Predict: The Next Generation Console Tech

Status
Not open for further replies.
Even Anand - who so couldn't be arsed that they didn't even list overclocking voltage other than to say "less than 10% extra" - got to 4.8 gHz under load with a total power draw of 294W at the wall. That's only 40W above Tech Power Up (and likely with higher voltage).

I'm not sure, but it seems that there is a thread of thought going on that the CPU may be overclocked, perhaps undervolted to keep power consumption/heat down?

I don't think this is feasible as cost need to be reduced, and yields need to go up, so "binning" isn't feasible to get cores that run acceptably at lower volts (such as LV/ULV bins). Historically, the opposite was actually true, to get yield rates higher, it's common to reduce frequency, and in some cases cores are deactivated to increase yields (Cell has 8 SPE's, 1 deactivated to increase yields).
 
Recent rumors say there are 3 unique features MS has request be added to the CPU/GPU. Likely audio DSP, esRAM or EDRAM, and an unknown third thing.

Thinking back to the Xenos gpu, I remember that there were 3 special features that were added to that: unified shaders, edram, and a hardware tessellation unit. ATI went on to implement 2 of the 3 in the next GPU designs. And IIRC, the tessellation unit in ATI/AMD's discrete GPUs was for the most part unused by PC games.

I was trying to think what this 3rd hardware feature could be, then I remembered that in AMD's 7000s series cards they had a new piece of hardware that will be used for their Partially Resident Textures technology, which is based on the ideas in John Carmack's Megatexture technology. I don't think there are any games out there that even use this feature, so why would AMD dedicate die space for it (similar to the tessellation unit in the R600)?

If Orbis is using Southern Island GPUs, it will have this as well, but it's currently limited to 32k textures. What if a more robust version of hardware PRT is the third item that MS wanted implemented in their GPU?
 
Partially resident textures are a coarser representation of GCN's general memory addressing capabilities and ability to use x86 page tables, to better fit the constraints of being an expansion board accessed with the current PC software models.

The die area was already spent because GCN's architecture can do much more.
 
I'm not sure, but it seems that there is a thread of thought going on that the CPU may be overclocked, perhaps undervolted to keep power consumption/heat down?

I don't think anyone is suggest that. As a console only has one spec of chip off a production line multiple bins and under/over clocking don't apply. The issue is about how suitable Piledriver based processors might be in a console. I don't think they'd be bad - certainly preferable to another Cell or Xenon.

Talk about overclocking and undervolting is really about the limits of existing processors. There's more headroom yet to Bulldozer derivatives in terms of clock or power (or perhaps both) so a console may be able to exceed the performance per watt of existing Piledriver chips. Trinity is the one to look at though as it doesn't have the L3 and is optimised for ppw over peak performance. Richland should be available for reviewers in a few weeks, and will hopefully improve on the already pretty good Trinity.
 
I don't think there are any games out there that even use this feature, so why would AMD dedicate die space for it (similar to the tessellation unit in the R600)?

You mean besides Rage?

Partially resident textures are a coarser representation of GCN's general memory addressing capabilities and ability to use x86 page tables, to better fit the constraints of being an expansion board accessed with the current PC software models.

The die area was already spent because GCN's architecture can do much more.

Would you say more games will then use that feature to fully exploit the architecture now, especially since the optimization would carry between both consoles being based on GCN?

It's other optimizations like this I am curious about developers being forced into that turns out to reverberate back into PC games and change how they are developed too.
 
Last edited by a moderator:
Thinking back to the Xenos gpu, I remember that there were 3 special features that were added to that: unified shaders, edram, and a hardware tessellation unit. ATI went on to implement 2 of the 3 in the next GPU designs. And IIRC, the tessellation unit in ATI/AMD's discrete GPUs was for the most part unused by PC games.

ATI's Radeon 8500 in 2001 already had a hardware tessellation unit....
 
You mean besides Rage?

Would you say more games will then use that feature to fully exploit the architecture now, especially since the optimization would carry between both consoles being based on GCN?

It's other optimizations like this I am curious about developers being forced into that turns out to reverberate back into PC games and change how they are developed too.

PRT's are interesting for any game that streams textures.
If it's there on a primary platform it'll likely get used.
How that gets ported onto a PC which may or may not support it will depend on the developer.
 
PS4 roadmap (from VGLeaks):

Hi everybody and Happy New Year !!!


We expect many great things for the new year.



To inaugurate the new year we have some new info about the next generation console manufactured by Sony: Orbis.


In this case we have the earlier roadmap.


2011 June/July – PC with Win7 64-bit and a “jailbroken” ATI r9 graphics card
2011 Sept/Oct – PC with Win7 64-bit and a “jailbroken” ATI r10 graphics card
2012 Q1 – disclosure to more developers
2012 E3 – potential unveiling window start
2012 Jul – devkits for engineers writing OS
2012 Q3 – first true hardware prototype devkits
2013 E3 – potential unveiling window end
2013 Q4 – launch


Soon we will put new interesting info about Orbis and Durango.
 
Last edited by a moderator:
What the hell are an ATI r9 and r10?

Maybe 9th gen and 10th gen Radeon but that's pretty stupid, you have to selectively omit some half-gens, or start the count at a certain generation such as 9700 pro. And I'd start at Rage pro anyway.

Forget that, "jailbreaking" a graphics card is totally nuts (or did Sony unlock Radeons into FireGL to build dev kits, riiight.)
 
What the hell are an ATI r9 and r10?

Maybe 9th gen and 10th gen Radeon but that's pretty stupid, you have to selectively omit some half-gens, or start the count at a certain generation such as 9700 pro. And I'd start at Rage pro anyway.

Forget that, "jailbreaking" a graphics card is totally nuts (or did Sony unlock Radeons into FireGL to build dev kits, riiight.)

r9 is probably R900, or Northern Islands in other words (Radeon HD6000), and r10 is probably R1000, which is known as Southern Islands (Radeon HD7000). That said, 2011 Sept/Oct for a Southern Island based card seems pretty early. Did AMD have any Southern Island based silicon back by then? I don't think TSMC's facilities were ready to start any production on 28 nm at that time.

As for 'jailbreaking', that might just be from poor translation or something. Or the entire thing is bogus....
 
r9 is probably R900, or Northern Islands in other words (Radeon HD6000), and r10 is probably R1000, which is known as Southern Islands (Radeon HD7000). That said, 2011 Sept/Oct for a Southern Island based card seems pretty early. Did AMD have any Southern Island based silicon back by then? I don't think TSMC's facilities were ready to start any production on 28 nm at that time.

As for 'jailbreaking', that might just be from poor translation or something. Or the entire thing is bogus....

I assume they're referring to firmware unlocked cards or something. Disabled some CU's or reset stock clocks or something along those lines.
 
I assume they're referring to firmware unlocked cards or something. Disabled some CU's or reset stock clocks or something along those lines.

Seems plausible, or actually very very likely. They're using engineering samples anyway.

That said, isn't the timeframe for their Southern Islands card a bit odd? 2011 Sept/Oct for a 28 nm HD7000 card. Anyone got some reliable information how when AMD got their first samples back from TSMC?
 
Status
Not open for further replies.
Back
Top