Predict: The Next Generation Console Tech

Status
Not open for further replies.
or get rid of the PPU and use x86 CPUs (cores) with the SPEs.

I can see the rationale behind this. it's like on cell phones where you have lots of silicon available but a power and bandwith wall, so this kind of processing is already appearing - CPU (that may use semi-heterogenous ARM cores, slower but with same instruction set), GPU, DSP with power shut down for elements unused at a particular time.

HSA would make this lower latency, finely grained and easier. but do we have so much silicon to waste on 28nm for a console that uses considerably more power and has a lot more bandwith, I'm not sure.
there would be power islands so unused SPE, or part of the GPU, or other hardware would be shut down but you would have to do this very fast, as in during the rendering of a frame.

capacitors or ultracapacitor could smooth the power use as I can imagine it being very jumpy.
 
Don't you also want to maximise utilisation though?
Yes, but it's a question of time. I guess once HSA works really well, maximizing utilization across all parts of the system won't be much of a problem. It's crucial to allow for a smooth transition, though.

No use having silicon on a board that many devs won't use. If you have an x86 CPU, what's the need for a 1PPU/4SPE module?
Yeah, maybe something like an updated version of Toshiba's SPURS chip would do. Probably no need for the PPE. The x86 CPU part seems like an important aspect of the cooperation with AMD. At the same time, it would be a pity to give up on the special capabilities of the SPEs.

Be that as it may: What I wanted to hint at is that, in the wake of HSA, we might see a return to more specialized hardware (at least to a certain extent). There might even be a case for certain parts of the Cell architecture to survive - if the average performance gain over using just "more of the same" outweighs the expenses of specialized silicon.

That's not to say you don't have a point concerning hardware redundancy. I'm just saying that (supposing you've got a rather hard limit of 400-500mm² of silicon to build your system from), it might make sense to dedicate a part of that overall space to more specialized sub-chips.

If 100mm² specialized logic + 200mm² CPU + 200mm² GPU is generally faster at the usual tasks running on your system than 250mm² CPU + 250mm² GPU, you might want to go with the former variant.

The main problem with more specialized hardware indeed seems to be putting it to good use - but isn't that exactly one of the core problems HSA is supposed to address (by making a system of diverse architectures easier to utilize and programm for)?


EDIT:
or get rid of the PPU and use x86 CPUs (cores) with the SPEs.
That's basically what I was driving at by referring to the SPURS chip - but I saw your more elaborate post only after finishing my reply. Sorry.
 
Last edited by a moderator:
Intel is quietly placing is marbles by the way, if you think of Sandy Bridge and its follow ups. there's an incredibly fast CPU in it, a GPU integrated into the memory hierarchy (that's somewhat more tight than current AMD Fusion), a DSP array (called "quicksync").

they even know how to make fifty-core CPU with smaller, high throughput cores. if they build up their version of HSA, they could integrate everything and make a monstrous chip. not that they are willing to make a console CPU, though.
 
Intel is quietly placing is marbles by the way, if you think of Sandy Bridge and its follow ups.
The fact that AMD is basically forced to open up their HSA push and closely work with partners might actually work in their favour in the long run, though. The more heterogeneous the architectures that are combined into a complex system get, the harder it will be for one company (as big as it may be) to keep up on all fronts.

Intel will probably still have some advantages with respect to efficiency of integration (given they can research and build most of the components in-house), but AMD might end up with the advantage of more specialized and efficient contributions from their partners.

It will be very interesting to watch.

For the time being, I'm most interested in how far SONY and Microsoft will respectively push the stacking of specialized hardware within their next gen consoles.
 
nice insight, this can explain why AMD had to suffered the pains of a slow and hungry bulldozer. the semi-disastrous chip was synthetizable, so they've made for themselves a basis for future HSA combination with "foreign" parts, if they get there.
they have an experience of a big, mass-produced and marketed chip as well.

I will be interested in what cooperation there will be between AMD and IBM, too.
 
I've said before having a block of 8 SPEs on the die with the x86 cores makes a lot of sense. You can emulate the PPE and GPU and get backwards compatibility, plus the SPEs are ideal to run a bunch of system level, but computationally intensive services like audio, decompression, video encoding for remote play to Vita and live internet broadcasting like the Vita does through Nico Nico, or Sony provided libraries for post processing and physics. Devs could get the benefit without every directly coding the SPEs, maintaining backwards compatibility would be awesome for users and 8 SPEs would be pretty small on 28nm.
 
This made me laugh :LOL:

66019497913364665701335371811EYYTYMYO6t5DKeUrNGxF.JPG


For those with bad eye sight

CPU - 22nm Cell @ 3.2Ghz with 16 PPE's and 128 SPE's
GPU - 22nm Custom Nvidia GPU @ 2Ghz based on Kepler GK104x2
RAM - 10Gb XDR2 + 10Gb GDDR6

What you guys reckon? 600-700w power consumption on full load?

Other little things like mandatory 1080p/30fps for all games....

EDIT : Changed GDDR6 amount :oops:
 
Last edited by a moderator:
I've said before having a block of 8 SPEs on the die with the x86 cores makes a lot of sense. You can emulate the PPE and GPU and get backwards compatibility, plus the SPEs are ideal to run a bunch of system level, but computationally intensive services like audio, decompression, video encoding for remote play to Vita and live internet broadcasting like the Vita does through Nico Nico, or Sony provided libraries for post processing and physics. Devs could get the benefit without every directly coding the SPEs, maintaining backwards compatibility would be awesome for users and 8 SPEs would be pretty small on 28nm.

Sony I think we've found your solution... I hope you're listening ;-)
 
This made me laugh :LOL:

http://www.picvalley.net/u/2564/66019497913364665701335371811EYYTYMYO6t5DKeUrNGxF.JPG

For those with bad eye sight

CPU - 22nm Cell @ 3.2Ghz with 16 PPE's and 128 SPE's
GPU - 22nm Custom Nvidia GPU @ 2Ghz based on Kepler GK104x2
RAM - 10Gb XDR2 + 10Gb GDDR6

What you guys reckon? 600-700w power consumption on full load?

Other little things like mandatory 1080p/30fps for all games....

EDIT : Changed GDDR6 amount :oops:

:oops:

This has to be the best (fake) leaked spec sheet ever made.

Unfortunately, it still won't be powerful to please the fans. Not enough of a generational leap.

(Epic will still ask them to double the RAM, trust me ;-))
 
or get rid of the PPU and use x86 CPUs (cores) with the SPEs.

Or drop the SPE's altogether and just use a CPU sporting AVX2. 4 cores at 3Ghz would offer huge amount of vector processing capability making the the addition of SPE's unnecessary except for backwards compatibility..
 
It must be true, it's in print and everything!
Anyone actually want 16 PPE(U?)/128 SPE processor?

You thought PS3 was difficult to program - you ain't seen nothing yet!

OK my firm 99.9% prediction - it will not be called PS4. (yeah Shifty I know wrong thread :p)
 
Status
Not open for further replies.
Back
Top