Xbox One (Durango) Technical hardware investigation

[Hecatoncheires]

Ah i see, so SoC = components bunched together, APU = component Hivemind, basically.

Exactly.

Would an MCM like the Wii U uses be somewhat similar or is that something completely different?

Completely different

 

[Inuhanyou]

Hm, i see. I guess i can see now why AMD would be so excited by the prospects of these machines(especially durango) and bet the farm on them. Durango will be the first fully integrated mass market HSA devices of their own creation, so its like a test run i guess as much as Cell in PS3 was.

So i'm assuming that a 1.2 tflop GPU with 8gb of DDR3 ram and 8 jaguar CPU is going to behave differently from a similar 'conventional' 1.2 tflop GPU set up in similar conditions, and that's even before the standard optimization process that comes with fixed hardware like consoles.

 

[Hecatoncheires]

So i'm assuming that a 1.2 tflop GPU with 8gb of DDR3 ram and 8 jaguar CPU is going to behave differently from a similar 'conventional' 1.2 tflop GPU set up in similar conditions, and that's even before the standard optimization process that comes with fixed hardware like consoles.

That's what I'm expecting as well.

But nobody who's not under a NDA can tell it for sure.

 

[cal_guy]

A good analysis from extremetech.

Assuming that this information is accurate, the major difference between Durango and a standard GCN is in the cache structure. Modern Radeons have a 16K L1 cache that’s four-way set associative and a shared L2 cache that’s 16-way associative. Durango reportedly has a 64-way associative L1 cache (still at 16K) and an 8-way associative L2.

http://www.extremetech.com/gaming/147577-xbox-720-gpu-detailed-merely-a-last-gen-radeon

 

[Inuhanyou]

They got a few things wrong though...like 4GB of DDR3 instead of the 5GB minimum rumor we have right now. And he neglected to mention anything about the CPU, even saying it was unknown when we know its Jaguar based.

 

[itsmydamnation]

he aint called Cal_guy for nothing. read into that what you will......lol

 

[sir doris]

It's very efficient.

Do you remember when the first dual cores emerged? Instead of having one single thread you divide your code into two threads and it will receive a sweet speedup. A Core2Duo with two 1.8Ghz threads was able to ditch a 3Ghz Pentium 4 easily. The HSA is basically the same, but instead of having just more cores, you integrate different kinds of cores into a single processor. Ony type of core is great for runtime intensive tasks, the other type is better at paralellizable tasks.

The advantage the Core2Duo had over the Pentium 4 was not the number of cores but the much more efficent architecture (much better IPC). Don't forget we already had dual core Pentium 4s.

 

[Hornet]

Interesting. I would find it odd if they went through the trouble of changing L1 cache associativity without any other tweaking of the CUs. Maybe it's one of the changes in GCN 2.0 or the GPU is indeed more custom than it looks, just in more subtle ways that we expected.

 

[Trejser]

Move engines http://www.vgleaks.com/world-exclusive-durangos-move-engines/

 

[Hecatoncheires]

The advantage the Core2Duo had over the Pentium 4 was not the number of cores but the much more efficent architecture (much better IPC). Don't forget we already had dual core Pentium 4s.

That's not the point. Exchange Core2Duo and Pentium4 with each other multi-core/single-Core combination if you like...

 

[LightHeaven]

On chip embedded DRAM would have lower latency than external memory. Otherwise IBM wouldn't bother putting gobs of EDRAM as L3 cache on Power 7 chips. "Low" is a relative term. Without knowing an actual point of comparison we can't conclude anything.



Or maybe they're just DMAs since they're where the DMAs should be and MS renamed everything else.

They very well could. But that wouldn't be any fun

 

[McHuj]

So it looks like the move engines are just as theorized, dma's with some on the fly compression/decompression capability.

I'm really curious to see how many ports the ESRAM has, with DMA's copying data to and from it, and while the GPU processes data in it, there could be a lot of contention. I'll guess its dual ported and banked in large chucks so neither the GPU nor the DMA interfere with each other too much.

 

[Betanumerical]

From the article.

The four move engines share a single memory path, yielding a total maximum throughput for all the move engines that is the same as for a single move engine.

Does this mean that regardless of how many of them you use you can only move a max of 25.6GB/s ?. I do not see these being a great deal of use to be honest (in the secret sauce way), and if I what I think above is true they seem a little slow as well.

Based off the above numbers (25.6GB/s at most).

For a game at 30FPS they can move 873.8MB/Frame.
For a game at 60FPS they can move 436.9MB/Frame.

And this is assuming that they are all doing only GPU work (they are not) and they are also all only being used to move Data to the GPU (they wont be).

EDIT:.
this article to be exact.
http://www.vgleaks.com/world-exclusive-durangos-move-engines/

 

[Shifty Geezer]

From the article.

Does this mean that regardless of how many of them you use you can only move a max of 25.6GB/s ?.

Yep. But it's not about BW.

I do not see these being a great deal of use to be honest, and if I what I think above is true they seem a little slow as well.

It's about using a tiny piece of silicon to do what you'd otherwise do on a CPU core or GPU shader. It's an economy of scale I believe, rather than a power feature that injects improved performance. From the sounds of it, Orbis's 4 CUs could be operating in a similar fashion, able to move and process data while the 14 GPU CUs do the graphics work.

One thing that's surprised me is the lack of functionality. No decompression of JPEG into DXTC means your stuck with bitmaps in ESRAM to draw. So I'm not seeing exactly where the JPEG compression is useful. It's not even JPEG2000.

 

[Hecatoncheires]

"Studios working with the next-gen Xbox are currently being forced to work with only approved development libraries, while Sony is encouraging coders to get closer to the metal of its box. Furthermore, the operating system overhead of Microsoft’s next console is more oppressive than Sony’s equivalent, giving the PlayStation-badged unit another advantage." EDGE

Seems as if Timothy Lottes was totally right after all. Windows 8 Box incoming?

 

[Betanumerical]

Yep. But it's not about BW.

It's about using a tiny piece of silicon to do what you'd otherwise do on a CPU core or GPU shader. It's an economy of scale I believe, rather than a power feature that injects improved performance. From the sounds of it, Orbis's 4 CUs could be operating in a similar fashion, able to move and process data while the 14 GPU CUs do the graphics work.

One thing that's surprised me is the lack of functionality. No decompression of JPEG into DXTC means your stuck with bitmaps in ESRAM to draw. So I'm not seeing exactly where the JPEG compression is useful. It's not even JPEG2000.

Yeah sorry I worded my post wrong, I can see there use I just don't see it as a big deal (ie megatonne secret sauce). But getting away from that I thought that the Orbis was also slated to have DMA controllers with hardware compression as well, so something although not quite as complex.

 

[dumbo11]

"Studios working with the next-gen Xbox are currently being forced to work with only approved development libraries, while Sony is encouraging coders to get closer to the metal of its box. Furthermore, the operating system overhead of Microsoft’s next console is more oppressive than Sony’s equivalent, giving the PlayStation-badged unit another advantage." EDGE

Seems as if Timothy Lottes was totally right after all. Windows 8 Box incoming?

As far I can tell, that is no different from Xbox 360 development. There is a lower level API than windows 8, but you "must" use it.

If the Durango is a partitioned box, then it's quite essential that Microsoft force developers through a properly tested API rather than letting them trample all over the APP/STB overlay.

My own guess is that Durango will have a Windows/DirectX API for the set-top-box, and an xbox/DirectX API for the game console.

 

[LightHeaven]

"Studios working with the next-gen Xbox are currently being forced to work with only approved development libraries, while Sony is encouraging coders to get closer to the metal of its box. Furthermore, the operating system overhead of Microsoft’s next console is more oppressive than Sony’s equivalent, giving the PlayStation-badged unit another advantage." EDGE

Seems as if Timothy Lottes was totally right after all. Windows 8 Box incoming?

As long its games are completely compatible with other W8 machines and durango is essentially a Ms steambox, i would be fine with that

 

[Gubbi]

That's not the point. Exchange Core2Duo and Pentium4 with each other multi-core/single-Core combination if you like...

Your point is wrong.

There were almost zero programs that were multi threaded when dual cores appeared. My Athlon X2 ran worse than my single core Athlon 64 it replaced because single threaded programs would bounce back and forth between cores unless process affinity was set.

Converting programs from general purpose CPUs to GPU/APU architecture isn't going to be straight forward just as multi threading programs weren't straight forward.

Cheers

 

[dumbo11]

One thing that's surprised me is the lack of functionality. No decompression of JPEG into DXTC means your stuck with bitmaps in ESRAM to draw. So I'm not seeing exactly where the JPEG compression is useful. It's not even JPEG2000.

I think people won't like this idea - but is it possible that this is simply related to the "set-top box/browser/whatever"?
AFAIR the DME is inside the CPU rather than the GPU?

Edit: Actually the article explicitly says that it isn't for the console - but that 1-3 are exclusively for use by the game.