AMD Radeon HD 8950 OEM vs NVIDIA GRID A100A
Processeur graphique
| Nom du GPU | Tahiti | GA100 |
|---|---|---|
| GPU variant | Tahiti PRO (215-0821056) | — |
| Architecture | GCN 1.0 | Ampere |
| Fonderie | TSMC | TSMC |
| Taille du processus | 28 nm | 7 nm |
| Transistors | 4,313 million | 54,200 million |
| Taille des matrices | 352 mm² | 826 mm² |
Carte graphique
| Date de publication | Jan 8th, 2013 | May 14th, 2020 |
|---|---|---|
| Génération | Sea Islands | GRID |
| Production | End-of-life | Active |
| Bus interface | PCIe 3.0 x16 | PCIe 4.0 x16 |
| Prédécesseur | Southern Islands | — |
| Successeur | Volcanic Islands | — |
Vitesse clock
| Base clock | 850 MHz | 900 MHz |
|---|---|---|
| Boost clock | 925 MHz | 1005 MHz |
| Mémoire clock | 1250 MHz 5 Gbps effective | 1215 MHz 2.4 Gbps effective |
Mémoire
| Taille de la mémoire | 3 GB | 48 GB |
|---|---|---|
| Type de mémoire | GDDR5 | HBM2E |
| Bus mémoire | 384 bit | 6144 bit |
| Bande passante | 240.0 GB/s | 1,866 GB/s |
Render Config
| Shading units | 1792 | 6912 |
|---|---|---|
| TMUs | 112 | 432 |
| ROPs | 32 | 192 |
| Compute units | 28 | — |
| Cache L1 | 16 KB (per CU) | 192 KB (per SM) |
| Cache L2 | 768 KB | 48 MB |
| Comptage SM | — | 108 |
| Tensor cores | — | 432 |
Performance théorique
| Taux de pixel | 29.60 GPixel/s | 193.0 GPixel/s |
|---|---|---|
| Taux de texture | 103.6 GTexel/s | 434.2 GTexel/s |
| FP32 (float) performance | 3.315 TFLOPS | 13.89 TFLOPS |
| FP64 (double) performance | 828.8 GFLOPS (1:4) | 6.947 TFLOPS (1:2) |
| FP16 (half) performance | — | 55.57 TFLOPS (4:1) |
Board Design
| Taille Slot | Dual-slot | IGP |
|---|---|---|
| Longueur | 267 mm 10.5 inches | — |
| TDP | 200 W | 400 W |
| Suggestions pour le PSU | 550 W | 800 W |
| Sorties | 1x DVI1x HDMI2x mini-DisplayPort | No outputs |
| Connecteurs de puissance | 2x 6-pin | None |
| Board number | C386-42 | — |
Caractéristiques graphiques
| DirectX | 12 (11_1) | — |
|---|---|---|
| OpenGL | 4.6 | — |
| OpenCL | 1.2 | 3.0 |
| Vulkan | 1.2 | — |
| Shader model | 5.1 | — |
| CUDA | — | 8.0 |
