| P100 on E5-2640 | P100 on Silver-4110 | V100 | ||
|---|---|---|---|---|
| Hardware | ||||
| Host | max-wng005 | max-p3ag028 | ||
| CPU | E5-2640 v4 @ 2.40GHz cpu MHz : 1202.343 cache size : 25600 KB | Silver 4114 CPU @ 2.20GHz cpu MHz : 800.000 cache size : 14080 KB | ||
| Memory | 256GB | 768GB | ||
| GPU | P100-PCIE-16GB | P100-PCIE-16GB | ||
| BUS & Numa node | bus:0x02 numa: 2 | bus:0x86 numa: 1 | ||
| CUDA Tests | ||||
| Bandwidth (MB/s) | Host→ Device 11709 Device→Host 12849 Device→Device 500636 | 12049 12863 500300 | ||
| p2pBandwidthLatencyTest (GB/s) | UNI P2P Disabled 346 UNI P2P Enabled 347 BI P2P Disabled 358 BI P2P Enabled 357 | 504 504 512 513 | ||
| convolutionFFT2D (Mpix/s) | built-in R2C / C2R 6088 custom R2C / C2R 6144 updated custom R2C / C2R 6069 | 6088 6107 7812 | ||
| simpleMultiCopy (GB/s) | Host→ Device 11.9 Device→Host 12.7 Kernel 417.1 Serialized exec 10.6 4 Streams 19.4 | 12.2 12.7 1248.0 12.0 20.6 | ||
| matrixMul (GFlop/s) | 420.0 | 1766 | ||
| matrixMulCUBLAS (GFlop/s) | 2321.7 | 5579 | ||
| Benchmarks | ||||
| shmembench | using 32bit operations : 1777.72 GB/sec using 64bit operations : 1815.44 GB/sec using 128bit operations : 1816.02 GB/sec | 7614.55 GB/sec 7890.97 GB/sec 7662.91 GB/sec | ||
| constbench | using 32bit operations : 650.68 GB/sec using 64bit operations : 2247.59 GB/sec using 128bit operations : 2402.28 GB/sec | 2778.73 GB/sec 9522.36 GB/sec 10218.13 GB/sec | ||
| cachebench | Read only accesses: int1: 1000.92 GB/sec int2: 2106.63 GB/sec int4: 2379.41 GB/sec max: 2379.41 GB/sec Read-write accesses: int1: 2229.74 GB/sec int2: 2210.64 GB/sec int4: 2127.50 GB/sec max: 2229.74 GB/sec | int1: 2121.66 GB/sec int2: 2310.82 GB/sec int4: 2379.64 GB/sec max: 2379.64 GB/sec int1: 2230.04 GB/sec int2: 2211.12 GB/sec int4: 2126.41 GB/sec max: 2230.04 GB/sec | ||
| gpu-burn (Gflops/s) | 7990 | 7990 | ||
| Others | ||||
| Tensorflow Resnet 50 | images/sec: 214.9 | images/sec: 210.41 | ||