64 mgby of

[03:21] cache memory there, unique to XVD chips, that's making all the difference there. So, we're going to be returning to Boulders Gate 3 later in this video, but here we noted a remarkable 39% increase in performance comparing the 9950X 3D to

[03:36] the 9950X. But it's fair to say that not every game will receive massive improvements. That's still significant though. So, for example, we logged an 18% increase in performance in Cyberpunk 2077. Again, though, we'll return to

[03:50] this one later with some intriguing results from custom testing. It's back to more winning ways for X3D and Far Cry 6, however, with a 36 percentage point increase in frame rates for 9950 X3D over 9950X,

[04:05] while it's a 33 point bump for Hitman World of Assassinations CPU limited benchmark when comparing the same processors. Now, you'll also note that in all of the visualizations I've included here, we're also seeing the

[04:19] included here, we're also seeing the Ryzen 9800 X3D compared to the 9700X. uh the 8 core counterparts to the 16 core behemoths. Again, X3D versus X. architecture, very similar clock speeds. The performance gain comes mostly, if

[04:36] not entirely, from the 3D Vcash. And there's Nion identical uplifts in gaming frame rates from the 8 core chips when compared to our 16 core processor results. It proves the point that CPUs thrive on more cache with X3D chips

[04:52] delivering an extra 64 MGB compared to the standard X models. More cache means less reason for the processor to access main system memory. Often a profound bottleneck for gaming workloads. Great numbers, right? But there are criticisms

[05:06] of this kind of lowresolution benchmarking. We paired all of these CPUs with an Nvidia RTX 4090 running at 1080p resolution. actually lower in many cases as we're using DLSS upscaling. This forces the test PC to focus on CPU

[05:22] performance alone. [music] Uh that's fine for a benchmark, but who in their right mind would actually use an RTX 4090 or a 5090 at 1080p resolution? And here's the thing. I can absolutely understand the sentiment. How many

[05:36] differentials that high when they're gaming at 4K? So, I set out to look into it. And this folded into work we're doing to update our visualization tools. We can now integrate any data from the industry standard present mon

[05:50] benchmarking utility. And there are two data points in particular to discuss. data points in particular to discuss. CPU busy and GPU busy. These millisecond metrics tell us how long it takes for a processor and graphics card to render a

[06:02] frame. Take 1,000 milliseconds, divide it by the CPU busy and GPU busy numbers, output of those components measured in frames per second. And as our first test

[06:14] example, let's take a look at Boulders Gate 3. Go to Act 3, navigate through the city of Boulders Gate, and you'll start to see some extremely heavy CPU limitations. As powerful as the 9950X3D is, you'll see that the RTX 5090 is

[06:30] usually faster. According to the data, we're limited by the CPU 62% of the time and limited by GPU 38% of the time. Across the entire sequence, the CPU averages 150 frames pers while the GPU averages 162. This is at 4K output

[06:48] resolution with DAA in play. In this game, we can't actually give the GPU a higher workload short of boosting internal resolution at the control panel level. So, this is a good example of how even at 4K, you can be CPU limited. So,

[07:02] let's bring in the nonX3D processor equivalent, the Ryzen 9 9950X, no 3D Vcash here means that frame rates are much, much lower, meaning that far more of the RTX5090's performance isn't being tapped into. The

[07:18] 9950X 3D delivers 34% more performance than the 9950X. So, GPU potential than the 9950X. So, GPU potential performance here, 162 FPS. 9950X3D performance here, 162 FPS. 9950X3D delivers 150, but a 9950X in identical

[07:32] content only gives us 112 frames per second average. So, this is a clear example of how even at 4K on max settings with an RTX 5090, X3D conclusively beats the equivalent nonX3D CPU. You'll also note that the 9950X

[07:50] frame times there are even less consistent than the X3DS. Boulders Gate 3, then it's an RPG. It doesn't have pretensions of high-end AAA graphics rendering. But the point is that every single game has its own balance of CPU

[08:03] and GPU requirements. And even with advanced graphics, I still found it relatively easy to find games where CPU is an issue, such as Warhammer 40,000, is an issue, such as Warhammer 40,000, Space Marine 2. 4K output here, maximum

[08:17] settings with RA tracing enabled. Um, specifically RAID traced ambient occlusion. That's the only RT feature in this one. So, here's the thing with Nvidia. Frame rate boosts haven't just happened through more shaders. There's a

[08:30] DLSS as well. Even with a 50/90 DLSS performance mode, especially with DLSS performance mode, especially with DLSS 4.5 is good to use for getting as many frames out of the GPU as possible uh while retaining excellent image quality.

[08:44] So, here's a gameplay slice of Space Marine 2 with the Ryzen 9 9950X, the nonX3D model, where we're plotting GPU busy on the 5090. CPU busy on the 9950X,

[08:57] and the actual output frame rate. As we're CPU limited, output frame rate is pretty much the CPU result with a little extra overhead. With plenty of enemies on the horizon, there's already a sizable gap between GPU and CPU here.

[09:11] But as the hordes get closer, the CPU load increases and there's anything up to a 60 frames pers delta between what the CPU is delivering and what the GPU can theoretically give you. Repeating the same gameplay segment with the 9950

[09:26] X3D sees the delta close up. Average frame rate is around 35% higher with X3D versus without. The lines on the frame rate graph are closing up, but this is gameplay. Is it possible to do a benchmark here? Well, it's actually

[09:40] difficult getting fully matched content, but the cutscene following this action remains CPUheavy. So, let's dig in. Let's take a look. Frame time stutters. They happen on cutscene camera transitions, typical to many games. But

[09:54] the bottom line here is that even in this limited slice, X3D has a 20% improvement to performance over the nonX3D cache processor. Also note that the RTX5090 and 9950X3D are kind of trading blows here. We are much closer

[10:11] to the CPU and the GPU being in balance in this scene. Bottom line though is that we're still significantly CPU limited even with X3D in some areas. the bottom line is that frame rates are as high as they are going to go and it's

[10:27] the CPU that's to blame. So, with that in mind, I increased the DLSS upscaling factor to quality mode. Since the GPU has spare cycles, why not increase image quality instead? Here's the result. The three data points are now much more

[10:42] closely aligned. The 9950X3D is now keeping pace more closely with the RTX5090, but the 9950X still some way behind. There's now a 17% output frame rate uplift for X3D over the nonX3D chip.

[10:59] gap narrows the higher the demand we place on the RTX 5090. So, let's take exercising the GPU to the next level with Cyberpunk 2077. A massively scalable game and at the high end, a hugely demanding GPU workout. Again,

[11:15] we'll start with some gameplay clips here pairing the nonX3D950X with the RTX5090 [music] on the Psycho RT setting DLSS Foundry viewers will know that the Cherry Blossom marketplace here is very

[11:29] heavy on CPU, even more so with ray tracing enabled where the CPU is running two sets of geometry. The world as you see it here along with the BVH structure, the geometry used to trace rays. So, the difference between the RTX

[11:44] 5090's theoretical output and what the 9950X is capable of delivering for us, it's actually a 24 fps difference. Now, carry the same test onto the X3D

[11:57] processor, and the gap between the 9950 X3D and the 5090 narrows, but there is still a gap, right? Uh, this is an example of just how powerful the RTX 5090 is really. But this isn't maxed settings. We can push further. So, let's

[12:11] settings. We can push further. So, let's engage path tracing via Cyberpunk's RT Overdrive setting. One of the greatest PC graphics accomplishments of this generation. With the graphics workload amped up, the 9950X is much more closely

[12:25] aligned with the RTX 590's output, but you'll note that we're still CPU limited. The 5090 still has more to offer. It's being underfed by that 9950X. And although the FPS graph lines may be more closely aligned, pay

[12:41] attention to the jagged frame time graph readings there. A sign that we're still CPU limited, and that's not good. Swap over to the 9950X3D, and we finally have the required CPU performance to keep the RTX 5090 fully

[12:54] fed. Almost complete alignment with the FPS graphs of the three readings there. Frame times on the bottom right graph are improved, but still a little spiky. So, there's the sense that we're perhaps not fully GPU limited. So, with a Psycho

[13:08] not fully GPU limited. So, with a Psycho RT and RT Overdrive tests there, uh, is it possible to consolidate that into a 4K benchmark? Well, yes, we can. For our general CPU testing, the Jerry Blossom Marketplace is one of our custom game

[13:22] mods, uh, where we commandeer the camera through a fixed run through the area, allowing us to get like for-like measurements. So the red line representing 9950X3D more closely hugs the green line which graphs RTX5090's

[13:37] rendering capabilities. Then consider the gap between 5090 and the nonX3D 9950X. The differential is considerably wider. You're losing 12% of performance wider. You're losing 12% of performance around 11 FPS by using the 9950X instead

[13:51] of X3D. Moving on to RT Overdrive, one of the most graphically demanding expected, the groupings are now considerably tighter, just a couple of frames pers between the two processors.

[14:05] limited at all in a path tracing scenario is kind of crazy, but nonetheless, it is measurable. Summarizing then, yes, 1080p benchmarks are fascinating for showing how X3D and nonX3D products compare when the GPU is

[14:21] fully out of the picture, not influencing results whatsoever. But we've shown here how the CPU can still be important in getting as much performance out of a GPU like the RTX5090 as possible, even on the

[14:35] highestend workloads. Now, the scenarios here are cherrypicked, of course, but nonetheless shows the relationship between the two most important processing components in your PC. And to be fair, in the case of Space Marine 2,

[14:48] well, the RTX 5090 is so fast that most of the game is actually CPU limited. So the CPU is super important. It processes game logic. It creates the commands necessary to tell the GPU what to do in race facing games. It's actually

[15:03] creating a secondary internal game world for rays to be traced against. But I richer the game world, the more entities that are in it. Well, the more of a

[15:15] workout it is for the CPU and where X3D makes a key difference. So yes, again, Digital Foundry regulars will know all about Bug Hill in Crimson Desert, which is our final test. 4K output resolution, DLSS performance mode, cinematic

[15:30] settings here, but we're freeing up further GPU resources here by taking the lighting quality down to ultra. [music] And to be honest, you don't really lose much in the way of fidelity, but it allows for top- end frame rates to rise

[15:44] on the graphics side. [music] A more challenging target for our CPUs. Therefore, across the entire sequence, the 9950X3D offers up a 20 FPS advantage over the 9950X. Another example of how it makes sense to pair the fastest

[15:58] graphics card on the market with the fastest CPU. But the key words there are across the entire sequence, an average. The differentials adjust according to context. And as you can see here, the delta between 9950X and 9950X3D can

[16:14] widen dramatically at key points in the sequence. So rationalizing the difference between those two processes down to a single number. It's nice to perhaps not fully indicative of performance on taxing graphics settings

[16:29] during in the- moment gameplay. And that's an important distinction to make as we rerun the test at 4K native resolution with DLAA 4.0 anti-aliasing, considerably increasing the graphics workload on the 5090. Now, I would have

[16:45] expected to achieved balance between CPU and GPU here on both of our CPUs as, you know, the top-end graphics limit drops to accommodate the much higher workload I'm placing on it. But we can still see differences despite the overall

[17:00] grouping. It's a 10 frames pers improvement with X3D across the bench. bunch of frames and different areas of the sequence see the delta shift. So, at

[17:12] the end of all of this testing, my overall takeaway is this. 1080p CPU benchmarking using a massive GPU, it makes sense. I think it's still the best way to isolate CPU performance and to produce a good group test with the

[17:26] closest you're going to get to a good ranking list. It's not often we see a new technology like 3D Vcash deliver startlingly large improvements, but those benchmarks show how potent X3D CPUs are, especially the latest Ryzen

[17:40] 9000 line. But those numbers still translate into palpable wins on higher depending on what you're testing, depending on your settings. Using the Alienware Area 51 specs here, we're able

[17:54] to zero in on one particular configuration, one CPU, one GPU, something you can't really do in a standard CPU review because everyone's PC is different, right? So, you know, in that case, isolating pure CPU

[18:06] performance alone makes sense. But within the confines of this content, it's really interesting to see how the best CPU and the best GPU interact on very very high settings at 4K resolution. But what I've learned here

[18:21] is that we've had generation after generation of big GPU performance bumps. Slowing down now perhaps, but the RTX 5090 is still a beast, right? Gen on genen CPU increases. That's been more of a mixed bag across the years. But Ryzen

[18:35] a mixed bag across the years. But Ryzen 9000X3D processors genuinely awesome and I think the numbers and the experiences shown today bear it out. But that's all thank Alienware for sponsoring our content. The research that went into

[18:48] this video, the brand new performance graphing and visualizations. That's actually a fully from the groundup revamping of our tools. Partly made possible thanks to that support. And I hope you enjoyed the content. But for

[19:00] now, thanks for watching and supporting Digital Foundry.