>muh nvidia >muh amd when will /g/ get a real coprocessor?

>>56209022
Define "real" coprocessor.

>>56209032
Something that exists

File: 1470576539509.jpg (110KB, 759x508px) Image search: [Google]

110KB, 759x508px

>atom cores

>>56209022
Maybe when Intel actually, IDK, finally releases this shit?

File: a80387dx___20_intel_very_rare_387_20mhz_math_processor_revision_sc21_vintage_1_lgw.jpg (52KB, 632x596px) Image search: [Google]

52KB, 632x596px

>>56209049
You mean like the 80387 released in 1987?

>>56209022
Sure is 1980's in here

File: Intel_C8087.jpg (17KB, 790x287px) Image search: [Google]

17KB, 790x287px

>>56209081
No. Something like the 8087 from 1980.

Why would you want several shitty pentium 4s? GPU's deliver better performance anyway
https://blogs.nvidia.com/blog/2016/08/16/correcting-some-mistakes/

>>56209064
atom cores are just as powerful as the "full sized" parts.
the key is they just have really low clock speeds to either be cheap, or use less power (like with this)
when you have 68 cpu cores crammed into something that small you can't expect a lot of per core performance.

>>56209022
Does that expand the processing power of the main CPU?

>>56210638
Depending on your needs, they provide better performance. Is your problem easy to parallelize or is it not?

>>56210738
Most games and emulators rely heavily on single thread performance, rarely two to four cores are used.

>>56210738
>my small penis is just like a "full sized" penis
this is you

>>56210993
>>56211014
MUH GAMES MUH SINGLE CORE

Can you /v/tards please fuck off and kill yourselves while you're at it?

>>56210993
Most? You must use a shitty engine. Pretty much all of them have job queues and thread pools. There's a fuckton that can be offloaded from the main thread.

>>56210993
>muh gaymes

>>56209081
it's built-in these days

File: 1471561856486.jpg (247KB, 1224x1445px) Image search: [Google]

247KB, 1224x1445px

>>56210993
>xeon phi
>muh gaymes

>>56210738
They kinda aren't. Atoms don't even have L3 cache.

>>56209022
what is the point of Phi/Knights (X) architecture?

it seems like scientific computing code has simple enough flow control and enough data parallelism that SIMT makes more sense than giant grids of nigger-tier pentiums with bolted-on SSE/AVX.

what types of workloads really benefit from it?

>>56210940
This is a GPU iirc

>>56210738
>when you have 68 cpu cores crammed into something that small you can't expect a lot of per core performance.

Meanwhile...
>>>56194667

So what fucking purpose does this piece of shit serve now?

>>56209022
This shit is fucking retarded to work with

>>56210993
Bad engine optimisation.