Predict: The Next Generation Console Tech

Status
Not open for further replies.
Charlie from S|A managed to take a picture of Radeon 8950. According to him, cards are already ready and are waiting for clearing of stock/Q2 marketing push. [he's hiding pictures behind pay wall]

If 8xxx series is already ready, it would be bad if PS4 has 7xxx [old vgleaks rumor].
 
He forgot it will make you toast. And -this is whats revolutionary- automatically butter it for you.

2*(8cores*4threads) might sound a bit over-the-top at first glance; but if MS threw enough money at IBM, they'd certainly be very capable of pulling that off within a reasonable die-size and power envelop.

High-frequency-optimized 8core Power7 (with 4 threads per core) launched at around 570mm² @45nm.

Sacrifice some peak fequency over transistor density, tweak the design a little more, move production to 20nm - and an adapted, customized version of that chip should be possible somewhere around the 110-120mm² mark. Lots of space left to add a reasonably sized GPU and some other stuff without making one final MCM much bigger than 200mm².

Would be a very long shot and entail a lot of risk - but it's certainly not as silly as it might seem.
 
Charlie from S|A managed to take a picture of Radeon 8950. According to him, cards are already ready and are waiting for clearing of stock/Q2 marketing push. [he's hiding pictures behind pay wall]

If 8xxx series is already ready, it would be bad if PS4 has 7xxx [old vgleaks rumor].

Isn't AMD APU been like a generation behind anyway. But the GPU that's going to be paired with either Steamroller or Jaguar, is going to be the 8xxx line won't it ?

I'm more curious about GDDR5 rumour. Have Sony given up on the 1024-bit Wide I/O desktop part and just going with GDDR5 ? I thought JEDEC is finishing off the specs around now ?
 
Just a heads up - we've another removed poster for going OT, and the likes of N2O, Rangers and (((interference))) are in the ironsights if they think bickering about who's rumours are best is worthwhile contribution to this thread.
 
Lol ok, (this is despite bgassassin being verified by lherre) but what about the ex-MS engineer who has indicated all reports of a large GPU are bs? And that we know that the pastebin rumour back in Dec 2011 has the same number? Or the fact lherre has said the 720 has the weaker GPU of the pair?

Who said the die was 450 mm^2?

The only I've seen post in this thread I trust is bkilian because he has been carefully vague and non committal on any number specifics. His wording leans towards low 1 TF numbers, but nothing exact. For others, I have to trust an unknown person that their unknown source is good. Sweetvar and pastebin info are over a year old now. That's enough time for it to have changed, especially if that info was correct and they feel Sony was reacting to it (like the rumors they upped their own memory in reaction to Durango).

Aegis is a real industry person with an actual reputation on the line. He has reason to have multiple industry sources. I'm not going to take his word as gospel, but I am going to allow it to give me pause about the existing rumors.

Just yesterday we received rumors that the ''main gpu" is "currently" 12 CUs at 800 MHz. That leads credence to the 1.2TF numbers, but it could just be building on it. What it does show us is that there's another block in there worthy of being called a GPU of some sort that no one has numbers on, at all. That would prove these 1.2TF rumors to be incomplete at best, and more likely they're based on dev kits than final silicon.

Right now, I believe a ~1.2TF sea islands GPU is most likely, but I have eason to believe that may be wrong or we may be significantly misunderstanding the actual power because of this second GPU.
 
The only I've seen post in this thread I trust is bkilian because he has been carefully vague and non committal on any number specifics. His wording leans towards low 1 TF numbers, but nothing exact. For others, I have to trust an unknown person that their unknown source is good. Sweetvar and pastebin info are over a year old now. That's enough time for it to have changed, especially if that info was correct and they feel Sony was reacting to it (like the rumors they upped their own memory in reaction to Durango).

Aegis is a real industry person with an actual reputation on the line. He has reason to have multiple industry sources. I'm not going to take his word as gospel, but I am going to allow it to give me pause about the existing rumors.

Just yesterday we received rumors that the ''main gpu" is "currently" 12 CUs at 800 MHz. That leads credence to the 1.2TF numbers, but it could just be building on it. What it does show us is that there's another block in there worthy of being called a GPU of some sort that no one has numbers on, at all. That would prove these 1.2TF rumors to be incomplete at best, and more likely they're based on dev kits than final silicon.

Right now, I believe a ~1.2TF sea islands GPU is most likely, but I have eason to believe that may be wrong or we may be significantly misunderstanding the actual power because of this second GPU.

I said it!. One pseudo-gpu to render lighting + traditional GPU!.
 
I said it!. One pseudo-gpu to render lighting + traditional GPU!.

That doesn't seem quite flexible enough to me. I'm presuming that whatever special sauce they add, it will be able to aid in more than just one task. Something more general purpose in other words. That's just my guess of course.
 
I personally trust bkilian. If bkilian who was an ex-MS employee pretty much indirectly confirmed that we shouldn't expect a 2-2.5tf gpu inside Durango much less a 3tf gpu, I think that pretty much leaves us with something like a 1.2tf gpu since he didn't like my 1500mm suspension travel bike tire either.
At this point I think people can narrow things down a bit and neglect wild claims a little easier. You an throw ideas around but that doesn't mean you shouldn't keep things consistent and apply logic to them.
 
Lol ok, (this is despite bgassassin being verified by lherre) but what about the ex-MS engineer who has indicated all reports of a large GPU are bs? And that we know that the pastebin rumour back in Dec 2011 has the same number? Or the fact lherre has said the 720 has the weaker GPU of the pair?

Who said the die was 450 mm^2?

I don't recall lherre verifying me, though at the same time I don't see why he would or would need to. Like I've mentioned before I just pass along what I think maybe ok to pass along and is worth discussing. I will add that last year one consistent thing I heard was that the overall performance between PS4 and Xbox 3 was debatable as to which one was better. There will probably be things PS4 does better than Xbox 3 and vice versa. I think things are still pointing in this direction and it will boil down to consumer preference.

I would also like to know where the large die info came from. Just for curiosity purposes.
 
I don't recall lherre verifying me, though at the same time I don't see why he would or would need to. Like I've mentioned before I just pass along what I think maybe ok to pass along and is worth discussing. I will add that last year one consistent thing I heard was that the overall performance between PS4 and Xbox 3 was debatable as to which one was better. There will probably be things PS4 does better than Xbox 3 and vice versa. I think things are still pointing in this direction and it will boil down to consumer preference.

I would also like to know where the large die info came from. Just for curiosity purposes.

I think there were three different rumors. 1. Charlie claimed it was a large APU, 2. aegis on GAF said it was a large APU, and 3. a Chinese forum posting claimed it was a large APU
 
That doesn't seem quite flexible enough to me. I'm presuming that whatever special sauce they add, it will be able to aid in more than just one task.
Lighting is such a huge job and very demanding potentially for high quality GI that I could see value in dedicated silicon to pull it off, but I don't know what such silicon would look like or how it'd fit into the render pipeline (it'd be easy to use in a deferred renderer).
 
GAF surfaced this patent filed by SCEA for a system for dynamic context switching between architecturally distinct graphics processors: http://goo.gl/m7NB3
Old patent aimed at mobile computing, switching graphics between two different graphics architectures for low-power and high-power activities. It's been discussed in this thread. I rate its relevance to PS4 as 0.5%. ;)
 
Did any of them give that specific number?

Now, I remember there was that Serenox poster that claimed a 450mm APU. That's the only number I remember, but he also claimed stuff like an NVIDIA GPU and that both Sony and MS have basically the same hardware. I don't believe there's any validity to his rumor.
 
Shifty Geezer said:
Old patent aimed at mobile computing, switching graphics between two different graphics architectures for low-power and high-power activities. It's been discussed in this thread. I rate its relevance to PS4 as 0.5%.

The patent was filed by SCEA though. Plus they describe a console kind of system, with Cell cpu, VRAM, etc.

Assuming there is APU + GPU, unless the APU-GPU is completely separate from communicating with the main one, I'm sure there exists something like this patent where they're both aware of each others state and could theoretically at least share tasks if not work on rendering stuff together ala SLI.

Edit: with HSA it's not really out of the picture anyways. So we might see it happen even if the patent has nothing to do with PS4
 
Last edited by a moderator:
Old patent aimed at mobile computing, switching graphics between two different graphics architectures for low-power and high-power activities. It's been discussed in this thread. I rate its relevance to PS4 as 0.5%. ;)

Ah, yeah, it's a continuation of a 2009 patent. Sorry about that.
 
The patent was filed by SCEA though. Plus they describe a console kind of system, with Cell cpu, VRAM, etc.

Assuming there is APU + GPU, unless the APU-GPU is completely separate from communicating with the main one, I'm sure there exists something like this patent where they're both aware of each others state and could theoretically at least share tasks if not work on rendering stuff together ala SLI.
Claim 1 specifies explicitly a just-in-time compiler for sending GPU instructions from CPU to one or other GPU. It's explicit in intent and not related to GPU communication in any other form. An AMD APU and GPU will be architecturally similar too and not want a JIT compiler.
 
Claim 1 specifies explicitly a just-in-time compiler for sending GPU instructions from CPU to one or other GPU. It's explicit in intent and not related to GPU communication in any other form. An AMD APU and GPU will be architecturally similar too and not want a JIT compiler.

How I read it to put it simply is it just switches on rendering between 2 GPUs depending on certain situations. And claim 1 gives the provisions for it to translate instructions if the architectures are distinct. But if they aren't that'll work too.

It's not unlike what a laptop would do with switching between 2 GPUs (but done better), but also open to perform like SLI.
 
Status
Not open for further replies.
Back
Top