NVIDIA Tesla K40s vs NVIDIA Tesla P40
Processeur graphique
| Nom du GPU | GK110B | GP102 |
|---|---|---|
| Architecture | Kepler | Pascal |
| Fonderie | TSMC | TSMC |
| Taille du processus | 28 nm | 16 nm |
| Transistors | 7,080 million | 11,800 million |
| Taille des matrices | 561 mm² | 471 mm² |
Carte graphique
| Date de publication | Nov 22nd, 2013 | Sep 13th, 2016 |
|---|---|---|
| Génération | Tesla | Tesla |
| Production | End-of-life | End-of-life |
| Prix de lancement | 7,699 USD | 5,699 USD |
| Bus interface | PCIe 3.0 x16 | PCIe 3.0 x16 |
Vitesse clock
| Base clock | 745 MHz | 1303 MHz |
|---|---|---|
| Boost clock | 876 MHz | 1531 MHz |
| Mémoire clock | 1502 MHz 6 Gbps effective | 1808 MHz 14.5 Gbps effective |
Mémoire
| Taille de la mémoire | 12 GB | 24 GB |
|---|---|---|
| Type de mémoire | GDDR5 | GDDR5X |
| Bus mémoire | 384 bit | 384 bit |
| Bande passante | 288.4 GB/s | 694.3 GB/s |
Render Config
| Shading units | 2880 | 3840 |
|---|---|---|
| TMUs | 240 | 240 |
| ROPs | 48 | 96 |
| Comptage SMX | 15 | — |
| Cache L1 | 16 KB (per SMX) | 48 KB (per SM) |
| Cache L2 | 1536 KB | 3 MB |
| Comptage SM | — | 30 |
Performance théorique
| Taux de pixel | 52.56 GPixel/s | 147.0 GPixel/s |
|---|---|---|
| Taux de texture | 210.2 GTexel/s | 367.4 GTexel/s |
| FP32 (float) performance | 5.046 TFLOPS | 11.76 TFLOPS |
| FP64 (double) performance | 1.682 TFLOPS (1:3) | 367.4 GFLOPS (1:32) |
| FP16 (half) performance | — | 183.7 GFLOPS (1:64) |
Board Design
| Taille Slot | Dual-slot | Dual-slot |
|---|---|---|
| Longueur | 267 mm 10.5 inches | 267 mm 10.5 inches |
| TDP | 245 W | 250 W |
| Suggestions pour le PSU | 550 W | 600 W |
| Sorties | No outputs | No outputs |
| Taille | — | 111 mm 4.4 inches |
| Connecteurs de puissance | — | 1x 6-pin + 1x 8-pin |
| Board number | — | PG610 SKU 200 |
Caractéristiques graphiques
| DirectX | 12 (11_1) | 12 (12_1) |
|---|---|---|
| OpenGL | 4.6 | 4.6 |
| OpenCL | 3.0 | 3.0 |
| Vulkan | 1.1 | 1.2 |
| CUDA | 3.5 | 6.1 |
| Shader model | 5.1 | 6.4 |
