Xbox One (Durango) Technical hardware investigation

[XpiderMX]

Okay, dare I ask? What are these 15 special-purpose processors?

More importantly, where are the DMEs?

The audio chip?

 

[Scott_Arm]

Okay, dare I ask? What are these 15 special-purpose processors?

More importantly, where are the DMEs?

DMEs are there. Swizzle copy and Swizzle LZ

 

[Rockster]

At least we have enough solid info now to update die size thread for Xbox One.

 

[eastmen]

that is a tiny chip all things considered , on 22nm its going to be cheap as heck to make.

 

[GrimThorne]

DMEs are there. Swizzle copy and Swizzle LZ

Oh, they actually gave them designations. Now I see them.

 

[XpiderMX]

Relevant?

Speaking at the Hot Chips engineering conference at Stanford University, Microsoft chip architect John Sell said the Xbox One’s main processor is a gargantuan beast, co-designed with Advanced Micro Devices and manufactured by contract chip manufacturer Taiwan Semiconductor Manufacturing Co.

The chip covers an area of 363 square millimeters and is built in a 28-nanometer state-of-the-art manufacturing process. It has 5 billion basic components, known as transistors, on a single chip that combines a microprocessor, graphics processor, and 47 megabytes of memory.


As one member of the audience remarked, that mean it’s a big chip. Such chips are hard to manufacture, as any small defect can render a chip inoperable. And the larger a chip, the more prone it is to small defects.


To deal with that, Sell said Microsoft designed in redundancies into the chip where it could, so that if one section was inoperable, it doesn’t kill the whole chip. Nevertheless, it means that it won’t be easy to churn out millions of such chips.

To make the chip more power efficient, Microsoft designed small islands and gates within the chip that could switch off certain areas so the overall chip consumes less power. Such techniques are standard in big chip projects these days, and they mean that only the parts of a chip that are being used are drawing power.

The central processing unit (CPU) has eight 64-bit Jaguar cores designed by AMD.

The graphics section of the main processor has 15 special processors that can handle things like graphics or physics processing. Data highways can transfer bits at a rate of 200 gigabytes per second or more across different parts of the chip. The audio subsystem has two dedicated vector cores.

Microsoft disclosed some details, but left many important pieces out. Evidently, Microsoft doesn’t want to tell all of its competitors about how well designed its system is.

http://venturebeat.com/2013/08/26/m...se-details-are-critical-for-the-kind-of-expe/

 

[Exophase]

Did you guys figured out the 47MB figure ?
I can only end up with 42480kb, where's the 4MB I'm missing?

12x CU:
- 4x64kb vector registers
- 64kb local store
- 8kb scalar registers
- 16kb L1
== 4128kb

3x groups of 4 CU:
- 16k icache
- 32k vector cache
== 144kb

4x memory controllers L2 128kb:
== 512kb

2x jaguar CU L2
== 4096kb

- 8x jaguar cores
- 32kb icache
- 32kb dcache
== 512kb

- 32mb esram

- 64kb GDS

- 240kb audio

At the very least, I expect all L1 caches and GPU registers have an extra bit per byte for ECC. Then add some more for redundancy.

 

[bagofsuck]

The graphics section of the main processor has 15 special processors that can handle things like graphics or physics processing.

i want to know which processors are they talking about,did we know about this?

 

[Kaotik]

Wait what? The "GPU side" has 15 "processors" for graphics & physics etc? That doesn't fit the leaks at all there. I suppose you could have 15 CUs in there (or 16 with 1 for redundancy?) but like said, doesn't fit the leaks at all

 

[Scott_Arm]

http://venturebeat.com/2013/08/26/m...se-details-are-critical-for-the-kind-of-expe/

One thing i noticed in this link is that the Kinect v2 latency is 20ms. From VGleaks we see that with processing you get 60ms of latency. So, assuming that is correct, 40ms in latency for processing the depth buffer and doing the skeletal tracking is what I'd take from that. That's interesting to me because it means you're unlikely to get a Kinect 3 mid-gen that would improve latency (maybe a few ms?). It also means if they put out a Kinect 3 for Xbox Two, latency could be very small, because you'd have more processing power to deal with it. I'll also be interested in seeing what kind of latency you could get on a PC using Kinect v2. I think with a high-end PC it could be very respectable.

 

[BRiT]

I suppose you could have 15 CUs in there (or 16 with 1 for redundancy?)

No.

 

[Scott_Arm]

Wait what? The "GPU side" has 15 "processors" for graphics & physics etc? That doesn't fit the leaks at all there

The audio block alone has four or five processors, I think. Then you have all the video encode/decode features, and the DMEs that do some swizzle and LZ functions. Those could count as well. I doubt there is anything that was not already included in vgleaks.

 

[Kaotik]

The audio block alone has four or five processors, I think. Then you have all the video encode/decode features, and the DMEs that do some swizzle and LZ functions. Those could count as well. I doubt there is anything that was not already included in vgleaks.

So the Venture Beat guy just understood something wrong and it's GPU + 15 other processors

 

[McHuj]

Add up the blocks on the GPU slide. There's 15. I think venture beat just screwed up and assumed the 15 processors refers to the blocks on the GPU slide.

 

[liolio]

It seems to say 8mb in each block.
So the digital foundry rumor had some truth after it after all.
But from what i remembered df was talking about a max of 192Gb/s.

Found an tweet on gaf.


https://twitter.com/Daniel_Bowers/status/372040673618182144

Full speed L2?

 

[3dilettante]

The uncore bandwidth to the CPU section does look to be higher than some other Jaguar chips. Tweaking the uncore and L2 interface for higher bandwidth might be the reason.

 

[DrJay24]

So from the slides, 2 compute and 2 graphics command processors per CU. Is this normal GCN 1.0 after all?

 

[Michellstar]

"As one member of the audience remarked, that mean it’s a big chip. Such chips are hard to manufacture, as any small defect can render a chip inoperable. And the larger a chip, the more prone it is to small defects.


To deal with that, Sell said Microsoft designed in redundancies into the chip where it could, so that if one section was inoperable, it doesn’t kill the whole chip. Nevertheless, it means that it won’t be easy to churn out millions of such chips."


Probably there are 14 CUs in there, and the 4 missing MB of sram...

 

[MrFox]

"As one member of the audience remarked, that mean it’s a big chip. Such chips are hard to manufacture, as any small defect can render a chip inoperable. And the larger a chip, the more prone it is to small defects.


To deal with that, Sell said Microsoft designed in redundancies into the chip where it could, so that if one section was inoperable, it doesn’t kill the whole chip. Nevertheless, it means that it won’t be easy to churn out millions of such chips."


Probably there are 14 CUs in there, and the 4 missing MB of sram...

Counting disabled parts wouldn't be very honest.

 

[BRiT]

No. They are not talking about unused bits.