Digital Foundry Microsoft Xbox Scorpio Reveal [2017: 04-06, 04-11, 04-15, 04-16]

The digitalfoundry article is from 2 October 2016 and Mark Cerny Interview about PS4 Pro was published 20th October 2016.

Edit: And it was the first time custom features of PS4 Pro GPU were revealed to the public...
 
Last edited:
http://www.gamasutra.com/view/news/283611/Inside_the_PlayStation_4_Pro_with_Mark_Cerny.php

And from Cerny Interview it seems it is the most important post Polaris feature of PS4 Pro GPU...

But Cerny seemed more excited about another one of these “post-Polaris” features that the PS4 Pro has: a significant improvement in the way it handles 16-bit variables like half-floats.

With the PS4 Pro, said Cerny, “it's possible to perform two 16-bit operations at the same time, instead of one 32-bit operation. In other words, with full floats, PS4 Pro has 4.2 teraflops of computational power. With half floats, it now has double that -- which is to say, 8.4 teraflops of computational power. As I'm sure you understand, this has the potential to radically increase the performance of games.”
 
http://www.gamasutra.com/view/news/283611/Inside_the_PlayStation_4_Pro_with_Mark_Cerny.php

And from Cerny Interview it seems it is the most important post Polaris feature of PS4 Pro GPU...
Well you can't blame him for preaching that PR line just like MS wanting everybody to think that ESRAM/EDRAM was the best thing since sliced bread...the impact of double-rate FP16 is going to be minuscule compared to having more ram, more CUs, or ROPS or Bandwidth.. etc..It's a nice thing to have for sure but this ain't going to light the world on fire.
 
Well you can't blame him for preaching that PR line just like MS wanting everybody to think that ESRAM/EDRAM was the best thing since sliced bread...the impact of double-rate FP16 is going to be minuscule compared to having more ram, more CUs, or ROPS or Bandwidth.. etc..It's a nice thing to have for sure but this ain't going to light the world on fire.

I have no idea I don't have the devkit doing any test. It seems it was the biggest element for Mantis burning racing team to achieve a 4k native resolution from 1080p 2xMSAA. They wouldn't talk about it breaking their NDA to Richard if it was not so important for their title...

Out of this it is wait and see...

And it was a launch PS4 Pro title, there is much more to do with it... It will be interesting to see what the custom version of Decima engine use for Death Stranding will do with the GPU...
 
Last edited by a moderator:
Yeah it will be more telling once we see how Death Stranding looks on both, we'll have to wait for one of those DF side-by-side videos I guess.

It will be interesting to compare PS4 version and PS4 Pro version of the game... ;)

Exactly what they have done with Mantis Burn Racing.

Out of comparing PS4 Pro and Scorpio, the comparison between each machine of the same family is interesting too...

Edit: PS4 against PS4 Pro and Xbox One, Xbox One S and Scorpio
 
Last edited:
"For the very small handful of titles that run at 720p today, our expectation is that they can checkerboard up to native 4K if they want to do that. I also expect variations of titles that are perhaps running at 900p at 30fps on Xbox One today that they can leverage the 31 per cent boost to CPU clock along with a bunch of other optimisations in conjunction with our D3D12 offload to potentially offer 1080p60 rather than 900p30. It's totally up to developers."

Now we talkin'.
 
it sounds like their own version of a way to help checker boarding in hardware. We don't know what that is yet, and if it could even be pretty standard new gcn tech.

it seems like their saying they have streamlined the whole apu around current engines running on x1.
x1 = esram needed to be used and optimised to get reasonable performance
ps4 = streamlined and easy to get performance.

Scorpio = streamlined to run x1 code with 4k assets, improved frame rate / resolution
4pro = rpm and checkerboarding to possibly make best use of hardware.

interesting how it seems to be turning out.
hopefully Scorpio has some interesting extensions, simply because i find things like that e.g. id buffer more fascinating.

Scorpio not having RPM seems to have caught everyone including myself by surprise. how much, and how often it can be used is yet to be seen.
maybe including it would've meant sacrificing general performance? Or simply, ill judged thinking that it would need to be specifically coded for, therefore not worth adding.

this is the most information we've had about a system before it's release, so I'm glad, although really would like to hear about those customizations.
I'm not expecting DF to give any more details about soc now though.
maybe MS is waiting for after release, when games can also be seen running.
 
Reading through it seems they valued clock speed as the best way to improve performance, because it improves the entire system performance vs just improving performance in a single aspect (eg CU). The comment about "raising all boats" kind of makes sense. Maybe they figured 6 tflops at a given clock speed gets them their performance mark without double-rate fp16, so it wasn't needed. There must be some design tradeoff to including it, like space, price, compatibility. Overall, if you do the math ps4 pro would have to do 2/3 of it's shader ops as fp16 to hit 6 teraflops. I imagine it'll become more and more useful as devs find suitable workloads, but I think in terms of general performance clock speed still probably clearly wins. Obviously both would be better.
 
If MS have really nailed throughput for the GPU with their "60+ simulation crafted customisations" then that could could potentially be of much greater benefit than fp16, with the huge perk of needing little to no additional work and not requiring the maintenance of two code branches.

Minimising additional work required for big Scorpio gains does seem to be one of the overriding goals of the hardware design team.

They're also very clearly claiming CPU optimisations for Scorpio beyond just running at a higher clocks. They're definitely talking about changes that have improved IPC as well as clock speed.
 
If MS have really nailed throughput for the GPU with their "60+ simulation crafted customisations" then that could could potentially be of much greater benefit than fp16, with the huge perk of needing little to no additional work and not requiring the maintenance of two code branches.

Minimising additional work required for big Scorpio gains does seem to be one of the overriding goals of the hardware design team.

They're also very clearly claiming CPU optimisations for Scorpio beyond just running at a higher clocks. They're definitely talking about changes that have improved IPC as well as clock speed.
And what if Sony made thousands of customisations?
It's a PR talk.
Double rate FP16 and ID buffer benefits were explained, these 60 customisations read like entirely marketing bullet point.
 
And what if Sony made thousands of customisations?
It's a PR talk.
Double rate FP16 and ID buffer benefits were explained, these 60 customisations read like entirely marketing bullet point.

Possibly PR talk or possibly not. We don't know yet because they were not explained in detail.
 
It lacks Double-rate FP16 not FP16 support all-together. Xbox One and PS4 where GCN2 based and didn't have support for single rate 16Bit ops which was introduced in GCN3 and supported in GCN4 (which is Polaris..thus Scorpio has is). PS4 Pro in addition supports Double-rate 16bit ops which will be part of GCN5 (Vega).
Which part cites Polaris being the design basis?
I interpret the article stating that the Xbox One was used as the base as a sign that it might still be similar at a CU feature level to GCN2, unless specifically cited as updated like DCC and some of the geometry changes.


Well you can't blame him for preaching that PR line just like MS wanting everybody to think that ESRAM/EDRAM was the best thing since sliced bread...the impact of double-rate FP16 is going to be minuscule compared to having more ram, more CUs, or ROPS or Bandwidth.. etc..It's a nice thing to have for sure but this ain't going to light the world on fire.
In his defense, Cerny's interest in the ID buffer may be more personal, given his involvement with it.
It might make more sense for the PS4 Pro given its co-development with the PSVR. Higher resolutions than the headset can support would not have the same worth than Microsoft's, and items like hardware-accelerated methods for detecting pixel velocities and more post-processing cleanup for a nearby screen might make sense.

Haven't seen either talk about whether they've improved the asynchronous compute capabilities, since there have been some improvements since the early days. Some of the hinted expansions of Scorpio's command processing might mean it snuck in, though.

They're also very clearly claiming CPU optimisations for Scorpio beyond just running at a higher clocks. They're definitely talking about changes that have improved IPC as well as clock speed.
One item was improved memory interface queues, although in part it would make sense since that block was replaced.
One core-area change is some unspecified memory translation changes for the always-present VM and hypervisor.
Perhaps specialized tweaks to VM instructions, or expansion and tweaks to the TLB hierarchy and page walker? It's a somewhat sensitive area of the core, which may have opened up for tweaking given the re-implementation at the new node.
 
And what if Sony made thousands of customisations?
It's a PR talk.
Double rate FP16 and ID buffer benefits were explained, these 60 customisations read like entirely marketing bullet point.

MS = "PR talk". Sony = "explained benefits". Aren't you tired of charging into Scorpio threads as the Sony guard dog?

No console vendor has ever talked about running the kind of simulations on hardware variations that MS has. This doesn't mean that they haven't, it's just that what they've described has never been described or even alluded to before.

I also clearly specified "if" MS have really nailed throughput due to the things they claim to have done. And MS have explicitly told DF that "the GPU supports extensions that allow depth and ID buffers to be efficiently rendered at full native resolution, while colour buffers can be rendered at half resolution with full pixel shader efficiency". This may be worse than Sony's implementation, or it may be better, but Secret Saucing an ID buffer is at this point silly because we know Scorpio can support something similar too. What is being described for Scorpio is remarkably similar to the implementation DICE used for their CB renderer.
 
Last edited:
I'm sure the fact that there would be roughly 80 million consoles already installed that do not support double data rate fp16, led ms to decide that support could only ever be a sku-specific optimization and have limited support. They also had no idea the pro existed or would support it so why use the die space on a feature that might only ever be on the Scorpio.
 
One item was improved memory interface queues, although in part it would make sense since that block was replaced.
One core-area change is some unspecified memory translation changes for the always-present VM and hypervisor.
Perhaps specialized tweaks to VM instructions, or expansion and tweaks to the TLB hierarchy and page walker? It's a somewhat sensitive area of the core, which may have opened up for tweaking given the re-implementation at the new node.

Given the areas of modification, would it be possible to estimate a ballpark for how much performance might have changed? Just how significant is the performance overhead of VM and hypervisor likely to be?
 
Back
Top