Nom du GPU | Polaris 23 | TU104 |
---|---|---|
GPU variant | Polaris 23 XT GL | TU104-850-A1 |
Architecture | GCN 4.0 | Turing |
Fonderie | GlobalFoundries | TSMC |
Taille du processus | 14 nm | 12 nm |
Transistors | 2,200 million | 13,600 million |
Taille des matrices | 103 mm² | 545 mm² |
Date de publication | Jul 2nd, 2019 | Nov 13th, 2018 |
---|---|---|
Génération | Radeon Pro | Quadro |
Production | Active | Active |
Prix de lancement | 199 USD | 899 USD |
Bus interface | PCIe 3.0 x8 | PCIe 3.0 x16 |
GPU clock | 1295 MHz | — |
---|---|---|
Mémoire clock | 1500 MHz 6 Gbps effective | 1625 MHz 13 Gbps effective |
Base clock | — | 1005 MHz |
Boost clock | — | 1545 MHz |
Taille de la mémoire | 4 GB | 8 GB |
---|---|---|
Type de mémoire | GDDR5 | GDDR6 |
Bus mémoire | 128 bit | 256 bit |
Bande passante | 96.00 GB/s | 416.0 GB/s |
Shading units | 640 | 2304 |
---|---|---|
TMUs | 32 | 144 |
ROPs | 16 | 64 |
Compute units | 10 | — |
Cache L1 | 16 KB (per CU) | 64 KB (per SM) |
Cache L2 | 512 KB | 4 MB |
Comptage SM | — | 36 |
Tensor cores | — | 288 |
RT cores | — | 36 |
Taux de pixel | 20.72 GPixel/s | 98.88 GPixel/s |
---|---|---|
Taux de texture | 41.44 GTexel/s | 222.5 GTexel/s |
FP16 (half) performance | 1.658 TFLOPS (1:1) | 14.24 TFLOPS (2:1) |
FP32 (float) performance | 1.658 TFLOPS | 7.119 TFLOPS |
FP64 (double) performance | 103.6 GFLOPS (1:16) | 222.5 GFLOPS (1:32) |
Taille Slot | Single-slot | Single-slot |
---|---|---|
Longueur | 167 mm 6.6 inches | 241 mm 9.5 inches |
Taille | 69 mm 2.7 inches | 111 mm 4.4 inches |
TDP | 65 W | 160 W |
Suggestions pour le PSU | 250 W | 450 W |
Sorties | 4x mini-DisplayPort | 3x DisplayPort1x USB Type-C |
Connecteurs de puissance | None | 1x 8-pin |
DirectX | 12 (12_0) | 12 Ultimate (12_2) |
---|---|---|
OpenGL | 4.6 | 4.6 |
OpenCL | 2.1 | 3.0 |
Vulkan | 1.2 | 1.2 |
Shader model | 6.4 | 6.6 |
CUDA | — | 7.5 |