Xbox One (Durango) Technical hardware investigation

[(((interference)))]

DaE has said the Durango specs people have are old, and come from the same European source. He said this as recently as March 20.

https://twitter.com/DJoe1899/status/314369969154584576

but he also said this today

https://twitter.com/RogueWave09/status/316946699505524736

so i dont even know what he's saying.

I asked Richard about DaE back in Feb:

With your recent Durango vs Orbis article, do you know why, if DaE is vgleaks source he constantly tweets that the vgleaks specs are old/fake and denies that Durango is less powerful than Orbis?

Re DaE, I think he’s having some fun with you.

DaE loves trolling the fanboys for teh lolz.

Try to resist getting sucked in.

 

[Rangers]

We all know he's a troll. So far he's never really said anything later proven to be blatantly false though, I dont think.

But I'm not sure if he's saying the Vgleaks specs as of now are accurate or inaccurate, per the last tweet I posted.

 

[dagamer]

Disseminating information isn't a crime. The only law stopping leaks is contract law on NDAs, where those found leaking information can be punished (sacked and effectively banned from the industry I believe. Dunno if there are greater reprimands like fiscal penalties). If someone gives me info, I'm free to pass it on. That's true of all information except that covered by official secrets acts or that illegally obtained by spying, so if I get info on you by a snoop bugging your phone, you could block me spreading that info I believe.

Otherwise, rumours are rumours. Word gets out (even government info like Wikileaks). As a company you can't really stop it. You can try and pressure a media outlet to withhold info, but info isn't IP that can be stopped via copyright laws. I guess logos can be, but if you hit a website with a cease and desist order, the internet will already have the image and they'll know that it's true because it's blocked. So instead you do nothing and leave the uncertainty of whether it's valid or fake.

That's rather unlikely thinking. Companies don't comment on rumour and speculation. It's equivalent to never giving in to terrorist demands. If they ever could be goaded into saying something, media could use rumour and speculation to extract real information. Ergo PR is controlled by the company. They'll release what they want to release when they want to release it. Regardless of whatever rumours float around, the moment the real information is released, that info becomes immaterial, so don't get goaded into saying to the public now, "we have always on by design but that doesn't mean you need an internet connection to play game..." when you can say that when you're ready with your whole message and set all those rumours to rest.

Don't forget the Striesand Effect: that which you go to great lengths to hide or remove from the Internet only publicizes it more. Besides, the only company they are honestly trying to hide this stuff from is Sony, and there are far better ways of figuring out what the other side is doing (via major 3rd party developers). Pretty obvious this time around they all wanted more memory, and unified memory at that.

In the end, leaks of raw hardware are boring. Leaks of software and business strategy are far more damaging.

 

[(((interference)))]

We all know he's a troll. So far he's never really said anything later proven to be blatantly false though, I dont think.

But I'm not sure if he's saying the Vgleaks specs as of now are accurate or inaccurate, per the last tweet I posted.

If he's saying the 720 is more powerful than the PS4, then that's blatantly false.

Are you sure you aren't just hoping that the vgleak specs are wrong, and so are willing to entertain his sly hints that, in fact, that is the case?

FWIW, I was asking Richard whether he has heard anything about last minute clock changes etc:

Nothing at all – aside from wishful thinking from GAF – about spec improvements.

No time to re-architect the hardware, and we’re dealing with one of the biggest APUs ever made – the clocks CPU-side are low for a reason.

So it looks like vgleaks is probably what we're going to get.

 

[McHuj]

Interesting. If he's saying that its one of the biggest APU's ever made, I'm going to guess that the embedded RAM is really standard SRAM and not some eDRAM variation. I doubt the current rumored specs and eDRAM would put in that size category.

 

[(((interference)))]

Interesting. If he's saying that its one of the biggest APU's ever made, I'm going to guess that the embedded RAM is really standard SRAM and not some eDRAM variation. I doubt the current rumored specs and eDRAM would put in that size category.

Well, on the Wii U the eDRAM already takes up most of the die:

And for Durango you'll have the move engines, audio blocks, hardware video encoder/decoders etc.

 

[AlphaWolf]

WiiU

LOL

 

[Rangers]

Interesting. If he's saying that its one of the biggest APU's ever made, I'm going to guess that the embedded RAM is really standard SRAM and not some eDRAM variation. I doubt the current rumored specs and eDRAM would put in that size category.

Sure they would! Unless I'm missing something... A10 5800K has 384 SP's, Durango has 768. Plus 32 MB of EDRAM, whatever size.

Sure the CPU side might be way bigger on the 5800k, but still.

I kinda think that's just Richard's speculation though. Bobcat's werent clocked very high.

Edit: well Google says the Trinity is 246mm. Cape Verde 123mm^2.

Still you're dealing with 12 Cu's not 10 on Durango, dual setup engines instead of one, the ESRAM, and other modifications. All told, dunno, but guessing ~300mm^2 for the whole APU.

Well, on the Wii U the eDRAM already takes up most of the die:

True, but it's on 40nm (at best, even mild speculation of 55nm I believe) vs 28nm for Durango.

So Durango's 32MB of EDRAM should be hopefully ~1/2 the size of Wii U. The Wii U's EDRAM was measured, IIRC it was ~46mm^2 or something. So 32MB in Durango at the same density could be close to 20mm^2.

I love the idea of EDRAM at these nodes. Didn't love it in Xenos at 90nm, but love it now, it makes much more sense at 28nm. The amount needed has scaled far less than the nodes have, so it's relatively much cheaper. You will save a lot more by going with DDR3 instead of GDDR5.

 

[(((interference)))]

Oh, it could well be 6T SRAM, all i'm saying is that eDRAM takes up a lot of die space as it is.

 

[AlphaWolf]

Oh, it could well be 6T SRAM, all i'm saying is that eDRAM takes up a lot of die space as it is.

The way you're backing the leak, you're backing eSRAM. Or are you now claiming they could be wrong?

 

[(((interference)))]

The way you're backing the leak, you're backing eSRAM. Or are you now claiming they could be wrong?

No, it's definitely described as ESRAM by MS in their docs (as vgleaks has).

I'm just not sure whether it's 1T or 6T SRAM (suspect the latter, as it's supposedly very low latency)

 

[Rangers]

Oh, it could well be 6T SRAM, all i'm saying is that eDRAM takes up a lot of die space as it is.

yeah, but i'm saying 32mb basic edram probably wont at 28nm judging by the wii u.

really no knowing till we see the size of the durango die, and/or chipworks shots of it.

The way you're backing the leak, you're backing eSRAM. Or are you now claiming they could be wrong?

from what i gather the definition of esram and the lines between it and edram are fuzzy and can be semantics.

ms calls it esram in the documentation apparently, so thats why we do. but it could be essentially edram by another name.

6t esram seems to be the one that could take a lot of space, specifically, but we dont know if that's what's in there.

 

[AlphaWolf]

The only semantics are mosys calling their edram implementation 1T SRAM.

 

[(((interference)))]

Also, we don't know how much space the four DMEs, audio cores and whatever else is in there take up.

I think it's likely to be 6T SRAM, as if it was just 1T SRAM they could have put in more than 32MB, even Wii U has 32 MB of it (eDRAM is similar in cost to 1T SRAM apparently).

Does anyone know the latencies of eDRAM vs 1T SRAM?

 

[Rangers]

"could have" is irrelevant. 32MB is probably the sweet spot for 1080p.

Isn't 6T ESRAM huge? EG, 6X as big as EDRAM?

At some point I'm pretty sure that doesn't make cost sense. The only way the whole DD3/small pool RAM works is as a cost play. Otherwise just use GDDR5.

 

[(((interference)))]

"could have" is irrelevant. 32MB is probably the sweet spot for 1080p.

Isn't 6T ESRAM huge? EG, 6X as big as EDRAM?

At some point I'm pretty sure that doesn't make cost sense. The only way the whole DD3/small pool RAM works is as a cost play. Otherwise just use GDDR5.

Except, Durango's ESRAM is not intended to be used for the framebuffer like in 360 but as a general purpose scratchpad.

I'm seeing it's 4x 1T ESRAM (which itself is 10-15% bigger than EDRAM on the die)
http://en.wikipedia.org/wiki/1T-SRAM

 

[Rangers]

That would seem to make the ESRAM as big as the whole GPU.

 

[AlphaWolf]

Not sure you can compare to the Quad density 1T SRAM which uses custom fabrication. The 6T SRAM is around double the size of the standard 1T SRAM.

I doubt you're looking at even 60mm2 for 6T SRAM.

 

[(((interference)))]

But the Quad density 1T SRAM is the one that's only 10-15% bigger than eDRAM? And it's half the size of regular 1T SRAM (0.14 vs 0.28 mm2/Mbit)

 

[Gubbi]

Except, Durango's ESRAM is not intended to be used for the framebuffer like in 360 but as a general purpose scratchpad.

That's how framebuffers are used nowadays. Fill a G-buffer and read it back as a texture. Render a Z buffer from a light's perspective and use it as a shadow buffer, etc.

All these techniques required the framebuffer to be resolved (exported to main memory) on the 360, not so on Durango.

Cheers