Predict: The Next Generation Console Tech

[anexanhume]

I believe that TSMC is only just now starting to see their supply keeping up with demand and they're still ramping up production.


Where is that quote from and is it reliable? I don't know who seronx is....

A poster from techpowerup forums. His latest post on next gen consoles was brought up on page 606 of this thread and dismissed out of hand. If we're going to assume a large APU, we should consider Aegis as the source, and he's never given mm^2 specifics.

 

[AlNom]

PD-SOI is a cost adder that has only partly justified itself for high-margin x86 (and not particularly strongly in terms of performance or power-efficiency these days), and AMD is doing what it can to get to bulk.
SOI adds to wafer costs and engineering costs over bulk, and its history is poor even for chips targeting markets with higher price points.

The drama with AMD and Globalfoundries should have made the idea of building a high-yield low-margin console device a rather uncomfortable proposition for the console makers.

Just a random thought... who has eDRAM capabilities and on what process?

:smile2:

 

[MarkoIt]

A poster from techpowerup forums. His latest post on next gen consoles was brought up on page 606 of this thread and dismissed out of hand. If we're going to assume a large APU, we should consider Aegis as the source, and he's never given mm^2 specifics.

I read his post, I agree with you

 

[Bagel seed]

A poster from techpowerup forums. His latest post on next gen consoles was brought up on page 606 of this thread and dismissed out of hand. If we're going to assume a large APU, we should consider Aegis as the source, and he's never given mm^2 specifics.

From what I've seen most of what he posts as far as complete system specs goes are just speculation, but he does seem to have inside knowledge into some AMD components.

Here's the link to his post in Nov
http://forums.anandtech.com/showpost.php?p=34281744&postcount=188

 

[anexanhume]

Just a random thought... who has eDRAM capabilities and on what process?

:smile2:

The thought of converting GCN design to 32nm SOI for the sake of eDRAM begs an analogy I am yet waiting for humanity to invent that is sufficient.

From what I've seen most of what he posts as far as complete system specs goes are just speculation, but he does seem to have inside knowledge into some AMD components.

Here's the link to his post in Nov
http://forums.anandtech.com/showpost.php?p=34281744&postcount=188

Perhaps he does an inside AMD line, but his previous line about Nvidia getting the design win for the GPU is simply incongruous with it being an AMD APU. Can you imagine AMD designing an APU with Nvidia's technology on it? No way they'd sign up to that.

 

[Helmore]

Just a random thought... who has eDRAM capabilities and on what process?

:smile2:

Good question. Anyone got an answer to that? I only know of IBM and I don't know to what extent they're sharing any of their process technology.

@anexanhume - We'd also have heard more about NVIDIA being involved in any of the consoles if it was the case.

 

[Bagel seed]

Perhaps he does an inside AMD line, but his previous line about Nvidia getting the design win for the GPU is simply incongruous with it being an AMD APU. Can you imagine AMD designing an APU with Nvidia's technology on it? No way they'd sign up to that.

Could be a mix a little truth into a lot of lies kind of thing to keep NDA demons away. But I agree with you.

Anyways, the possibility of a 450 mm² chip is a good talking point.

 

[JasonLD]

According to the rumors, it is likely that the Durango may feature eDRAM, while Orbis doesnt. Especially if Durango goes lower bandwidth 8GB while Orbis does high speed 4GB.

 

[XpiderMX]

Another ~450 mm² chip before Kryptos? Kepler?

 

[anexanhume]

So I asked @aegis about foundry for Durango:

https://twitter.com/aegies/status/290920369173450752

i thought rumors were pointing to global foundries but that was ages ago.

I'm keeping my hope alive for FD-SOI

 

[JasonLD]

Arent GF, TSMC, and IBM only options available?

 

[3dilettante]

Just a random thought... who has eDRAM capabilities and on what process?

:smile2:

For SOI, it would be IBM, and there were announcements that GF gained the capability. I'm not sure what products have been built for IBM at GF using that tech.

The really well-known chips that have SOI and eDRAM from IBM are a bit pricey.

 

[fehu]

you can put a lot of things in a 450mm die

 

[anexanhume]

you can put a lot of things in a 450mm die

Indeed, and the question of why would you not want more GPU if you had that kind of space is a very interesting question.

 

[MrFox]

Indeed, and the question of why would you not want more GPU if you had that kind of space is a very interesting question.

My three uneducated guesses would be Cost, Wattage, and the 450mm2 die is a pipe dream?

 

[scently]

Errr I think you guys need to calm down a bit. First of all how sure are you that the chip is 450mm2? if it does, how sure are you that the GPU isn't more powerful than the 1.2. teraflop been bandied around? We need more info before we jump to conclusion.

 

[Shifty Geezer]

Indeed, and the question of why would you not want more GPU if you had that kind of space is a very interesting question.

See the thread on custom hardware. Custom silicon can provide greater efficiencies for a given task than programmable hardware.

 

[Helmore]

Arent GF, TSMC, and IBM only options available?

For what exactly? There's also Samsung and UMC....

 

[MrFox]

Does it all give more credibility to the rumored eSRAM then?

If IBM's eDRAM is so finicky to produce, maybe they decided it's better to use a ram cell that can be made with a normal process instead, so they can put it inside the SoC from the start (versus never for the 360). Maybe it's worth the size tradeoff?

 

[anexanhume]

Does it all give more credibility to the rumored eSRAM then?

If IBM's eDRAM is so finicky to produce, maybe they decided it's better to use a ram cell that can be made with a normal process instead, so they can put it inside the SoC from the start (versus never for the 360). Maybe it's worth the size tradeoff?

I would think so. Either port it to your process of choice for integration or make a different choice. Integrating everything on one SoC has to be extremely appealing for a ton of reasons.