Digital Foundry Article Technical Discussion [2022]

Status
Not open for further replies.
..but PC's should fare quite a bit better in this regard this generation than last once Direct Storage removes the CPU decompression burden..
How does DirectStorage deliver that? This requires hardware changes. As in hardware is doing the decompression that the CPU is currently doing. And hopefully with less need to move decompressed data to <here> (CPU or GPU) when the actual data in needed in the other memory pool.
 
The 3700X is actually way more powerful than the PS5. The PS5 has one core occupied by the OS, so that leaves 14 threads available for games running at 3.5 GHz, the 3700X certainly boost above 4 GHz all core turbo.

The 3700X has 32 MB (!) of L3 cache instead of 8 MB. And the CPU is driven by DDR memory instead of GDDR, which has much less reduced latency.

The Ryzen 5 3600 is a better comparison for the PS5 imo. But the ultimate comparison is the 4700S, a shame it's only available with limited PCIe bandwidth.
 
How does DirectStorage deliver that? This requires hardware changes. As in hardware is doing the decompression that the CPU is currently doing. And hopefully with less need to move decompressed data to <here> (CPU or GPU) when the actual data in needed in the other memory pool.

Direct Storage will utilise any DX12 (SM6.0) capable GPU to handle the decompression task for GPU targeted data. No hardware changes required. This should significantly reduce the burden on the CPU for a negligible impact on the GPU.
 
Direct Storage will utilise any DX12 (SM6.0) capable GPU to handle the decompression task for GPU targeted data. No hardware changes required. This should significantly reduce the burden on the CPU for a negligible impact on the GPU.
No hardware changes apart form DX12 GPUs and NVMe SSDs. That's change for a lot of PC users.
 
The 3700X is actually way more powerful than the PS5. The PS5 has one core occupied by the OS, so that leaves 14 threads available for games running at 3.5 GHz, the 3700X certainly boost above 4 GHz all core turbo.

The 3700X has 32 MB (!) of L3 cache instead of 8 MB. And the CPU is driven by DDR memory instead of GDDR, which has much less reduced latency.

The Ryzen 5 3600 is a better comparison for the PS5 imo. But the ultimate comparison is the 4700S, a shame it's only available with limited PCIe bandwidth.

I don't hugely disagree, but the OS and other system processes don't come for free on the PC either so the 3700x still has to dedicate some of it's resources to that. It's just not as clear cut as it is with the PS5 and a whole dedicated core.

And although the 3700x can boost to 4.4 Ghz, that's a single core boost and it's base clock is only 100Mz faster than the PS5. If all 8 cores are being pegged out by a game it may not be boosting all that much higher overall, I've read around 4Ghz on a good liquid cooler which is roughly 14% more performance.

Then there's the cache which can clearly offer major benefits for unpredictable PC code, but in consoles where the code is tailored to the available cache of the CPU it's running on, that advantage can be somewhat mitigated. So overall, there's no denying the 3700x is a better, stronger CPU, but I'm not sure the advantages are great enough for it to lose 25% of it's cores and still be considered equivalent to the PS5 CPU. In the case of Spiderman where they're apparently dedicating a whole core to decompression alone, what you're left with is essentially 5 cores on the PC running both the system and game, vs 7 cores on the PS5 dedicated to the game. Even with it's higher clock speed and extra cache is it really any wonder that the 3600x struggles to keep up in that scenario - especially when you add extra API overhead into the mix? That's why I think the 8 cores of the 3700x make for a better comparison. There it's essentially 7 slightly faster cores on the PC for the system + game vs 7 cores on the PS5 dedicated for the game.
 
Last edited:
No hardware changes apart form DX12 GPUs and NVMe SSDs. That's change for a lot of PC users.

I'm really not sure what the relevance of that statement is. I pointed out that Direct Storage will help equalise the CPU overhead between consoles and PC's this generation. Obviously, you actually need a PC capable of taking advantage of Direct Storage for that to be the case, I'm not sure why that would need explicitly stating. For equivalence you also need a decent CPU, probably at least 16GB of RAM, an RT capable GPU for graphics parity, a PSU to turn the thing on etc... but all these things go without saying, and PC gamers are no strangers to having to change (upgrade) their PC's... it's kinda what PC gaming is all about, otherwise we'd all still be using 8086's.

The point here is that the technology not only already exists, but is in fact very widely available in the current PC gaming market. If you have any GPU since Maxwell or above (>94% of the market according to Steam), and an NVMe drive, you can take advantage of Direct Storage. And in fact, we don;t yet know that the NVMe drive is actually required for the GPU based decompression.
 
I don't hugely disagree, but the OS and other system processes don't come for free on the PC either so the 3700x still has to dedicate some of it's resources to that. It's just not as clear cut as it is with the PS5 and a whole dedicated core.

And although the 3700x can boost to 4.4 Ghz, that's a single core boost and it's base clock is only 100Mz faster than the PS5. If all 8 cores are being pegged out by a game it may not be boosting all that much higher overall, I've read around 4Ghz on a good liquid cooler which is roughly 14% more performance.

Then there's the cache which can clearly offer major benefits for unpredictable PC code, but in consoles where the code is tailored to the available cache of the CPU it's running on, that advantage can be somewhat mitigated. So overall, there's no denying the 3700x is a better, stronger CPU, but I'm not sure the advantages are great enough for it to lose 25% of it's cores and still be considered equivalent to the PS5 CPU. In the case of Spiderman where they're apparently dedicating a whole core to decompression alone, what you're left with is essentially 5 cores on the PC running both the system and game, vs 7 cores on the PS5 dedicated to the game. Even with it's higher clock speed and extra cache is it really any wonder that the 3600x struggles to keep up in that scenario - especially when you add extra API overhead into the mix? That's why I think the 8 cores of the 3700x make for a better comparison. There it's essentially 7 slightly faster cores on the PC for the system + game vs 7 cores on the PS5 dedicated for the game.
Good points.

I guess we will see. Personally, I'm really interested to see Miles Morales compared. Miles Morales AFAIK has a Performance RT mode with uncapped framerate and by then, Nixxes will have more CPU optimizations implemented they might carry over to this title. This might give us a better idea how PC CPUs and GPUs compare in a real world scenario with added API overhead to the PS5.

Miles Morales also uses ML for muscle deformation, I am curious wheter that is accelerated on RTX GPUs using tensor cores, could be a nice little optimization from Nixxes part.
 
I'm not actually sure how valid that oft-cited Jaguar argument is. Throughout the last generation PC CPU's had far higher IPC but generally fewer physical cores than the consoles. PC versions of games were therefore re-engineered to depend more on IPC and less on multithreading. It's therefore expected that if you try to run the PC version of the game on a very low IPC processor like a Jaguar, it will run poorly. In the majority of cases though those games will run just fine, and often better than the consoles on a mere dual core PC CPU. Those dual cores could have twice the IPC of the Jaguar and twice the clock speed, for roughly 4x more performance per core, but with 1/4 the number of cores the overall potential performance is roughly the same.

In terms of the PS5, it certainly stands to reason that you'd need a bit more CPU power to cope with the thicker API's and the software based decompression (which I must admit I underestimated the impact of if they're really using a whole core for this relatively modest level of streaming), but PC's should fare quite a bit better in this regard this generation than last once Direct Storage removes the CPU decompression burden, and of course thanks to DX12 which is a much thinner API than DX10/DX11 used throughout most of last gen. It'd certainly be interesting to understand how much performance is needed to make up the difference on the API front though. It'd also be nice to see how a 3700x fares at PS5 settings which is the CPU I consider closest to that in the PS5.
I don't think this has actually materialized in typically better CPU performance in real world results.
 
3700x actually does very well in Spiderman at 1080p, the internal resolution the PS5 is using when activating RT performance mode/60fps. 3700x with a 3060 (non Ti) obtaining close to 100fps with very high settings (above the PS5's), 1080p and DLSS Quality. Now lets not either forget that the PC version doesnt dial down as the PS5 version does, coupled to higher settings aswell as improved RT.
A 3060 is often considered to be the match for the PS5 in NV-land. The game hasnt got all its optimization patches yet as there will be some improvements according to Nixxess. Its a good showcase that hardware roughly equal to PS5 will give you ballpark PS5-performance. API overheads, bad-architecture problems seem overstated, as you get quite good performance day one from a title that is one of the studios first to port over.
Heck, you get better performance probably due to the 3060 being more performant at RT and being able to use ML reconstruction. 3700x is higher clocked with a much larger l3 cache too.

 
Last edited:
I think that depends on developer skill / experience. I imagine Nixxes for example can extract more performance out of DX12 then they can from DX11.
Agreed, I just don't think typical CPU performance has improved over DX11. There were a few outlier DX11 titles that showed excellent CPU performance scaling and coming up on a decade into DX12 usage the same statement can be made.
 
Agreed, I just don't think typical CPU performance has improved over DX11. There were a few outlier DX11 titles that showed excellent CPU performance scaling and coming up on a decade into DX12 usage the same statement can be made.

Yeah it was basically Nixxes with Rise of the Tomb Raider and Shadow of the Tomb raider that showed significantly better CPU scaling in DX12. However taken as the whole with games that provide both renderers, that hasn't necessarily been the case.

The confounding factor though is the vast majority of those games which provide both API's are ones built with the Jaguar CPU's of the PS4/Xbone in mind and as such, aren't likely close to really stressing modern X86CPU's in the first place. Games that do - such as Spiderman - don't have the option to compare, so kind of tough to determine what impact DX12 has.

I think Vulkan has a better track record in providing tangible results, at least over OpenGL - this is especially true for emulators, Cemu and the Switch emus provide dramatically better performance with Vulkan. Of course this comparison has its own problems, as AMD's OpenGL drivers were particularly awful so of course they're going to show a huge advantage. Nvidia performance still shows an appreciable uptick though.

The advantage of a Windows lower level API in general was also more prominently felt in the very early days with dual-core CPU's and Vulkan's 'predecessor' - AMD's Mantle. I had a dual core CPU and it provided markedly improved results in BF4 and Thief when it existed from what I recall.
 
I think Vulkan has a better track record in providing tangible results, at least over OpenGL - this is especially true for emulators, Cemu and the Switch emus provide dramatically better performance with Vulkan. Of course this comparison has its own problems, as AMD's OpenGL drivers were particularly awful so of course they're going to show a huge advantage. Nvidia performance still shows an appreciable uptick though.
The main game I've played that uses OpenGL and Vulkan is GZDoom, and the performance in Vulkan is much higher, but I've also experienced more inconsistent framerates under Vulkan as well.
 
Xess talk at 52 minutes in
My question is how does an intel Xe(not arc) gpu compare to the Xbox Series S and could some one do a dp4a comparison. The intel video said that the dp4a implementation was for older intergrated graphics chip but they seem to be talking more about the Intel's Xe gpus then intel's UHD.
XeSS isn't released yet, so won't get comparison yet. Not until the higher end models of the discreet GPUs are released.

The compatibility version can run on anything that supports dp4a and SM 6.4.
I don't know anything about Intel GPUs but maybe UHD doesn't support dp4a or SM 6.4, or they just happen to not mention it as its far from their focus.
 
The kind of people interested in the things DirectStorage brings already have those things, let's not kid ourselves.
That's nuts. A lot of people who don't have a DX12 GPU and NVMe drive will be interested in I/O improvements that DirectStorage will bring, they just don't have that hardware right now. The majority of PC users aren't updating their hardware on a regular basis, plenty wait for hardware that will bring a significant leap in gaming performance. If you've had a non-M2/NVMe SSD until now, there's been little reason to upgrade it to a NVMe drive.
 
That's nuts. A lot of people who don't have a DX12 GPU and NVMe drive will be interested in I/O improvements that DirectStorage will bring, they just don't have that hardware right now. The majority of PC users aren't updating their hardware on a regular basis, plenty wait for hardware that will bring a significant leap in gaming performance. If you've had a non-M2/NVMe SSD until now, there's been little reason to upgrade it to a NVMe drive.

DX12/SM6.0 encompasses every GPU since the Maxwell launch 8 years ago. According to the Steam Hardware Survey ~94% of users already have one. I doubt there is a significant proportion of users that are still running a Kepler based GPU who are worried about loading times on the latest PS5 ports. They're more likely to be concerned about the game starting at all.

Similarly, if your actual concern is loading times then you're either going to already have an NVMe drive, or be quite ready and willing to get one. No one has suggested that DirectStorage is going to turn a potato PC into a PS5 equivalent IO monster. You're obviously going to need the correct hardware in place to support it, but there is no significant hardware barrier to entry here. The bar in the vast majority of cases will be even lower than a PS5 owner who wants to purchase an external SSD for extra storage. Not a high bar I'm sure we can both agree.
 
Status
Not open for further replies.
Back
Top