High Core Count CPUs Software Renderering Video Games?

[ArbitraryAffection]

Hiya.

I think this is the right place to ask this, because my question is about the hardware. Anyway sorry if it's not. Ahem. Anyway~

Microsoft had some kind of DX12 capable software renderer that allows a CPU to render a video game, for example, but in software (so it emulates the function of a GPU). Is that a thing?

And if so, I have a silly question. Would this software renderer scale with AVX2/512 and if so could a 28-core Intel Xeon on Skylake with AVX512 be able to render a modern video game at a decent FPS at say, 1080p? Without the need for a dedicated GPU doing the work? Also the 32-core EPYC processor using AVX2, how would that compare to for example a GTX 1080 or RX Vega 64?

Sorry if it's a dumb question but I was just honestly curious.

 

[Deleted member 178884]

It wouldn't work well at all, and for starters, Intel and AMDs HEDT don't have a iGPU to output a display anyways.

could a 28-core Intel Xeon on Skylake with AVX512 be able to render a modern video game at a decent FPS at say, 1080p?

no

 

[Vya Domus]

Short answer, no. Long answer :

Even the fastest single socket CPUs have a FP32 performance in the range of about 1 TFLOPS and while that would be somewhat comparable to a low end GPU it is much more difficult to extract that sort of performance out of a CPU than it is out of a GPU. Moreover, even if they would get close to the same level of compute they still lack the dedicated hardware for texture mapping, rasterization , tessellation, etc. All of that functionality would have to be emulated.

CPUs and GPUs are separate entities for a reason.

 

[Deleted member 178884]

There's plenty of limitations anyways, GPUs are heavily optimized for games and if this was possible at decent frame rates already then intel wouldn't have tried to enter the GPU market and failed multiple times.

 

[ArbitraryAffection]

Okay. Thanks for the replies.

Could someone tell me the actual peak FP32 throughput of the 28-core Skylake and the 32-core EPYC please? Just out of curiosity. That is with AVX.

Sorry if it seemed dumb.

 

[Disparia]

In theory, sure. Intel has been showing off CPU rendered games for over a decade, starting with Quake on four networked dual-CPU machines. The frame-rate wasn't anything special but was an important milestone, demonstrating ray-traced scenes rendered over multiple cores. These demonstrations over the years have increased in both the resolution and complexity, as well as the underlying system used, such as 4P and 8P multi-core servers, and multiple Xeon Phi co-processors. With AMD and Intel putting out Threadripper, Epyc, Xeon-W, etc; we're on the cusp of a single socket with more than enough power to handle complex screens at high FPS. I don't know the intricacies of DX12 well enough to speak to your question specifically, but this kind of thing scales perfectly, like 99.9%. If AMD/Intel continues to commit to CPUs with high core counts, this could be a future.

Short answer no, don't have any quick way to accomplish what you're asking, but long answer, it's a not a stupid question. Vast amounts of money has been invested in this very idea and it will come around when it's possible to do it. It's kind of like VR. I've experienced 3 major VR "feel the water" events in my life, including what is happening now. Should the market bare it, VR becomes common, if not, try again in some years.

 

[Vayra86]

Short answer, no. Long answer :

Even the fastest single socket CPUs have a FP32 performance in the range of about 1 TFLOPS and while that would be somewhat comparable to a low end GPU it is much more difficult to extract that sort of performance out of a CPU than it is out of a GPU. Moreover, even if they would get close to the same level of compute they still lack the dedicated hardware for texture mapping, rasterization , tessellation, etc. All of that functionality would have to be emulated.

CPUs and GPUs are separate entities for a reason.

Heh. And even IF the CPU could push the frames, how would it transport all that data... there is a massive bandwidth problem here