So this thing is basically a cpu pretending to be a gpu?
What the fuck were they thinking?
Sony pioneered high levels of parallel processing for consumer gaming devices, the PS2 aimed for the same thing, as does the PS4 now.
The Cell Broadband Engine has these units called SPUs which are capable of processing SIMD with fairly high throughput. Its a CPU that does a bit of what a GPU does, with a close connection to a dedicated GPU.
>>51637124
Did you know that a GPU is a processing unit? Did you know that a CPU is a processing unit? Is it really surprising to learn that a CPU can do similar things to a GPU?
>>51637124
They weren't, which is why no one used it.
Cell is a pretty cool architecture, but it is too complex which is why game devs never put in the necessary effort to hyper-optimise their game engines. I don't think we ever saw what the PS3 was fully capable off.
>>51637124
No, it was just a really slow CPU connected to 7 even slower CPUs.
>>51637124
it's a ken kutaragi wank, that sony had to accept because of sucess of the PS2
>>51637124
...and this is the real GPU. What's the problem?
Will this fucking thing ever be emulated?
Considering we still don't have perfect PS2 emulation it could be decades before we get something playable out of a PS3 emu
>>51637124
>A GPU with 10 cores
That'd be a really shitty GPU.
>>51639305
Yes, HLE will make it happen.
>>51637422
But PS4 is just a x86 pc.
>>51638124
nvidia 7950gt from 2006 to be exact.
>>51639327
Its an HSA compliant APU with a flat address space between the CPU and GPU. It was purpose designed to excel in compute workloads for its given power.
Its not similar to a X86 desktop with discrete GPU.
>>51639305
Believe in Play!.
>>51639620
It is very similar. The only real difference is that the CPU and GPU share the same memory and have to compete for the same bandwidth with equal priority. You can do the same kind of GPU compute tasks as with regular discrete solutions, the only difference is that all you need to do is pass along a pointer rather than write to and read from the GPU memory all the time.
In high performance computing applications they've just concluded that this is pointless as most of the run time is going to actually crunching the numbers anyway and that you lose more than you gain from the performance losses you get when doing this. As for GPGPU compute in every day applications it's not faster than a discrete chips ether, you can get the same kind of results by just using AVX instructions.
The only real benefit from the APU's AMD has pushed out so far is decent game performance on a budget. This is obviously the main reason why the PS4 and XB1 use them.
