Xbox One (Durango) Technical hardware investigation

[Gipsel]

It's not something shown for the original version of GCN. It seems to be something the coming consoles have brought up.

The funny thing is, if one looks a bit to chapter 10 (pipe priority settings) of this documentation from september 2012 jlippo just linked to, one can find hints to a second graphics pipe with the same name (HP3D = high priority 3D) as in the Orbis docs on vgleaks.
Edit: That is for Sea Islands/CIK aka GCN 1.1, not the original GCN (which has there just priority setting for the three queues it supports [1 graphics + 2 compute]).

 

[3dilettante]

I wasn't sure if it would be there like so many other console tweaks have made their way into Sea Islands, and whether the expansion of the front-end in the Hawaii could have included it.

Maybe Mantle can make use of it by giving the OS a front end that its driver can link to as a safety measure in case the Mantle code does something wrong or takes too long for the response time threshold.

There's also a RESTORE bit, which mentions something about continuing after a context switch, whatever that may mean.

 

[Cyan]

Can someone help me understand if this is standard GCN architecture or customization!


-eurogamer article


Basically it appears the GPU can execute CU and ROP work concurrently!

So the "Title(GAME)" is processing via the CU and at the same time the "System(APP)" is processing thru ROP's.. All concurrently, this is how i'm interpreting the statement.



Also the statement "Two Concurrent render pipes" is confusing me. In OpenGL language a "render pipe" is a GPU. From the OpenGL documentation "... OpenGL context is created for each rendering pipe, i.e. for each GPU" .. So i interpret this as the Xbox One with its two concurrent render pipes, behaves like a dual GPU setup???

Can anyone shed light on what the Xbox One architect is saying here?!

What makes you think that? Saying that next gen consoles have two GPUs seems rather improbable, awesome and farfetched at the same time.

I was hoping to read something about those claims this weekend in the new DF article, but I won't at home by then and there isn't an internet connection where I am going to stay at.

 

[liquidboy]

What makes you think that? Saying that next gen consoles have two GPUs seems rather improbable, awesome and farfetched at the same time.

I was hoping to read something about those claims this weekend in the new DF article, but I won't at home by then and there isn't an internet connection where I am going to stay at.

I know it sounds far fetched... as I said I'm trying to make sense of the architects definition of a "Render Pipe", of which he is saying the Xbox One has two of.

I have worked in architectures/systems where a "RENDER PIPE" is defined as an explicit GPU ... And in a multi-gpu system, eg. 2 gpu's, would be called "two render pipes" ... In my mind it sounded like he was suggesting it was behaving like 2 gpu's , that is all!

P.S. I too am hoping to read more of the architecture in the follow up post. Truthfully the Xbox One architecture intrigues me coupled with the idea that we as devs can get our hands on it and develop on it!..

 

[Betanumerical]

I know it sounds far fetched... as I said I'm trying to make sense of the architects definition of a "Render Pipe", of which he is saying the Xbox One has two of.

I have worked in architectures/systems where a "RENDER PIPE" is defined as an explicit GPU ... And in a multi-gpu system, eg. 2 gpu's, would be called "two render pipes" ... In my mind it sounded like he was suggesting it was behaving like 2 gpu's , that is all!

P.S. I too am hoping to read more of the architecture in the follow up post. Truthfully the Xbox One architecture intrigues me coupled with the idea that we as devs can get our hands on it and develop on it!..

Hes talking about the highspeed / priority GFX pipe (OS stuff) and the normal game GFX pipe, thats what he means with 'dual render pipes'.

I hate to bring the other console into it but its the only diagram thats available publicly.

http://www.vgleaks.com/wp-content/uploads/2013/02/gpu_queues.jpg

The far left, HP GFX and GFX 3D and CS. 'dual render pipes'.

 

[Jay]

The final part of DF article is up:
http://www.eurogamer.net/articles/digitalfoundry-the-complete-xbox-one-interview

 

[pjbliverpool]

Awesome article answers so many questions. Kudos to those guys for being so open.

 

[Gipsel]

The final part of DF article is up:
http://www.eurogamer.net/articles/digitalfoundry-the-complete-xbox-one-interview

So now it is also officially confirmed: XB1 has 2 ACEs with 8 queues each (16 total).

 

[dumbo11]

Awesome article answers so many questions. Kudos to those guys for being so open.

Indeed - a very good article, and kudos to both Eurogamer and MS for doing it.

Just like our friends we're based on the Sea Islands family.

I was assuming Sea Islands is partly based on the XB1/PS4 rather than vice-versa.

Edit: they also do not take the opportunity to mention PRT, which is fascinating.

 

[Rangers]

Awesome article answers so many questions. Kudos to those guys for being so open.

Most interesting new points to me:

-Talk about why ESRAM was chosen.

-Mention 8GB was in mind early on (contrary to the fact that reliable sources say the spec was 4GB as late as late 2011)

-They brush aside a ESRAM latency question. Hmm, at first blush, not a good sign for latency benefits.

-Confirms there's no major customizations to the Jaguar CPU's.

-They paint a picture of higher bandwidth vs more CU's for compute, and claim to not know how the battle will turn out.

 

[Nisaaru]

They still play this ridiculous power consumption game while the PSU specs are known.

 

[MarkoIt]

Most interesting new points to me:

-Talk about why ESRAM was chosen.

-Mention 8GB was in mind early on (contrary to the fact that reliable sources say the spec was 4GB as late as late 2011)

-They brush aside a ESRAM latency question. Hmm, at first blush, not a good sign for latency benefits.

-Confirms there's no major customizations to the Jaguar CPU's.

-They paint a picture of higher bandwidth vs more CU's for compute, and claim to not know how the battle will turn out.

They have done some customization at the command processor. But overall, the interview doesn't put Xbox One in a better light. Going on the territory of bandwidth figure fight, doesn't seem a good idea. They may hit an higher figure than Ps4 but with more effort and only with some payload. And we are not talking of a night/day difference, but 20-30% higher.

They do have a much higher bandwidth/flops figure, but it's only because they have less flops.

 

[Shifty Geezer]

They still play this ridiculous power consumption game while the PSU specs are known.

(Angry Eyes) This isn't a stupid, fanboy versus or corporate PR thread. This is technical investigation to be held on a technical level by only those both capable and willing. We have unprecedented hardware info prior to release to consider, discuss and evaluate - if that doesn't excite you, you're on the wrong forum.

 

[liolio]

That DF interview was a great read, it is obvious that a lot of thoughts went into Durango, the OS side thing looks like a pretty great piece of software engineering. It is pretty awesome to have this much info from the horse mouth and prior to the system release, any geek should be happy
If I get things right lot of sounds and kinect processing is providing "for free" /done on the system side with its reserved resources.

The part that caught my attention the most is the part about the custom command processor and how those engineers went how of their way to mention that they wish that those features makes it to vanilla PC hardware. MSFT cares about PC gaming, nice to hear, and as AMD hinted MSFT might know about Mantle and to me it makes Mantle announcement a bit "transitional".

 

[Jay]

I would've liked to hear if it's a tier 1 or 2 DX11 part.
More details about the scaler, although maybe they've said enough and I just don't understand enough of it lol.
Interesting their take on gpgpu and cpu gpu bandwidth, anyone here have any views on their take on it?

Overall I thought it was a good read and regardless of the reasons behind it, I'm appreciative of it, and all things considered seemed to be as honest and a little less PR as you have to expect with such things.

 

[liolio]

I would've liked to hear if it's a tier 1 or 2 DX11 part.
More details about the scaler, although maybe they've said enough and I just don't understand enough of it lol.
Interesting their take on gpgpu and cpu gpu bandwidth, anyone here have any views on their take on it?

Overall I thought it was a good read and regardless of the reasons behind it, I'm appreciative of it, and all things considered seemed to be as honest and a little less PR as you have to expect with such things.

I do not remember precisely the presentation AMD made a week ago but iirc only their last GPUs support Tiers 2. Durango and Orbis are based on Sea Island I would guess that both are tiers 1.

 

[Grall]

Durango and Orbis are based on Sea Island I would guess that both are tiers 1.

Seeing as both the new consoles and "hawaii" AMD GPUs were developed during the same timespan, it's within the realm of possibilities that consoles include tier 2 functionality. Completely unconfirmed, of course. Also, tier 2 seem to be of rather minor benefit from my limited understanding, so even if it's missing it's not as if we're going to miss it much.

Most games today don't even offer proper anisotropic filtering for textures and such (and gamers overall also don't take notice of the quality reduction), so the possibility of missing hardware MIP mapping support for virtual texturing and whatnot won't be a great loss I wager.

 

[Jay]

I do not remember precisely the presentation AMD made a week ago but iirc only their last GPUs support Tiers 2. Durango and Orbis are based on Sea Island I would guess that both are tiers 1.

They already have gpu's out that are both tier 1 and tier 2 (gcn 1.1), so it wouldn't surprise me if either/both console are also tier 2.

Also, tier 2 seem to be of rather minor benefit from my limited understanding, so even if it's missing it's not as if we're going to miss it much.

This I'm unsure about, in the build presentation it sounded like their was decent performance benefits, or was that talking about if it was only software based?
I don't see why AMD bothered releasing a gcn 1.1 part otherwise to be honest.

Can't remember the name of the co-presenter but they said they were going to do some performance benchmarks soon after, and they never seemed to put it up on their site like they said they would.

Wonder if their able to move cpu resources over to the game partition over time with optimisation, ore if it would have to be a full core?
If not, if they find spare capacity and make some standard libraries resident on the OS side of cpu, something like FMODD etc.

They sound confident on gpu processing power, cpu sounds a bit more tighter. Which I always thought would be the case.

 

[pjbliverpool]

Most interesting new points to me:

-Talk about why ESRAM was chosen.

-Mention 8GB was in mind early on (contrary to the fact that reliable sources say the spec was 4GB as late as late 2011)

-They brush aside a ESRAM latency question. Hmm, at first blush, not a good sign for latency benefits.

-Confirms there's no major customizations to the Jaguar CPU's.

-They paint a picture of higher bandwidth vs more CU's for compute, and claim to not know how the battle will turn out.

Indeed, they do seem to imply that esram was chosen to allow for 8GB of main memory within a reasonable cost/power envelope while maintaining reasonable bandwidth as opposed to any latency specific advantages.

The confirmation of sea islands architecture for both (strangely) consoles was also interesting. Would love to hear Dave B's take on that considering Sea Islands isnt officially an IP designation.

Im also loving the (finally) clear explanation around the esram bandwidth. We now have a real peak figure, an explanation for the missing write cycle and a believable sustainable bandwidth utilisation rate.

Its really great theyve been so honest.

 

[expletive]

Whereas we've said that we find it very important to have bandwidth for the GPGPU workload and so this is one of the reasons why we've made the big bet on very high coherent read bandwidth that we have on our system.

I actually don't know how it's going to play out of our competition having more CUs than us for these workloads versus us having the better performing coherent memory.

Is this the ESRAM 109GB/s figure or the 30GB/s figure or something else?