AMD Radeon Instinct MI210 vs NVIDIA A100 PCIe
Processeur graphique
| Nom du GPU | Aldebaran | GA100 |
|---|---|---|
| GPU variant | Aldebaran | — |
| Architecture | CDNA 2.0 | Ampere |
| Fonderie | TSMC | TSMC |
| Taille du processus | 6 nm | 7 nm |
| Transistors | 58,200 million | 54,200 million |
| Taille des matrices | unknown | 826 mm² |
Carte graphique
| Date de publication | Dec 2021 | Jun 22nd, 2020 |
|---|---|---|
| Génération | Radeon Instinct | Tesla |
| Production | Active | Active |
| Bus interface | PCIe 4.0 x16 | PCIe 4.0 x16 |
Vitesse clock
| Base clock | 1000 MHz | 765 MHz |
|---|---|---|
| Boost clock | 1700 MHz | 1410 MHz |
| Mémoire clock | 1600 MHz 3.2 Gbps effective | 1215 MHz 2.4 Gbps effective |
| GPU clock | — | 1410 MHz |
Mémoire
| Taille de la mémoire | 64 GB | 40 GB |
|---|---|---|
| Type de mémoire | HBM2e | HBM2E |
| Bus mémoire | 4096 bit | 5120 bit |
| Bande passante | 1,638 GB/s | 1,555 GB/s |
Render Config
| Shading units | 6656 | 6912 |
|---|---|---|
| TMUs | 416 | 432 |
| ROPs | 0 | 160 |
| Compute units | 104 | — |
| gpu.details.mcm | 1 | — |
| Cache L1 | 16 KB (per CU) | 192 KB (per SM) |
| Cache L2 | 16 MB | 40 MB |
| Comptage SM | — | 108 |
| Tensor cores | — | 432 |
Performance théorique
| Taux de pixel | 0 MPixel/s | 225.6 GPixel/s |
|---|---|---|
| Taux de texture | 707.2 GTexel/s | 609.1 GTexel/s |
| FP16 (half) performance | 181.0 TFLOPS (8:1) | 77.97 TFLOPS (4:1) |
| FP32 (float) performance | 22.63 TFLOPS | 19.49 TFLOPS |
| FP64 (double) performance | 22.63 TFLOPS (1:1) | 9.746 TFLOPS (1:2) |
Board Design
| Taille Slot | Dual-slot | Dual-slot |
|---|---|---|
| Longueur | 267 mm 10.5 inches | 267 mm 10.5 inches |
| Taille | 111 mm 4.4 inches | — |
| TDP | 300 W | 250 W |
| Suggestions pour le PSU | 700 W | 600 W |
| Sorties | No outputs | No outputs |
| Connecteurs de puissance | 2x 8-pin | 8-pin EPS |
| Board number | — | P1001 SKU 200 |
Caractéristiques graphiques
| DirectX | — | — |
|---|---|---|
| OpenGL | — | — |
| OpenCL | 3.0 | 3.0 |
| Vulkan | — | — |
| Shader model | — | — |
| CUDA | — | 8.0 |
