Big Blue talks up 45nm PlayStation 3 processor possibilities
Higher clock speeds or a smaller power drain, Sony?
It's no great surprise, perhaps, but Cell Broadband Engine chip co-developer IBM will produce a 45nm version of the processor, paving the way for smaller, cooler-running PlayStation 3 consoles.
Originally produced using a 90nm process, Cell is currently being punched out at 65nm. That version debuted with the 40GB PS3. The 45nm version is expected to consume just a third of the power that the 90nm Cell did and less than half of that drawn by the 65nm chip, IBM indicated.
Those measurements assume all three are running at 4GHz. Alternatively, Sony could use the new power characteristics to boost the clock speed at the expense of power draw. That could see the chip clock to 6-7GHz yet operate in the same power envelope as today's PS3s provide.
There's also the opportunity to use the process shrink to redesign the core to add more processing cores and local memory.
More likely, Sony would use the 45nm Call as an opportunity to shrink the console's cooling system, making for a leaner machine. Such a product was recently rumoured to be in the pipeline, but since it's common Sony practice to reduce the size of its consoles, that's no great revelation.
The 45nm Cell has an area of 114.5mm² - about half the space taken up by the 90nm die. It has a PowerPC based processing core on board, along with up to eight Synergistic Processing Engines (SPEs) - essentially vector-data maths processors. The die has 512KB of L2 cache for the main core, while each SPE has 256KB cache of its own. The die incorporates a memory controller that connects it to Rambus XDR high-speed memory.
I attended a talk given by a PS3 game developer in which I was told that future IBM Cell processors would contain 100s of Synergestic Programming Elements. He mentioned this fact in order to emphasise the importance of making increasing use of SPEs if one was to produce competetive high performance games.
If future Cell processors run at no more than the current clock frequency of 3.2 GHz then this has the implication that existing games would work without any problem on the future Playstation. The only limitation would be that the games would not be making full use of the 100s of SPEs in the new Cell processor.
Another snippet of interesting imformation from this developer is that he said that Java might replace C++ in future game devwelopment despite the inefficiency of BYTE codes. He said that the reason for this is that it was easier to generate parallel code for SPEs in a Java compiler than in a C++ compiler.
Heck, port FFTW, and Mathematica, and maybe Gauss to these little beasties - 8 per card - and sell it in the HPC market. They're too hard to program straight up in C for most researchers, but with a decent library interface they could prove very popular.
Good IBM, now put 4-8 of these on clusterable cards and write directx drivers.
The Cell was designed from the ground up to be stackable. Now that they are fast enough, lets start scaling these into huge arrays and write some directx drivers for them :)
My 8800GT runs at 600Mhz. If a cell can do 6ghz, and you can cluster a bunch of them, even with some wacky emulation, there has to be a way to get up to speed.
I'll even buy the first version if it is only as fast as 10 8800GT's. :-0
P.S. Add native h.264 encode and decode to each of the cores please. I have some home movies that need compressed.
Hot air blasting from the vents? It should be hotter! Stood in front of an IBM Z-series mainframe recently. Honestly it's like a furnace combined with a wind tunnel.
Now THAT's computing :-)
re: changing clockspeed.
Don't put it past Sony. They doubled the memory capacity of the slim and light PSP over the fat and heavy one, which is why the official skype firmware doesn't run on the older model.