It think Gubbi already gave definitive examples in the last two pages of what types of workloads completely avoid the ddr3. We can start there. I look forward to hearing more from developers on exactly what the data fetch process wrt to the esram can be. I wouldn't think it would be too different from how they managed the edram on 360 - with the added benefit of it being both a cache and scratch pad.
Technical Comparison Sony PS4 and Microsoft Xbox
- Thread starter BRiT
- Start date
- Status
It think Gubbi already gave definitive examples in the last two pages of what types of workloads completely avoid the ddr3. We can start there. I look forward to hearing more from developers on exactly what the data fetch process wrt to the esram can be. I wouldn't think it would be too different from how they managed the edram on 360 - with the added benefit of it being both a cache and scratch pad.
Betanumerical
Veteran
He only avoided having to discreeting read something into the eSRAM, his examples went like this.
DDR3 -> GPU.
GPU -> eSRAM.
It stilled used the DDR3 bandwidth just not purely for copying into the eSRAM.
Particles?? Shadowing??
Probably almost any poorly optimized or memory bound compute job where the source data is in the low latency memory.
Anytime you can't cover the cache misses with in flight wavefronts, the lower latency will be a win.
In my all be it limited experience with GPU compute, almost everything short of trivial (n in to 1 out) compute is memory bound not computation bound.
You'd use the exact same amount of bandwidth on PS4.
You'd read material textures and write your light prepass or G-buffer.
The only reason to do distinct copies into ESRAM would be if it was a win (reuse ends up saving DDR3 bandwidth).
Cheers
I think video game renderers are far too complex for us at this point to speculate about. Also, I'm quite sure that different game genres have very different resource utilization patterns, and then there are the differences between a fully deferred renderer, a forward renderer, a light pre-pass renderer, or a renderer with multiple components like Crysis 3, or any renderer relying on fully unique virtual texturing.
So the short answers are:
- XB1 and PS4 memory bus and pool requirements will be highly different among the various titles
- XB1 will need more developer attention to fully utilize the memory subsystem
However, PS2 is a good example of such a multilevel memory system (32MB RAMBUS RAM and 4MB EDRAM) and developers have managed to master it quite well.
And also, this past generation we've seen a lot of consolidation in development studios; I'd say ~70% of AA console development is done by the inhouse studios of the big publishers and they're sharing resources and engines as much as possible. So we can fully expect IW/Ubi/DICE/Epic etc. to do all the optimizations necessary to make the best use of XB1 they can.
So the short answers are:
- XB1 and PS4 memory bus and pool requirements will be highly different among the various titles
- XB1 will need more developer attention to fully utilize the memory subsystem
However, PS2 is a good example of such a multilevel memory system (32MB RAMBUS RAM and 4MB EDRAM) and developers have managed to master it quite well.
And also, this past generation we've seen a lot of consolidation in development studios; I'd say ~70% of AA console development is done by the inhouse studios of the big publishers and they're sharing resources and engines as much as possible. So we can fully expect IW/Ubi/DICE/Epic etc. to do all the optimizations necessary to make the best use of XB1 they can.
TheWretched
Regular
Looking at PS2 numbers, it's quite astonishing to see 1/8th of the main RAM as EDRAM. That'd be 1GB with todays consoles.
The edram in ps2 was also massively faster than the main memory. So if Xbox One had 1gb of edram at 1TB/sec that would be a similar design.
I think that, for a fairer comparison you need to compare the HD 7850 to the HD 7790 Bonaire GPU.
I am amazed at how accurate and premonitory this article was when it was published time ago, even Eurogamer linked to it in a Digital Foundry article:
http://www.extremetech.com/gaming/151367-amd-launches-radeon-7790-meet-the-xbox-720s-gpu
I understand that you are under NDA and you can't say much but the execution timeline -both revealed practically at the same time- clearly shows that the Xbox One GPU is basically an HD 7790 Bonaire in many ways.
But is the 200GB/s a real thing? I mean, is it a "hypothetical" value?or is it a "reachable" value (most time) ? talking about the GPU side.
1.4TF + 410GF = 1.8TF
Bagel seed
Veteran
About as real as this one for 360
No.. 102+68+30 is real.
About as real as this one for 360
Well, in some grade it is real, and usefull.
But it is not "official" I guess, Sony said "1.8TF" for GPU.
- Status
Similar threads
- Locked
