Current Generation Hardware Speculation with a Technical Spin [post GDC 2020] [XBSX, PS5]

[pjbliverpool]

Yeah I dunno...I am not overly optimistic about RT performance on the consoles now. AMD being quiet about yesterday seemed a bit weird....and that was with hardware that has Infinity Cache...

Well working backwards the RX 6800 has 15% more core resources (based on gaming clock) than the XSX. It also has 50% more fill rate and roughly the same memory bandwidth but with the addition of Infinity Cache. On balance that puts it, at least according to AMD's own testing at around 15-20% faster than a 2080Ti. So it's reasonable to assume the XSX is going to have ballpark 2080Ti performance outside of RT if both are utilised equally (more than expected I must admit) and assuming the XSX has sufficient bandwidth to keep it's core from being bandwidth limited (the 6800 probably has way more than it needs).

Now the 6800XT is about as fast as the 3080 which is in turn about 30% faster than the 2080Ti but according to the 2 information points we have so far, may be more like 2080Ti performance in RT.

So that would put the XSX at somewhere between 2070 and 2070S in RT performance and the PS5, assuming it's RT performance scales comparably to the XSX, at around 2060 level. That's assuming the RX6xxx cards aren't getting a disproportional RT performance improvement from the Infinity cache.

So not game changing, but still pretty reasonable.

disclaimer: I'm well aware that many factors could throw the above comparison off significantly.

 

[thicc_gaf]

Well working backwards the RX 6800 has 15% more core resources (based on gaming clock) than the XSX. It also has 50% more fill rate and roughly the same memory bandwidth but with the addition of Infinity Cache. On balance that puts it, at least according to AMD's own testing at around 15-20% faster than a 2080Ti. So it's reasonable to assume the XSX is going to have ballpark 2080Ti performance outside of RT if both are utilised equally (more than expected I must admit) and assuming the XSX has sufficient bandwidth to keep it's core from being bandwidth limited (the 6800 probably has way more than it needs).

Now the 6800XT is about as fast as the 3080 which is in turn about 30% faster than the 2080Ti but according to the 2 information points we have so far, may be more like 2080Ti performance in RT.

So that would put the XSX at somewhere between 2070 and 2070S in RT performance and the PS5, assuming it's RT performance scales comparably to the XSX, at around 2060 level. That's assuming the RX6xxx cards aren't getting a disproportional RT performance improvement from the Infinity cache.

So not game changing, but still pretty reasonable.

disclaimer: I'm well aware that many factors could throw the above comparison off significantly.

This falls in line with what Alex at DF said when he analyzed the Spiderman footage a few weeks back. Honestly? If that is just "2060" levels of RT, I'm a happy gamer. Because for good RT, you need not just the technical power but an artistic eye for utilizing it with flair. Insomniac are showing that off extremely well IMHO, with what is "just" 2060-tier RT performance.

Bodes well for RT on both systems next-gen on the 1P front IMHO. I'm still really impressed with the RT shown in other games too like Bright Memory Infinite and Exo-Mecha, and I hope that level of RT fidelity is present with the final versions once they're ready to release.

 

[dten]

The infinity cache is interesting. AMD chose 128MB regardless of the card so they are betting on this.
It's a huge gambit on AMD's part. Anandtech estimates 128MB is 6 billion transistors. It feels like one of those things that could be a game changer but could also be scrapped in future architectures if AMD decides the trade offs are too much, like HBM.

 

[Kugai Calo]

The infinity cache is interesting. AMD chose 128MB regardless of the card so they are betting on this.
It's a huge gambit on AMD's part. Anandtech estimates 128MB is 6 billion transistors. It feels like one of those things that could be a game changer but could also be scrapped in future architectures if AMD decides the trade offs are too much, like HBM.

6 billion!?

 

[AlNom]

6 billion!?

128*1024*1024 bytes*8bits/byte * 6T SRAM = 6.4 billion transistors

 

[Kugai Calo]

128*1024*1024 bytes*8bits/byte * 6T SRAM = 6.4 billion transistors

Ah, forgot the 8 bits/byte multiplier. And that's excluding the control logic! When you put it into perspective like that, it really shows how prohibitively expensive SRAM cache is!

 

[BRiT]

Ah, forgot the 8 bits/byte multiplier. And that's excluding the control logic! When you put it into perspective like that, it really shows how prohibitively expensive SRAM cache is!

Cache is $$$$.

 

[function]

I find this interesting, and I'll be very curious to see if MS chooses to reveal what specific RDNA2 features they chose to wait for that were apparently not included in PS5. Is this just PR speak, or is it more than that? Was it for the sole purpose of making sure PC/XB have uniform DX features? The specific feature(s) MS decided to wait for did apparently come at the cost. That cost being delayed dev kits, which leads to immature dev tools, which is far from ideal. So I have to assume their decision to wait for the full RDNA2 feature set was for a good reason.

They do seem to be hinting at a number of things, and Sony seem to be strangely quiet about (particularly) VRS. SFS and int4 / int8 are specific to MS, but Sony could still have some variation on Sampler Feedback as that's in 'vanilla' RDNA 2. And I suppose MS have to be a bit careful about mentioning Sony directly by name. I guess time will reveal the particulars...

It's safe to say that MS have been planning for DX12U for years and that it was always the plan for XSX to have it too. I think they always knew they were going to be close to the wire with XSX | XSS, but probably banked on DX12 PCs being a reasonable substitute platform in the mean time (even without DX12U). I think they would probably have liked to not be quite this close to the wire though. Perhaps RDNA2 slipped a little.

Yup. There's no single right or wrong decision here. If it is about timing and you want to ensure consistently between console and PC then you want to fork later and accept this gives you less time to iterate/improve your custom changes. If you don't care about consistency, or about feature sets beyond point X then you fork earlier and get spend more time in custom changes and experimenting/iterating those.

It's interesting to think about Nintendo. They use almost 'withered' hardware at times, but it means they have a long time with almost final hardware before launch (Switch was what, nearly 2 years for the off the shelf chip they used?). Nintendo seem to produce amazing launch window games that see them out of the gate really strong, where as MS always seem to push it right to the edge. XSX and 360 seem to have been a bit of a scramble and even X1 didn't have a finalised, low resource version of DX11 ready at launch iirc.

Perhaps there's a sensible middle road. Could it be that Sony are benefiting from taking it?

 

[function]

There's also the possibility of some of Microsoft's customizations becoming part of a discrete GPU design. Cerny pointed to that collaboration that Sony had with AMD and how a few minor PS4 tweaks showed up in one GPU. It's likely Sony isn't the only party that this can happen for. Of course, if one console's feature shows up in a discrete product, it could be interpreted that the discrete product + vendor-specific tweak is the "full" architecture.

It would be really interesting if Navi 21 and Navi 22 end up having some minor differences that reflected XSX and PS5. Although could such differences ended up hidden by the drivers?

That possibility aside, there's a few discrepancies that have been mentioned in the past that could be other sources of a timing discrepancy or a choice to skip a feature. The inference-focused instructions could be something Microsoft waited for, although in fairness specific formats like that tend to be inconsistent among AMD GPUs.
Sony's geometry front end had some explicitly different naming, and primitive shaders were named for Sony rather than the DX12 mesh shaders. That could be the result of timing, since Mesh shaders weren't an AMD initiative and could have led to AMD's primitive shader functionality being replaced or modified. Sony may have committed prior to that transition, or decided the transition was uncompelling versus what it already had.

This reminds me of some discussion a while back about Sony patents on geometry processing from Cerny himself IIRC.

There were some Sony patents that described dividing screen space up into areas to be rendered at different resolutions to accelerate FOVeated rendering. I recall this being done at the geometry processor level rather than the pixel shader level like VRS. I could see a more VR focused Sony perhaps wanting to spend time customising an offshoot of Geometry Engines to support this, rather than waiting to take a newer geometry processor whose changes they didn't value as highly.

Mesh shaders seem like a fantastic concept to me, but then again I have no idea what Sony's capabilities are, and what they might have given up but gained instead.

 

[Johnny Awesome]

I think what we are seeing is that MS are pushing the envelope on features more than Sony, but at the expense of not being as ready from a GDK & software standpoint. MS are counting on BC to carry them with early adopters this year. They're probably fine as long as they get some showcase titles out for display at E3 next year and ship some of them before Black Friday 2021.

 

[RDGoodla]

We have known a lot of info about resolution/frame rate of games in the launch window. It's very interesting that all of these games have the same
resolution and frame rate between Xsx and PS5 versions. At least nominally.

Previously it indicates that actual performance may differ only 5~15%. Do we have further info about these launch games (or any leak)?

 

[Nesh]

https://twitter.com/i/web/status/1321555899811565568

So this kind of confirms no VRS and some other RNDA 2 features on PS5?

 

[Jay]

So this kind of confirms no VRS and some other RNDA 2 features on PS5?

No it doesn't confirm that at all.
It could have a variation that Sony felt worked better as a whole with their VR ambitions.

We don't know what it actually means, hopefully we'll be getting more details today via Major Nelsons podcast with Mr Beard Guy.

Up until then all we can do is speculate.

 

[cheapchips]

We have known a lot of info about resolution/frame rate of games in the launch window. It's very interesting that all of these games have the same
resolution and frame rate between Xsx and PS5 versions. At least nominally.

Previously it indicates that actual performance may differ only 5~15%. Do we have further info about these launch games (or any leak)?

It'll be interesting to revisit the deltas between launch titles and actual next gen titles in a few years.

 

[Deleted member 86764]

I think what we are seeing is that MS are pushing the envelope on features more than Sony, but at the expense of not being as ready from a GDK & software standpoint. MS are counting on BC to carry them with early adopters this year. They're probably fine as long as they get some showcase titles out for display at E3 next year and ship some of them before Black Friday 2021.

I think we have to be careful saying Sony have less features. It's entirely probable they have non-standard hardware modifications that wouldn't make logical sense on DX12 or for a PC. We can see that already with how they've setup their IO system. They haven't named it something like "Velocity Engine" and created a promotional video for it. And that's despite having a significantly better IO system.

So while PS5 may very well not have "full RDNA2" features. There's a very high chance that they have modified their hardware in some way that are better than standard RDNA2 and some that may be less.

I would be reluctant to take an Xbox "marketer" (as stated on that guy's profile) at complete face value. Although I'm sure he relishes in people doing it.

The proof is in the eating and if we go by what's currently stated for all promotional material for all next-gen games to date; they're both apparently at the same resolution / framerate. I derive that this means that the PS5 drops variable resolution more often, or Sony have performed some unexpected miracle with the hardware. Although, I'm inclined towards the former.

 

[chris1515]

I think we have to be careful saying Sony have less features. It's entirely probable they have non-standard hardware modifications that wouldn't make logical sense on DX12 or for a PC. We can see that already with how they've setup their IO system. They haven't named it something like "Velocity Engine" and created a promotional video for it. And that's despite having a significantly better IO system.

So while PS5 may very well not have "full RDNA2" features. There's a very high chance that they have modified their hardware in some way that are better than standard RDNA2 and some that may be less.

I would be reluctant to take an Xbox "marketer" (as stated on that guy's profile) at complete face value. Although I'm sure he relishes at people doing it.

The proof is in the eating and if we go by what's currently stated for all promotional material for all next-gen games to date; they're both apparently at the same resolution / framerate. I derive that this means that the PS5 drops a variable resolution more often, or Sony have performed some unexpected miracle with the hardware. Although, I'm inclined towards the former.

Or like Digitalfoundry told about platform comparison, there will be surprise.

 

[Deleted member 86764]

Or like Digitialfoundry told about platform comparison, there will be surprise.

Literally just finished watching that video just now. That does seem to imply that the PS5 may be performing unexpectedly well.

I loved Richard's joke about not wanting to power his dreams. Fell a bit flat with John though. Ha.

 

[function]

The wording threw me a bit. Didn't equate dual stream to dual command processor.

Yeah the wording is a bit different this time round. Maybe they've gone from one processor per stream to something more complex (3+ processors? Something more like threads on a CPU with load management? I dunno!).

 

[Jay]

Literally just finished watching that video just now. That does seem to imply that the PS5 may be performing unexpectedly well.

I thought it could be a few different things.
Like xbox frame rate doubling with maybe increase resolution.
XSS
PS5 BC system features
PS5 Loading times, simular quick resume feature

I personally didn't think there was enough to say about PS5 performaning particularly unexpectedly well.
General BC it should be able to max out DR and FPS, maybe games with unlocked fps performing well due to possibily higher fixed resolution compared to XO.
New games, didn't sound like was talking about that, but that could be my interpretation of how it was said. The wording platform made me think this.

 

[Deleted member 86764]

I thought it could be a few different things.
Like xbox frame rate doubling with maybe increase resolution.
XSS
PS5 BC system features
PS5 Loading times, simular quick resume feature

I personally didn't think there was enough to say about PS5 performaning particularly unexpectedly well.
General BC it should be able to max out DR and FPS, maybe games with unlocked fps performing well due to possibily higher fixed resolution compared to XO.
New games, didn't sound like was talking about that, but that could be my interpretation of how it was said.

Yeah, you're right. We can only speculate on what they meant. The PS5 should be slower.

I can imagine the BC of PS5 working well, since the hardware is so similar only at a much higher frequency.