Xbox Scorpio Hardware Prediction Thread [April 2017 Revision]

Can the B3D gestalt accurately predict Scorpio hardware?


  • Total voters
    58
  • Poll closed .
Status
Not open for further replies.
You're suggesting Microsoft went to Digital Foundry of all places to attempt to pull the wool over the eyes on people of the specs/performance? Most who even know about this reveal or watch/read Digital Foundry are the most hardcore of fans and tech enthusiasts.

I am hoping for a simple breakdown and analysis of the specs whatever they end up being.

I expect MS to show the strengths of the system which need to lay in CPU offloading to GPU if MS wants to impress the general public with a Jaguar
 
Think about it, though. To get the bandwidth you want you either have to go wide or go fast. Using less memory to get to the same bandwidth would therefore require you to:
  • Use smaller chips on the same size bus, meaning you're not getting as much as you could out of the investment in that bus and the mainboard real estate that the chips and traces take up, and then adding *another* bus and taking up yet more mainboard real estate with your secondary memory chips and traces
  • Run faster, more expensive, chips on a smaller bus to get your bandwidth (paying more per GB for that memory) so some of your memory can be cheaper.
I can't say it's impossible that either of those approaches ends up cheaper, but it seems a little shaky to me.
My 'concern' was always board layout and are they called pads (going from chip to pins), so yea don't disagree, just don't know the economics.
That's why hbm would be better, but it's too soon and expensive for that.
If it was using zen, then may have more pad space as it includes some of the functionality that would be done of chip, etc.
I do think this will be the way that amd laptops will go though.
 
CPU: 8 core Jaguar at 2.1 GHz. The extra CPU power is reserved for OS functionality.

12GB GDDR5 ram on 384 bit bus. 6GB reserved for OS.

5.8 teraflop 16bit RX450 derivative with Vega features. 2.9 teraflop 32bit. 1.85 teraflop 64bit. 200 gigaflop 256bit.

Secret Sauce: Checkerboard and motion smoothing black silicon that emulates the feel of native 4k and 60 fps.

5% of GPU reserved for OS. 10% reserved for Kinect 3.0

Cloud computing makes this the most powerful console but an always online connection is required.

Used games are no longer allowed.

$599 with Kinect 3.0, 4K Blu-ray, Keyboard and mice. It is a premium console after all.
 
I mean yeah, that'd be one way to look at it I guess, LOL. I guess the worst case scenario (cue Bear Grylls) deserves play too.
 
Same as Pascal/Ps4S/X1S/Ps4Pro then, TSMC 16nm FF+
Is GoFlo 14nm? Is that what this part of the discussion is about? I'm assuming GoFlo 14nm is just not ready for prime time manufacturing yet? So everyone is waiting for it to happen for cost reductions?
 
Is GoFlo 14nm? Is that what this part of the discussion is about? I'm assuming GoFlo 14nm is just not ready for prime time manufacturing yet? So everyone is waiting for it to happen for cost reductions?

It's another reason to think no Zen because Zen was built for GloFo's (Samsung's) 14nm process.
 
and we're back to square one. The polls represent our dilemma well. On CPU we're having a hard time guessing where the hardware will land. There's just too much support on both sides of the argument.

So I was doing some research into Execute Indirect again. I know, I know, obsessive. And GPUs being able to spawn their own draw calls is a big deal, even if in a limited fashion. I know we talked about how XBO has customizations to make Execute Indirect more flexible, but as I continued to do some research I had to ask why it wasn't used.
my thought process was the following:
a) there was enough CPU available in the console space such that there was no reason to offload draw calls to the Command Processor
b) it is coming but will be the final bit of optimization that will happen over the course of this generation
c) it's not available on every platform

a) didn't seem interesting to me, reason being, MS engineers noted their biggest failure point was the CPU for them, which is why they clocked XBO higher. They really needed that boost to get their frame rate higher. Long story short, Xbox OS is heavier and they needed more lift. And that was all they could give it. A majority of the games on XBOX are probably GPU limited, but that's not to say there are probably instances where they may hit CPU bottlenecks from time to time. Considering how weak the GPU is on XBO, being CPU limited before GPU limited is likely not frequent.

b) this is a possibility because we are now starting to see this feature show up in slide decks in GDC. Perhaps it's true that 2-3 years from now we will see it in games, I'm not sure... but that leads me to dealing with (c)

c) I went to check if execute Indirect as per function, available on every platform. It's not. It's actually _not_ available on Vulkan. They require a nvidia extension to support it and I imagine that only applies to nvidia cards. Looking at the forums for Vulkan there is demand for this feature, but from what I've been reading, this feature would require a big lift for them to support and that demand is noted but expectations on delivery of this feature soon is unlikely. that got me thinking, GNM is apparently modelled very close to Vulkan, what if GNM doesn't support execute indirect either -- that a very large change to the API would need to be done to support it. So if Vulkan and PS4 don't support execute indirect, why would any 3P support it?

Summarizing that up, the biggest feature that would offload CPU work, just won't be a thing for this generation. And that to me is a big strike against keeping with Jaguar. You've got 4K resolution as a marketing goal, but that's just one aspect. As DF folks have alluded to, the rate in power differential between generations will lessen because we're slowing down in terms of how many transistors we can fit. So the new game is how to use that power, and there's going to be a lot more focus on hardware features supporting new rendering techniques. But this generation doesn't yet align in supporting that, so brute forcing the solution with more CPU seems to make more sense now.
 
Polaris is GloFo and both X1S and Ps4 Pro ended up with TSMC 16nm

How different do you think the CUs in PS4 Slim and XB1S are from those in PS4 Pro? I'd wager not that different, so they could directly apply the work done to shrink the PS4 and XB1's APUs to 16nm towards building PS4 Pro and (potentially) Scorpio on that process with the only work needing to be on the parts of the chips that have changed.
 
I was honestly a bit split between Zen and Excavator+ for CPU.
I don't think it'll be Jaguar/Puma based, simply because even at 16FF the architecture gets its sweetspot at 2.1GHz (according to Mark Cerny) and that will probably hold back a 6TFLOPs GPU too much.

3/4-module Excavator+ at ~2.5GHz would probably be a great fit for a console with a 6 TFLOPs GPU. But this is only because so many people have been claiming Zen is too green to go into a SoC.

However, AMD wants to forget Bulldozer architectures ASAP and they want developers to work with Zen at all costs.
Plus there's all the hints and AMD rep "winks" whenever someone asked why "Scorpio" logos were being put between Zen and Vega posters during technology fairs.

For me, in terms of probability we have:
1st: Zen
2nd: Excavator (only if AMD couldn't put Zen on time)
distant 3rd: Jaguar


How different do you think the CUs in PS4 Slim and XB1S are from those in PS4 Pro?
Different enough to support rapid packed math on each ALU, at least. That certainly doesn't come for free.
Scorpio is definitely bringing 2*FP16 and 4*uint8, I have very little doubt about this.
 
Status
Not open for further replies.
Back
Top