NVIDIA Tesla C870 vs NVIDIA Tesla P100 PCIe 16 GB
Processeur graphique
| Nom du GPU | G80 | GP100 |
|---|---|---|
| Architecture | Tesla | Pascal |
| Fonderie | TSMC | TSMC |
| Taille du processus | 90 nm | 16 nm |
| Transistors | 681 million | 15,300 million |
| Taille des matrices | 484 mm² | 610 mm² |
| GPU variant | — | GP100-893-A1 |
Carte graphique
| Date de publication | May 2nd, 2007 | Jun 20th, 2016 |
|---|---|---|
| Génération | Tesla | Tesla |
| Production | End-of-life | End-of-life |
| Prix de lancement | 1,499 USD | 5,699 USD |
| Bus interface | PCIe 1.0 x16 | PCIe 3.0 x16 |
Vitesse clock
| GPU clock | 600 MHz | — |
|---|---|---|
| Shader clock | 1350 MHz | — |
| Mémoire clock | 800 MHz 1600 Mbps effective | 715 MHz 1430 Mbps effective |
| Base clock | — | 1190 MHz |
| Boost clock | — | 1329 MHz |
Mémoire
| Taille de la mémoire | 1536 MB | 16 GB |
|---|---|---|
| Type de mémoire | GDDR3 | HBM2 |
| Bus mémoire | 384 bit | 4096 bit |
| Bande passante | 76.80 GB/s | 732.2 GB/s |
Render Config
| Shading units | 128 | 3584 |
|---|---|---|
| TMUs | 32 | 224 |
| ROPs | 24 | 96 |
| Comptage SM | 16 | 56 |
| Cache L2 | 96 KB | 4 MB |
| Cache L1 | — | 24 KB (per SM) |
Performance théorique
| Taux de pixel | 14.40 GPixel/s | 127.6 GPixel/s |
|---|---|---|
| Taux de texture | 38.40 GTexel/s | 297.7 GTexel/s |
| FP32 (float) performance | 345.6 GFLOPS | 9.526 TFLOPS |
| FP16 (half) performance | — | 19.05 TFLOPS (2:1) |
| FP64 (double) performance | — | 4.763 TFLOPS (1:2) |
Board Design
| Taille Slot | Dual-slot | Dual-slot |
|---|---|---|
| TDP | 171 W | 250 W |
| Suggestions pour le PSU | 450 W | 600 W |
| Sorties | No outputs | No outputs |
| Connecteurs de puissance | 2x 6-pin | 1x 8-pin |
| Board number | P357 | — |
| Longueur | — | 267 mm 10.5 inches |
Caractéristiques graphiques
| DirectX | 11.1 (10_0) | 12 (12_1) |
|---|---|---|
| OpenGL | 3.3 | 4.6 |
| OpenCL | 1.1 (1.0) | 3.0 |
| Vulkan | — | 1.2 |
| CUDA | 1.0 | 6.0 |
| Shader model | 4.0 | 6.4 |
