10 ms·
If you're treating that number as accurate, then I'm running on a 896-core x86 computer. As a comparison between GPUs and CPUs, each "CUDA core" is approximate
by jcranmer 7y ago
If you're treating that number as accurate, then I'm running on a 896-core x86 computer.
As a comparison between GPUs and CPUs, each "CUDA core" is approximately equal to a SIMD lane on an x86 core. So 56 cores, times 16 lanes (AVX-512) per core, is 896 cores.
- mattnewport 7y ago> each "CUDA core" is approximately equal to a SIMD lane on an x86 core It's not really, that's why Larrabee failed.
- lostmsu 7y agoCould you ELI the difference.
- vardump 7y agoBingo. It has bothered me a long time how the whole industry has fallen to GPU manufacturer marketing speak. That Nvidia Titan RTX thingie is a monster, but it has only 72 SMs (Streaming Multiprocessor), the closest equivalent to a CPU core. That is, SM is the smallest unit that can individually execute a true branch. Just like a CPU core. Not counting hyperthreads here, GPUs have a metric ton of them, because they have to do useful work while waiting for very high latency memory. CPUs need much less of them to cover "bubbles in the pipeline".
- mattnewport 7y agoThe Turing architecture is a bit more sophisticated than that: "Like Volta, the Turing SM is partitioned into 4 sub-cores (or processing blocks) with each sub-core having a single warp scheduler and dispatch unit" "Like we saw in Volta, these changes go hand-in-hand with the new scheduling/execution model with independent thread scheduling that Turing also has, though differences were not disclosed at this time. Rather than per-warp like Pascal, Volta and Turing have per-thread scheduling resources, with a program counter and stack per-thread to track thread state, as well as a convergence optimizer to intelligently group active same-warp threads together into SIMT units. So all threads are equally concurrent, regardless of warp, and can yield and reconverge." GPU 'cores' are still not as general as a full CPU core but they are a fair bit more flexible and sophisticated than a SIMD lane. https://www.anandtech.com/show/13282/nvidia-turing-architecture-deep-dive/4 https://www.anandtech.com/show/13282/nvidia-turing-architect...
- vardump 7y agoInteresting. So what are the sub-core limitations execution flow wise? Or is it truly generic? Can there be execution level conflicts? Like execution flow could be taken, but execution resources constraining (stalling) it somehow? I'm only talking about "per clock" view, not about hardware threading. Sometimes these terms are mixed in a bit hard to decipher fashion.
- mattnewport 7y agoIt's hard to get detailed information on a lot of this as much of it is considered proprietary by NVIDIA. My understanding is that threads are not truly independent as they would be on full CPUs but that the hardware is able to dynamically detect thread divergence (when different lanes / threads of the SIMDish hardware take different branches) and intelligently regroup the threads to minimize divergence. I'm not sure how they then deal with instructions that operate across a warp, it's possible those interfere with the regrouping but they're not commonly used in many typical pixel shader workloads so it may be an overall win. I suspect some of these changes were to make raytracing more efficient (and the Anandtech article suggests as much too). A common problem trying to implement a GPU raytracer on older architectures is performance tanking when rays diverge too much. NVIDIA's Optix did some complicated magic to try and regroup divergent threads but having direct hardware support for it probably makes the raytracing GPU code both simpler and more efficient.
- zozbot234 7y agoThe description is unclear, but these look more like hyperthreads/SMT units than true CPU cores. It's talking about "scheduling resources" implemented in hardware, which suggests a SMT model.
- jcranmer 7y agoGPUs are already SMT processors, since that's how they handle main memory latency. Instead, I suspect it's essentially dynamic warp formation to handle cases where there's still enough threads to keep all the lanes busy, but they're spread out over several static 16-thread chunks instead of being right next to each other.
- chrchang523 7y ago(initial comment deleted since one part addressed integer 'lanes' instead of the 32-bit FP that's more directly relevant in a comparison with a GPU, and the other part was incorrect.)
- vardump 7y ago> They have 16 SIMD registers. AVX-512 has 32 SIMD registers.
- jcranmer 7y agoI'm counting the number of single-precision FP units. Although, strictly speaking, AVX-512 hardware actually has two FMA units per core.
- vardump 7y agoI thought AVX-512 ALU resources varied between models?
- jcranmer 7y agoThis is Skylake server part (which should be implied by the high core count), which is using this block diagram: https://www.researchgate.net/figure/Skylake-Microarchitecture-CPU-Core-Block-Diagram-6_fig5_332543387 https://www.researchgate.net/figure/Skylake-Microarchitectur...