Digital Foundry Article Technical Discussion [2022]

Status
Not open for further replies.
palpatine-star-wars.gif
The latest news letter:
* Alex is spending some time behind the scenes this week looking at new image quality analysis techniques.

Let’s go!!
 
So I've been trying to get some kind of rough idea of how the SM 6.4 version of XeSS might run on Series S. And yes, my entire hypothesis is based off of this one part of the DF video (y) :

View attachment 6939

The RTX 3070 is shown upscaling using XeSS from 1080p to 4K in 3.8 ms. With default clock it has specs of, I think:
- 20.3 TF fp32
- 10.2 TOPs int32
- (afaik) 40.6 non tensor TOPs int8 (I'm multiplying int32 x 4)

Xbox Series S has specs of:
- 4 TF fp32
- 4 TOPs int32
- 16 TOPs int8

So just going by the numbers (which is always risky!) you're looking at XSS being ~ 1/5th speed at fp32 and int32, and ~ 2/5th speed at int8. So depending on the balance of instructions used, and assuming the 3.8 ms represents all the cost of XeSS, you might expect the XSS to be 20% to 40% as fast in this scenario, upscaling from 1080p to 4K. So taking 9.5 to 19ms where the 3070 takes 3.8 ms.

If DP4a is leant on heavily in XeSS, then that might be expected to skew towards the better end of the range for Series S, as it is in less of a deficit for int ops than flops.

However, you probably wouldn't be trying to scale from 1080p to 4K on the Series S. Assuming that XeSS workload scales broadly linearly with resolution, upscales from 540p to 1080p (2.4 ~ 4.8 ms?) or even 720p to 1440p might be within reach.

All very highly speculative of course, so take with a big 'ol pinch of salt!

I don't see any specs that say RTX is able to spit out non-tensor int8 at 4X the rate of its int32 units. Plus, XESS seems pretty heavy. Isn't the A770 at 200+ int8 TOPs? A380 puts out int8 at 64 TOPs and only has 1/4th the Xe cores of A770. If so XESS on the RTX 3070 is probably using tensor cores.

If that's true, XESS doesn't seem like a natural alternative to solutions like FSR 2 for lower-end non-ARC gpus (outside of a RTX 2060/3050).
 
Last edited:
At 19:20 and onwards, MS states that costs per transistor is having foundational impacts on console development. I noted this before some days ago here, it was refuted largely. However its going to bring problems there too. Prices are going to climb excessively if we want same performance for the same price, even on consoles. Unless companies are going to eat it all ofcourse, a 1000+ dollar BOM and the customer paying half that.....
Their gobsmacked by what RTX can do, both to old games using rtx remix, but also CP2077 with a total reworked RT and Portal. RacerX is a look at true next gen.
 
Last edited:
Interestingly a Portal dev felt like the auto-RT treatment from NV looked horrible and altered the look of the game in bad ways.

Regards,
SB
You'll always have purists who prefer the look of the original. For instance, I think Blue Print did a bad job with Demon's Souls remake despite it looking objectively much more advanced than the original. I still prefer the color palette and style of the original. The remake looks like a generic fantasy ARPG to me.
 
Would prefer to see comparison using dlss quality mode. Performance mode usualy is below native quality and also cant imagine rtx4090 user to use performance mode.
 
Status
Not open for further replies.
Back
Top