Betanumerical
Veteran
As do i now, but its clearly two compute pipes but anyway.
If it is GCN1.1 its odd that they have decided to forgo the extra pipes that the PS4 went with, theres not much reason to axe them.
Xbox One (Durango) Technical hardware investigation
- Thread starter Love_In_Rio
- Start date
- Status
As do i now, but its clearly two compute pipes but anyway.
If it is GCN1.1 its odd that they have decided to forgo the extra pipes that the PS4 went with, theres not much reason to axe them.
What does AMD call it?
Also, which link can I read for more on "and what the leaks said about Durango I would think that Durango is based on GCN 1.1"?
Betanumerical
Veteran
GCN1.1 doesn't bring that many changes its a minor refresh not a major revision. So it would not be that surprising to be honest.
FWIW the whole hardware SPURS 64 pipes thing was driven entirely by Mark AFAIK, so I don't think MS would have had the option of adopting.
They probably could have requested additional ACE's, but I don't think they have the same emphasis on asynchronous compute as Sony does.
Both chips will probably have revisions similar to parts shipping shortly after they do, but there are a lot of things not publicly disclosed that could impact performance register file sizes, internal cache sizes, the basic building blocks look to be similar, but thats the 10000ft view.
Calling them GCN 1.X or 2.X or whatever isn't useful without being able to define the distinction.
They probably could have requested additional ACE's, but I don't think they have the same emphasis on asynchronous compute as Sony does.
Both chips will probably have revisions similar to parts shipping shortly after they do, but there are a lot of things not publicly disclosed that could impact performance register file sizes, internal cache sizes, the basic building blocks look to be similar, but thats the 10000ft view.
Calling them GCN 1.X or 2.X or whatever isn't useful without being able to define the distinction.
AMD at this point doesn't describe the implementations with a specific GCN version number, and going forward there's going to be varying mixes of IP. Generational code names, numbers, versions, and other meaningful methods of identification are not forthcoming.
The numerical designations being thrown around are at the determination of the writer, sometimes for the same chip.
We can try to go by codenames for specific chips, and possibly say how much they are related to one another, but there's no clear hierarchy when things can be mixed differently for devices that have overlapping development periods.
I don't know what term AMD has used for the console chips, other than "the chip for this console" or somesuch.
True. What I was trying to get at was whether or not the Xbox One could have features/improvements in common with VI/Hawaii a bit like Xenos and R600.
(And then what are those and where can we read about them.)
So I wanted to know what AMD calls VI/Hawaii and then find leaks/documents talking about GCN ?X? if that is want AMD calls it.
I don't know how minor or significant the VI/Hawaii changes are, but perhaps worth a discussion.
Sorry I read the thing again and I might have missed again, nothing point to two ACEs in VGleaks leaks.
The article makes clear that the GPOU has a dual geometry set-up but I'm not sure that we can assert from that that the GPU has 2 ACEs.
We don't know if they axed the extra command queue or not, I though it could be the refresh because of that part:
But may be it is not needed for the CPU and the GPU to efficiently share the same virtual memory space.
I see. Not as clear a situation as discussing Llano, Trinity or Richland.
I was referring to the two orange rectangles on the left of this graphic (labeled Graphics Commands):
http://www.vgleaks.com/durango-gpu/block_gpu/
So I asked if that corresponded to two ACEs. (Actually I think I asked about two Command Queues.)
Betanumerical
Veteran
Everything we have seen so far aside from the changes to the memory controller setup and eSRAM interfaces etc, have been stock GCN. Until other information comes out lets not go pretending it contains anything from a unreleased generation.
It wouldn't be any kind of argument, Xenos & R600 are quite a bit different architectures, starting with the fact that Xenos uses Vec4+Scalar "shader processors" which for closest match would be R5x0's vertex shaders while R600 uses VLIW5 units.
Well one say graphic the other compute, it is just a diagram but the article give not hint of 2 ACEs at all.
I think what betanumerical is saying is that dual ACEs is "part" of GCN hence his ref to the white papers.
Kabini has four ACEs (32 active queues) for example, without insiders saying something VGleak is not good enough to say one way or another.
Durango should have 2 ACEs, though which? I would think the last because of the support for coherent memory space.
I am not sure it falls into the classification of "pretending". I could say it is more unreasonable to "pretend" it will be based upon two year old architecture at launch than current architecture.
I think the Xenos/R600 example is reasonable and a good argument to expect that it would not be dated by two years by release.
Obviously not definitive but I think "pretending" is uncalled for and not accurate.
Betanumerical
Veteran
We have seen nothing to support your suggestion, therefore Im going to call it 'pretending' until some real evidence comes around.
Show me one thing aside from memory interfaces that goes beyond standard GCN1.0/1/1/current cards. that is in XBONE and not the PS4 (i.e. not something for consoles such as the low latency graphics pipe).
The vgleaks article describes to practically a T standard GCN and it is honestly getting a little annoying seeing people pop up everywhere suggesting there is some undiscovered secret sauce in it.
Yes, they are different. But Xenos was not "out of date" by two years. It did contain advancements along the lines of where GPUs were going. Can't expect MS and AMD to go in lockstep. The different teams will have different ideas. Some the same and some similar but customized and then others which might be totally "one off" and not in common perhaps like eDRAM and perhaps now eSRAM. I think the UMA and unified shaders fit into the former categories reasonably well and that eDRAM and now perhaps eSRAM could possibly fit the latter categories.
I think two year old GCN 1.0 (by release date) is the much more tenuous argument.
Betanumerical
Veteran
Why, this console should have been nailed down a year or so ago, they can't keep changing the technology right up until people buy the machine thats not how things work.
Granted the vgleaks seems to be doing pretty well so far.
But it is also pretty annoying that some keep pointing to one leak and saying that is gospel before MS confirms and officially releases. (And 100% rejecting all other leaks and speculations.)
It could be true but until MS officially releases I think it is perfectly reasonable for there to be interested in the possibilities. Based upon the research MS does (based upon their publications and staff) it seems MS is interesting in strong customization. The Xenos seems to suggest that too, although things could be quite different this generation.
Plus "xbone", "secret sauce" and "pretending" don't tend to appeal in terms of responses to technical ideas and questions. The adherence to "xbone" being GCN 1.0 and nothing more seems too narrow and restrictive to fit the range of possibilities.
Especially once you consider all simulations MSFT supposedly ran.
That is not a logical argument. The 360 was innovative at the time of release. 7970 was innovative at the time of release. Xbox One could be innovative at the time of release too. All three had to be nailed down a year to so prior to release.
Anyway how do you know that the PS4 is based on GCN refresh vs something custom that looks close to it?
And GCn 2.0 doesn't exist for now and may never exist if AMD marketing department decides so.
- Status
Similar threads
