The diagram is bullshit. Well its not a graphics card diagram that much is sure.
likely nothing to do with VI or the xbox. May have nothing to do with AMD either.
I will reveal: this is the steam/apple console.
joke of course, maybe.
Xbox One (Durango) Technical hardware investigation
- Thread starter Love_In_Rio
- Start date
- Status
I will reveal: this is the steam/apple console.
joke of course, maybe.
Hold on.
Those are not 16 CUs. They are groups of 4 CUs that even Tahiti sports in its diagram. That part has 64 CUs.
Also, there are some obvious errors in that article. As that diagram shows 72 bit ECC DDR3 not GDDR5 on a 512 bit interface. Mistake me if I am wrong but would that mean this part has less bandwidth than Orbis?
Also, 20nm must be magical as Tahiti is 352 mm2 at 28nm. Doubling up on the ALUs, ROPs and TMUs while throwing in what looks like something analogous to 16 Steamroller cores would seem to lead to a gigantic chip.
Last edited by a moderator:
Looks like something more applicable to a server part maybe a new iteration of AMD's Sky line of GPUs.
Was initially my thought, too. But why would it have display port?
It could potentially be an Xbox part... If Microsoft is moving into having Xbox support in windows and they will be using it as a "games will work guaranteed" type thing or they potentially will be licensing out (as per the old yukon rumor) and they can shove it in any device... laptops, cable boxes, whoever wants it, gets it
Having the on chip/board security is really baffling to me
Last edited by a moderator:
AMD is implementing TrustZone on their APUs.
http://www.anandtech.com/show/6007/...cortexa5-processor-for-trustzone-capabilities
That diagram doesn't look like a console chip to me.
True
Load balancing and SMASH seems to point to server
But doesn't it seem a little... Low power? (I'm completely unaware of thermal or power demands of servers)
(((interference)))
Veteran
Durango has an ARM security chip.
can Infinity/ps4 be similar to intel/amd when coming to gp-gpu computing?
take an Intel’s Core i7-3770k and AMD’s A10-5800k, the intel have a weak HD4000 as gpu and amd a much stronger HD7660, this AMD gpu have more than double the Flops when compared to intel gpu, Intel’s HD 4000 has 16 EUs and AMD’s HD 7660D has 384 VLIW4 cores arranged into six groups
but
even if it's weaker, the HD 4000 can use the i7-3770k’s eight megabyte L3 cache to improve its performance.
results: the HD4000 is a best performer, in some cases really outperform HD7660
I'm believing that low-latency eSram can do the same thing as the L3 cache on the i7.
This is only for GP-GPU, not for general 3D, of course
take an Intel’s Core i7-3770k and AMD’s A10-5800k, the intel have a weak HD4000 as gpu and amd a much stronger HD7660, this AMD gpu have more than double the Flops when compared to intel gpu, Intel’s HD 4000 has 16 EUs and AMD’s HD 7660D has 384 VLIW4 cores arranged into six groups
but
even if it's weaker, the HD 4000 can use the i7-3770k’s eight megabyte L3 cache to improve its performance.
results: the HD4000 is a best performer, in some cases really outperform HD7660
I'm believing that low-latency eSram can do the same thing as the L3 cache on the i7.
This is only for GP-GPU, not for general 3D, of course
Last edited by a moderator:
Love_In_Rio
Veteran
can Infinity/ps4 be similar to intel/amd when coming to gp-gpu computing?
take an Intel’s Core i7-3770k and AMD’s A10-5800k, the intel have a weak HD4000 as gpu and amd a much stronger HD7660, this AMD gpu have more than double the Flops when compared to intel gpu, Intel’s HD 4000 has 16 EUs and AMD’s HD 7660D has 384 VLIW4 cores arranged into six groups
but
even if it's weaker, the HD 4000 can use the i7-3770k’s eight megabyte L3 cache to improve its performance.
results: the HD4000 is a best performer, in some cases really outperform HD7660
I'm believing that low-latency eSram can do the same thing as the L3 cache on the i7.
This is only for GP-GPU, not for general 3D, of course
In compute (Open CL benchmark you posted) is clear a cache will boost performance. In rendering as already said multiple times not so much.
in standard rendering I Agree, but a lot of post-processing can be done with gp-gpu computing, included AA and others things, for what came in my mind at the moment. is not meant for this the 4 CU on Orbis? to help rendering with gp-gpu computing?
Surely AA doesn't count as GPGPU as it's actually dealing with the graphics and presumably uses the graphics hardware? future AA techniques will use sophisticated graphics shaders.
There is no '4 CUs' in Orbis.
AA in post process can be implemented using shading, but maybe could be possible to achieve better results with a smartest approach.
The point is this:
is the hardware that have to be tought for software or viceversa?
I strongly believe that the software have the duty to exploit hardware capabilities, and if a machine have a strong point in GP-GPU, talented developers will find a way to make it shine. or they could change occupation
even in ps3 development, they has a weak gpu, but some spe useful in computing, and they pushed to use cell'spe in some way to help rendering. I think they've done a good work, in the end
now they removed the 4 CU isolation, but there was
Shaders are just computer programs that run on GPUs. Whether they are computing graphics or physics, they are shaders. Compute is just running shaders that calculate non-graphics tasks (or rather don't hit the graphics-specific components as a recent discussion on this board suggested the definition should be).
I've lost the thread of what you're trying to say. What is your point regards Durango and compute/GPGPU/CPU?
Or it was never in and that was a misunderstanding. But that's an Orbis discussion.
I count only 16 CUs with 4 SIMD 16-way each, for a total of 1024 ALUs.
The ECC memory controller, and the 16 Steamroller cores makes me think of a APU for high-performance computing, with around 1 Teraflops DP performance. As for size, at 20nm it shouldn't be much bigger than current Bulldozer or Tahiti chips.
On the paper it's seems a nice part for HPC market.
The ECC memory controller, and the 16 Steamroller cores makes me think of a APU for high-performance computing, with around 1 Teraflops DP performance. As for size, at 20nm it shouldn't be much bigger than current Bulldozer or Tahiti chips.
On the paper it's seems a nice part for HPC market.
Its 16 CU arrays not 16 CUs. Its just tahiti X 2 plus changes that arent so readily described by the diagram. One array equals 4 CU and tahiti sports 8 arrays.
Im guessing that AMD and Nvidia are going to offer graphics parts with cpus on them with the intention of creating parts that are capable of CPU based rendering in realtime. Basically a hybrid chip thats can deploy fixed/traditional programmable gpu hardware, gpgpu schemes as well as cpus for tasks. Anything done in reyes or ray tracing that cant be readily realized using gpgpu can simply be performed by a cpu array.
I dont thinks its applicable to durango as it may sport a capable configuration but lacks the necessary complexity and amount of hardware to pull it off. That diagram looks like a chip that if ran at durango speeds would be 7 TFLOPs or more.
Last edited by a moderator:
I would not attribute the merit only to "L3" or eSRAM, I think Intel has a really good design:can Infinity/ps4 be similar to intel/amd when coming to gp-gpu computing?
take an Intel’s Core i7-3770k and AMD’s A10-5800k, the intel have a weak HD4000 as gpu and amd a much stronger HD7660, this AMD gpu have more than double the Flops when compared to intel gpu, Intel’s HD 4000 has 16 EUs and AMD’s HD 7660D has 384 VLIW4 cores arranged into six groups
but
even if it's weaker, the HD 4000 can use the i7-3770k’s eight megabyte L3 cache to improve its performance.
results: the HD4000 is a best performer, in some cases really outperform HD7660
I'm believing that low-latency eSram can do the same thing as the L3 cache on the i7.
This is only for GP-GPU, not for general 3D, of course
cache seems to offers a lot of bandwdith are low latency, the mem controller is awesome (Pete linked the data for uncached access somewhere).
THe architecture seems narrower than the competition I don't know to which extend it helps with utilization of ALUs. It also seems that media processing and GPGPU were every bit as important (may be more) for Intel as 3d when they designed those GPU.
Still there is a lot more to those GPU performances than eSRAM /Intel engineering greatness.
liquidboy
Regular
Its what mistercteam and misterxmedia have been saying all along.. that Durango have been working towards this goal, they have the right minds at MS that could solve the problem of making this work in a console system..
There is a possibility, however remote, that Durango will be use a gpu very similar to a Volcanic Island Hawaii gpu. Consider these separate clues from the past:
1)Charlie saying the chip was having yield problems.
2) Discussion here on Beyond3d indicating that by looking at the behind of Durango dev kits, that they had two hd7970s in them.
3)Thuway on neogaf saying that the alpha dev kit had 7970s (note the plural, not singular) in them.
4)Thuway saying that the beta kits were running hot.
5)Paul Thurott saying Durango will cost $500, that even he admits, is expensive. (I ask Proelite, if Durango is meant to be a low-end console, why is it priced like a "luxury" console. Proelite, did you know that the base 30GB model of the Playstation 3 was priced at $499, initially?)
And finally, this week, the appearance a "leak", a AMD gpu coming at the end of this year, having the same computing power of two HD7970s, roughly the same time Durango is supposed to debut.
There might be nothing to these clues. But isn't there a possibilty?
1)Charlie saying the chip was having yield problems.
2) Discussion here on Beyond3d indicating that by looking at the behind of Durango dev kits, that they had two hd7970s in them.
3)Thuway on neogaf saying that the alpha dev kit had 7970s (note the plural, not singular) in them.
4)Thuway saying that the beta kits were running hot.
5)Paul Thurott saying Durango will cost $500, that even he admits, is expensive. (I ask Proelite, if Durango is meant to be a low-end console, why is it priced like a "luxury" console. Proelite, did you know that the base 30GB model of the Playstation 3 was priced at $499, initially?)
And finally, this week, the appearance a "leak", a AMD gpu coming at the end of this year, having the same computing power of two HD7970s, roughly the same time Durango is supposed to debut.
There might be nothing to these clues. But isn't there a possibilty?
No, because devs working with devkits now have a 1.2 TF GCN APU with eSRAM. If the intention was a powerhouse as far back as that alpha kit, why has MS gimped their developers with such low performance compared to what the console is actually going to have? It could only be true if the existing devkits with their custom APU hardware, which needed full design and manufacture, is just an elaborate cover-story.
- Status
Similar threads
- Locked
