Digital Foundry Article Technical Discussion [2022]

[PSman1700]

TLDR: They get 6 to 12 GB/s per 1 TF GPU Decompression. A mere 2 TF will outperform the PS5.

Post in thread 'Blazing Fast NVMEs and Direct Storage API for PCs *spawn*' https://forum.beyond3d.com/threads/...-storage-api-for-pcs-spawn.61761/post-2158006

That would be on top of the raw 7gb/s (up to 13gb/s for some drives) i suppose. If youve got such a drive, you wouldnt need even 1TF to outpace the PS5's nvme in read speeds.

 

[PSman1700]

Another thing is the Xbox already has IO and Decompression hardware. It just doesn't have the Kraken scheme of decompression.

It's really quite something to see a single post contain such completely wrong statements. It makes me believe they are deliberately doing so in bad faith only to prop up their little plastic box called PlayStation in relation to PC and Xbox.

The Xbox's equalivent to Kraken would be BCpack. Both PS5 and Xbox series consoles are equipped with hardware decompression blocks.

Xbox Series X Can Reportedly Decompress Textures Better Than The PlayStation 5

A former Valve employee believes that the Xbox Series X has a better texture decompression technique than the PS5. This decompression technique is called

segmentnext.com

'A former Valve employee believes that the Xbox Series X has a better texture decompression technique than the PS5. This decompression technique is called ‘BCPack’ which will rival the PlayStation 5’s ‘Kraken’ technique.'

 

[Remij]

I'll echo what I said in my initial write-up: It's overall good, but ultimately while the quality of life nods in terms of UI and features are great to see, they still don't necessarily trump final performance for me. As is stands now, you basically need the fastest x86 on the planet to have a hope of maintaining above 60fps with RT, and probably still wouldn't reach the framerate that the PS5 version gets when running with VRR in many situations.

Granted you can do this at a higher res with a top-end GPU (and have a selection that will be much more so in a few months), but I think this is the first time we've seen a release where top-end PC hardware doesn't greatly outpace the native console version. With the poor frame pacing on lower-end GPU's (and that means anything below a 2060 Super), this is a pretty restrictive title in terms of PC hardware that can just match a console experience, if not beat it. This is not a dig at Nixxes mind you, they may be doing the best they can - but if so, that does speak to either some significant inefficiencies in the PC's architecture or the potential difficulty of ports that are more 'native' to the next gen consoles going forward - most of the ports that have brought over until this have largely been PS4 versions with PS5 patches.

I think it's important to keep in mind that these ports are done on a limited budget. They produced this port in under 1 year.. coming into an entirely new codebase, of a game that is highly tuned for the specific console architectures its on... while also coming right on the heels the company merging with Sony. They don't have time to completely rewrite everything from scratch. It had to have been an incredible busy time period.. and like they alluded to in Alex's interview, they were doing the best they could to collaborate with Insomniac while being on completely different time schedules across the world.

They've likely been developing Miles Morales during most of that time in parallel as well, so the learnings should carry forward. They'll be more familiar with the codebase in general now, and if we assume that they will collaborate more in the future for PS5 games, they will likely be able to make more informed design decisions during the console game development which makes things more efficient on the PC side.

 

[Remij]

That would be on top of the raw 7gb/s (up to 13gb/s for some drives) i suppose. If youve got such a drive, you wouldnt need even 1TF to outpace the PS5's nvme in read speeds.

More like games (and that goes for Sony as well) aren't streaming in 7+GB/s during gameplay anyway. Not even close. So the concern of it affecting performance while gaming is grossly misguided. As Brit already stated, that work can be done asynchronously to help keep the GPU actually fed.

It's going to have a negligible impact on performance. Much less than if it were to be done on the CPU.

 

[Remij]

The Xbox's equalivent to Kraken would be BCpack. Both PS5 and Xbox series consoles are equipped with hardware decompression blocks.

Xbox Series X Can Reportedly Decompress Textures Better Than The PlayStation 5

A former Valve employee believes that the Xbox Series X has a better texture decompression technique than the PS5. This decompression technique is called

segmentnext.com

'A former Valve employee believes that the Xbox Series X has a better texture decompression technique than the PS5. This decompression technique is called ‘BCPack’ which will rival the PlayStation 5’s ‘Kraken’ technique.'

I believe Rich said that before he knew anything about Oodle Texture compression.

Not sure where it all lands nowadays though. Rich is always talking about UFOs on twitter and since RAD was bought by Epic you don't hear much anymore.

 

[PSman1700]

I believe Rich said that before he knew anything about Oodle Texture compression.

Not sure where it all lands nowadays though. Rich is always talking about UFOs on twitter and since RAD was bought by Epic you don't hear much anymore.

Time will tell, but my guess is that it wont matter much if at all. Both encompass the same ideas just approaching them differently.

 

[Remij]

Time will tell, but my guess is that it wont matter much if at all. Both encompass the same ideas just approaching them differently.

Well, it basically has more to do with reducing the size of the game package on disc and less with regards to upfront loading times. PS5 still wins for the most part.

 

[PSman1700]

Well, it basically has more to do with reducing the size of the game package on disc and less with regards to upfront loading times. PS5 still wins for the most part.

For the Xbox MS hasnt really started utelizing the nvme all too much as of yet. Its hard to call a winner in any department, if theres going to be a 'winner' anyway.

 

[pjbliverpool]

From the previous generation atleast it was generally said that UMA and split memory pools had each its own dis- and advantages. I remember AF being one of them, and memory contention ofcourse. UMA is easier to develop for. This has always been true as far back as the OG Xbox (which had a single memory pool too).

Yes, my expectation is that provided you avoid the pitfalls of using the wrong memory pool for a given operation you're going to have more raw performance available with a split memory pool than you would with a UMA, even accounting for the extra copies that are going to be required.

So basically the total load time is about what it takes for the RT BVH to be built, and without that step it's around half a second quicker. Not much that can be done to further reduce that, unless they made an option in the game's menu to compile all shaders ahead of time to reduce loading times.

As far as I understand it, the shader compilation and BHV creation are different activities. So there's potential to improve both. The game menu shader compilation option should be able to remove the shader compilation delay, but for the BHV creation it's not so clear. Nixxes mentioned caching them in the interview but didn't implement it for reasons I wasn't clear about.

While it's not perfect without RT, the biggest CPU hit by comparison to the PS5 is with ray tracing, so I'm not sure how much of the CPU limitation most are complaining about is really due to texture decompression as a main factor.

If they're really dedicating a whole core to it then on something like a 3600x that's a fairly big percentage of overall CPU performance. Possibly enough to take it from the current high 40's lows to something more like a locked 60fps.

I'll echo what I said in my initial write-up: It's overall good, but ultimately while the quality of life nods in terms of UI and features are great to see, they still don't necessarily trump final performance for me. As is stands now, you basically need the fastest x86 on the planet to have a hope of maintaining above 60fps with RT, and probably still wouldn't reach the framerate that the PS5 version gets when running with VRR in many situations.

Granted you can do this at a higher res with a top-end GPU (and have a selection that will be much more so in a few months), but I think this is the first time we've seen a release where top-end PC hardware doesn't greatly outpace the native console version. With the poor frame pacing on lower-end GPU's (and that means anything below a 2060 Super), this is a pretty restrictive title in terms of PC hardware that can just match a console experience, if not beat it. This is not a dig at Nixxes mind you, they may be doing the best they can - but if so, that does speak to either some significant inefficiencies in the PC's architecture or the potential difficulty of ports that are more 'native' to the next gen consoles going forward - most of the ports that have brought over until this have largely been PS4 versions with PS5 patches.

I think we need to remember here about the object distance setting which has a huge impact of RT performance. Most sites presumably test with this set to 10 where-as thanks to @Dictator we know the PS5 runs between 7 and 8. As per @Dictators video on DF we know that a 12900k can handle 90+ fps while swinging around the Times Square area. So I don't think you do need the fastest x86 to maintain 60fps, it seems the fastest x86 can easily exceed this. But what we have seen at the appropriate settings is that an x86 that should be roughly equal to the PS5 CPU (the 3600x) is not achieving the same level of performance. But as noted earlier that sounds like it's down to CPU decompression and API overhead - and possibly a less efficient BHV update than the PS5 at present.

More like games (and that goes for Sony as well) aren't streaming in 7+GB/s during gameplay anyway. Not even close. So the concern of it affecting performance while gaming is grossly misguided. As Brit already stated, that work can be done asynchronously to help keep the GPU actually fed.

Yes I agree, I don't think SSD speed is the limiting factor here. Any PCIe 4.0 NVMe should be plenty sufficient to match or exceed the PS5 in this regard. The bottleneck is still CPU performance where it seems the PC still has a few areas where it will have to work harder than the consoles for the same result (API overhead, decompression, shader compilation, BHV creation)

 

[Riddlewire]

For the Xbox MS hasnt really started utelizing the nvme all too much as of yet. Its hard to call a winner in any department, if theres going to be a 'winner' anyway.

Quick Resume has gotten plenty of usage and, if you discount the first load, has given Xbox loading time wins against PS5 in most cases. Resume from rest is also almost always faster on Xbox.
I know you're talking about in-game application, but quick resume has been a pretty nice feature.

 

[Remij]

As far as I understand it, the shader compilation and BHV creation are different activities. So there's potential to improve both. The game menu shader compilation option should be able to remove the shader compilation delay, but for the BHV creation it's not so clear. Nixxes mentioned caching them in the interview but didn't implement it for reasons I wasn't clear about.

Sorry, I guess with the way my post was worded it wasn't clear, but I'm very well aware of the fact that BVH creation and shader compilation are completely different things. I meant outside of the BVH creation taking as long as it's going to take.. there's not much they can do to further reduce the loading time currently... without having an option to compile all shaders beforehand.

Basically what I'm saying is that they've done a great job at ensuring that everything is as compact as it can be. With RT, the game doesn't take that much longer to load, considering the extra processing it has to do.

 

[Remij]

Quick Resume has gotten plenty of usage and, if you discount the first load, has given Xbox loading time wins against PS5 in most cases. Resume from rest is also almost always faster on Xbox.
I know you're talking about in-game application, but quick resume has been a pretty nice feature.

It's an amazing feature. I just really wish we had hard data on what the actual console is pushing through.

Where the hell are all the GDC talks about utilizing next-gen storage and memory!

 

[PSman1700]

Quick Resume has gotten plenty of usage and, if you discount the first load, has given Xbox loading time wins against PS5 in most cases. Resume from rest is also almost always faster on Xbox.
I know you're talking about in-game application, but quick resume has been a pretty nice feature.

Yeah right, true. Probably the only feature so far i'd want over on the PC, and the feature that actually matters the most.

 

[Flappy Pannus]

I think we need to remember here about the object distance setting which has a huge impact of RT performance. Most sites presumably test with this set to 10 where-as thanks to @Dictator we know the PS5 runs between 7 and 8. As per @Dictators video on DF we know that a 12900k can handle 90+ fps while swinging around the Times Square area.

Well yes, albeit the big differentiator with the 12900k is likely that DDR5 though as we've seen from benchmarks, DDR5 is still relatively esoteric for rigs currently.

 

[pjbliverpool]

Well yes, albeit the big differentiator with the 12900k is likely that DDR5 though as we've seen from benchmarks, DDR5 is still relatively esoteric for rigs currently.

I think you may be right but it'd be great to see some controlled benchmarks around this.

 

[Flappy Pannus]

I think it's important to keep in mind that these ports are done on a limited budget. They produced this port in under 1 year.. coming into an entirely new codebase, of a game that is highly tuned for the specific console architectures its on... while also coming right on the heels the company merging with Sony. They don't have time to completely rewrite everything from scratch. It had to have been an incredible busy time period.. and like they alluded to in Alex's interview, they were doing the best they could to collaborate with Insomniac while being on completely different time schedules across the world.

Oh sure like I said I don't intend to dump on Nixxes and it's not a 'bad' port by any stretch, it just has a more constricted range of hardware that can run it 'well' than what we normally witness - currently. It has been only 2 weeks since release and 2 patches so far after all, so like I said - we'll see.

They've likely been developing Miles Morales during most of that time in parallel as well, so the learnings should carry forward. They'll be more familiar with the codebase in general now, and if we assume that they will collaborate more in the future for PS5 games, they will likely be able to make more informed design decisions during the console game development which makes things more efficient on the PC side.

Yeah I thought about that and if they take Insomiac's lead like they did with Miles Morales, they could potentially back-port any engine improvements back to SM:RM. Insomniac kept up with a steady stream of rather significant patches for months since MM initial release that massively improved RT quality, added the 60fps RT mode, and consistently tweaked performance - MM would usually get them first, them they would release a patch for Petey.

 

[snc]

Digital Foundry: I think you as a team have proven that you are good at getting a foreign codebase and making magic out of it. So, you mention getting ray tracing on PC and the last push before launch was offering more settings to adjust the load on the CPU with the BVH ray tracing object range. Trying it on the 12900K or recent Alder Lake CPUs, the game flies, but then on a Ryzen 3600, what I consider a mid-range CPU very similar to the PS5: Zen 2, 6 cores/12 threads available for game usage, so on and so forth. And on that CPU at the highest settings it will be drop below 60fps while moving through the city and swinging as it is CPU limited. It has lurches down to the upper 40s. So what exactly is the CPU limitation bottleneck? What work do you want to do there in the future?


Jurjen Katzman: There are a few things going on there that are interesting and we're actually still making some changes to the code right now. So as I think you know, from your early analysis, to achieve 60fps on PS5 with ray tracing it makes other compromises [even beyond ray tracing]. So it turns down crowd density for example, or there's fewer cars around. And so that compensates for some of those CPU things, and we didn't make that very easy for the user to do in the [early review] build you played. So we are actually offering up some more options to make that to allow that to be better balanced [in the retail build].


And in general, the game originally came from the PS4 right? The PS4 CPU cores were not so stellar and the PS5 and the PCs were far more powerful. With the PS5, that gap has certainly gotten smaller. And there's still quite a few things on the PC where there's more overhead, like the APIs have more overhead, we don't have the decompressor for example, we don't have hardware doing decompression for us as we're streaming in content - that gets left to the CPU. So we certainly have more CPU challenges to go around even when we're doing the same things. And then if we don't dial down things that are dialled down on the console, we now have even more work to do on the CPU.


So okay, so if you have a PS5 game that fully loads all the CPU cores, then yeah, PC CPUs that don't have the same core count, for example, or the same processing power, they'll be in a tricky spot, right? And they will have to rely on lower settings of scalability, as well. But I think that's important about PC, right, that we do have that scalability, we do offer all those options. And you can run it in a way that works well for your system, no matter what.


Michiel Rosa:It's even worse for us because we also have the added overhead of the abstraction layer to DX12 and the DXR abstraction layer, which is obviously very lean on the Sony side. So even if you have a more powerful CPU than on the PlayStation 5, you might still end up with a lower frame-rate.

Thats a good question and nice explanation, its nothing surpsise for me as so many examples that you need more powerful cpu on pc than ps5 have (btw it was also true for last generation, on jaugar level cpu pc gaming would be disaster).

 

[PSman1700]

Thats a good question and nice explanation, its nothing surpsise for me as so many examples that you need more powerful cpu on pc than ps5 have (btw it was also true for last generation, on jaugar level cpu pc gaming would be disaster).

At the same time they aknowlegde things arent dialed down as much in the pc version, the PS5 version is making compromises far beyond just ray tracing to obtain 60fps, with a less aggressive DRS mode to boot. He also explains that if a game 'fully loads all cores' on the PS5, you might run into trouble on a fewer core count CPU like the 3600 with its 6 cores. The game doesnt really scale to 8 cores, thats more of a optimization issue i suppose. Also, this is a rather quick port (under eleven months) and one of their first from PS5 to PC, before further optimizations and gpu decompression.
3700x, which is closest to PS5, albeit clocked almost a Ghz higher, does spiderman just as good if not better considering the same settings.

If you'd be on a jaguar CPU last generation, you'd be limited to console framerates, which yea, is a disaster for most pc gamers. Just like if you'd be on a 2019 zen2 3700x thats heavily downclocked, would pose limitations to more capable GPUs and UE5-engined titles.

 

[Shifty Geezer]

Well, it basically has more to do with reducing the size of the game package on disc and less with regards to upfront loading times. PS5 still wins for the most part.

If you want to pit two Olympic marathon runners against each other to see who's fastest, there can be only one, and a 74 second advantage is a 74 second advantage, 1% faster. If you want to get a message from Marathon to Athens, 74 seconds is meaningless as to your choice of runner or consideration of the outcome.

I'd prefer it if, unless there's specific reason to identify a fastest performer in a specific benchmark, we looked more at the overall tiers of performance, whether a platform/system is playing in the same league for it not to be a consideration for developers. That can be a margin of 2x faster/slower maybe in some workloads. Our interests are only in the techniques required to make our pretty visuals and the algorithms and hardware devs can tap into. Scaling up and down within those algorithms is a specific type of discussion better suited to Top Trumps cards than this tech forum IMO.

 

[pjbliverpool]

Thats a good question and nice explanation, its nothing surpsise for me as so many examples that you need more powerful cpu on pc than ps5 have (btw it was also true for last generation, on jaugar level cpu pc gaming would be disaster).

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.