AMD Radeon Instinct MI25 vs NVIDIA GRID K340
Processeur graphique
| Nom du GPU | Vega 10 | GK107 |
|---|---|---|
| GPU variant | Vega 10 XT GL (215-0894124) | — |
| Architecture | GCN 5.0 | Kepler |
| Fonderie | GlobalFoundries | TSMC |
| Taille du processus | 14 nm | 28 nm |
| Transistors | 12,500 million | 1,270 million |
| Taille des matrices | 495 mm² | 118 mm² |
Carte graphique
| Date de publication | Jun 27th, 2017 | Jul 23rd, 2013 |
|---|---|---|
| Génération | Radeon Instinct | GRID |
| Production | Active | End-of-life |
| Bus interface | PCIe 3.0 x16 | PCIe 3.0 x16 |
| Prix de lancement | — | 3,299 USD |
Vitesse clock
| Base clock | 1400 MHz | — |
|---|---|---|
| Boost clock | 1500 MHz | — |
| Mémoire clock | 852 MHz 1704 Mbps effective | 1250 MHz 5 Gbps effective |
| GPU clock | — | 954 MHz |
Mémoire
| Taille de la mémoire | 16 GB | 1024 MB |
|---|---|---|
| Type de mémoire | HBM2 | GDDR5 |
| Bus mémoire | 2048 bit | 128 bit |
| Bande passante | 436.2 GB/s | 80.00 GB/s |
Render Config
| Shading units | 4096 | 384 |
|---|---|---|
| TMUs | 256 | 32 |
| ROPs | 64 | 16 |
| Compute units | 64 | — |
| Cache L1 | 16 KB (per CU) | 16 KB (per SMX) |
| Cache L2 | 4 MB | 128 KB |
| Comptage SMX | — | 2 |
Performance théorique
| Taux de pixel | 96.00 GPixel/s | 7.632 GPixel/s |
|---|---|---|
| Taux de texture | 384.0 GTexel/s | 30.53 GTexel/s |
| FP16 (half) performance | 24.58 TFLOPS (2:1) | — |
| FP32 (float) performance | 12.29 TFLOPS | 732.7 GFLOPS |
| FP64 (double) performance | 768.0 GFLOPS (1:16) | 30.53 GFLOPS (1:24) |
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 | 225 W |
| Suggestions pour le PSU | 700 W | 550 W |
| Sorties | No outputs | No outputs |
| Connecteurs de puissance | 2x 8-pin | 1x 8-pin |
| Board number | D051 | — |
Caractéristiques graphiques
| DirectX | 12 (12_1) | 12 (11_0) |
|---|---|---|
| OpenGL | 4.6 | 4.6 |
| OpenCL | 2.1 | 3.0 |
| Vulkan | 1.2 | 1.1 |
| Shader model | 6.4 | 5.1 |
| CUDA | — | 3.0 |
