NVIDIA GRID A100A vs NVIDIA GRID RTX T10-8
Processeur graphique
| Nom du GPU | GA100 | TU102 |
|---|---|---|
| Architecture | Ampere | Turing |
| Fonderie | TSMC | TSMC |
| Taille du processus | 7 nm | 12 nm |
| Transistors | 54,200 million | 18,600 million |
| Taille des matrices | 826 mm² | 754 mm² |
| GPU variant | — | TU102-875-A1 |
Carte graphique
| Date de publication | May 14th, 2020 | Unknown |
|---|---|---|
| Génération | GRID | GRID |
| Production | Active | Active |
| Bus interface | PCIe 4.0 x16 | PCIe 3.0 x16 |
Vitesse clock
| Base clock | 900 MHz | 1065 MHz |
|---|---|---|
| Boost clock | 1005 MHz | 1395 MHz |
| Mémoire clock | 1215 MHz 2.4 Gbps effective | 1750 MHz 14 Gbps effective |
Mémoire
| Taille de la mémoire | 48 GB | 8 GB |
|---|---|---|
| Type de mémoire | HBM2E | GDDR6 |
| Bus mémoire | 6144 bit | 384 bit |
| Bande passante | 1,866 GB/s | 672.0 GB/s |
Render Config
| Shading units | 6912 | 4608 |
|---|---|---|
| TMUs | 432 | 288 |
| ROPs | 192 | 96 |
| Comptage SM | 108 | 72 |
| Tensor cores | 432 | 576 |
| Cache L1 | 192 KB (per SM) | 64 KB (per SM) |
| Cache L2 | 48 MB | 6 MB |
| RT cores | — | 72 |
Performance théorique
| Taux de pixel | 193.0 GPixel/s | 133.9 GPixel/s |
|---|---|---|
| Taux de texture | 434.2 GTexel/s | 401.8 GTexel/s |
| FP16 (half) performance | 55.57 TFLOPS (4:1) | 25.71 TFLOPS (2:1) |
| FP32 (float) performance | 13.89 TFLOPS | 12.86 TFLOPS |
| FP64 (double) performance | 6.947 TFLOPS (1:2) | 401.8 GFLOPS (1:32) |
Board Design
| Taille Slot | IGP | Dual-slot |
|---|---|---|
| TDP | 400 W | 260 W |
| Suggestions pour le PSU | 800 W | 600 W |
| Sorties | No outputs | No outputs |
| Connecteurs de puissance | None | 1x 6-pin + 1x 8-pin |
| Longueur | — | 267 mm 10.5 inches |
| Board number | — | PG150 SKU 220 |
Caractéristiques graphiques
| DirectX | — | 12 Ultimate (12_2) |
|---|---|---|
| OpenGL | — | 4.6 |
| OpenCL | 3.0 | 3.0 |
| Vulkan | — | 1.2 |
| CUDA | 8.0 | 7.5 |
| Shader model | — | 6.6 |
