Nvidia RSX and perhaps a way PC's could learn from CELL+GPU processing?

Mikjul · October 25, 2015

I may be a bit out of touch with latest advancements in tech so do excuse me if this thread is pointless.

PS3 had a weird but genius way of making it's graphical performance greater than the base spec of it's RSX GPU by using the CELL to do GPU calculations along side it... Kind of like dual graphics but CELL was limited to what it could process obviously by being a CPU.

Now i'm having a console session with some PS3 exclusives and can't hekp but notice that a GeForce 7900 GTX or even some 8000 cards would be pulled under by what PS3 is delivering on screen. Not nonumentally massive

jumps in petformance but enough to be 5-10 fps ahead of said GPU's.

Why is this not usable on the PC platform?

Imagine an i7 doing the tesselation calculations and vertex work... It would speed up GPU's quite a lot and allow for mote advanced usage of this tech.

Another thing... These new consoles just cannot pull this off.. CPU being a huge reason.. CELL is not powerful in normal things but for parallel tasks it's still a beast.

Our PC's should in theory be able to do this... Even desktop FX chips.

LukaP · October 25, 2015

Not all CPUs are powerful enough to do this, resulting in ahuge performance hit over GPU only for those systems

Not many things GPUs process is very efficient to do on a CPU (eg physics, lighting, AI are all embarrasingly parallel and 2 threads running at 4GHz will always be beaten by 64 threads with a much lower IPC running at like 0.5GHz)

Slow interconnect introduces alot of latency (even PCIe 4.0 will be slow compared to having it in HBM/GDDR5X)

You need to double up the resources into system ram, increasing the minimum specs by a fair amount

Its a nightmare to code for.

APasz · October 25, 2015

LukaP3 has basically hit the nail on the head. Its just not practical considering how prominent the x86 and x64 architectures are. Cell tech is just so very different.