Nom du GPU | TU102 | GA102 |
---|---|---|
GPU variant | TU102-875-A1 | — |
Architecture | Turing | Ampere |
Fonderie | TSMC | Samsung |
Taille du processus | 12 nm | 8 nm |
Transistors | 18,600 million | 28,300 million |
Taille des matrices | 754 mm² | 628 mm² |
Date de publication | Aug 13th, 2018 | Apr 12th, 2021 |
---|---|---|
Génération | Quadro | Quadro |
Production | Active | Active |
Prix de lancement | 6,299 USD | — |
Bus interface | PCIe 3.0 x16 | PCIe 4.0 x16 |
Base clock | 1440 MHz | 1170 MHz |
---|---|---|
Boost clock | 1770 MHz | 1695 MHz |
Mémoire clock | 1750 MHz 14 Gbps effective | 2000 MHz 16 Gbps effective |
Taille de la mémoire | 24 GB | 24 GB |
---|---|---|
Type de mémoire | GDDR6 | GDDR6 |
Bus mémoire | 384 bit | 384 bit |
Bande passante | 672.0 GB/s | 768.0 GB/s |
Shading units | 4608 | 8192 |
---|---|---|
TMUs | 288 | 256 |
ROPs | 96 | 96 |
Comptage SM | 72 | 64 |
Tensor cores | 576 | 256 |
RT cores | 72 | 64 |
Cache L1 | 64 KB (per SM) | 128 KB (per SM) |
Cache L2 | 6 MB | 6 MB |
Taux de pixel | 169.9 GPixel/s | 162.7 GPixel/s |
---|---|---|
Taux de texture | 509.8 GTexel/s | 433.9 GTexel/s |
FP16 (half) performance | 32.62 TFLOPS (2:1) | 27.77 TFLOPS (1:1) |
FP32 (float) performance | 16.31 TFLOPS | 27.77 TFLOPS |
FP64 (double) performance | 509.8 GFLOPS (1:32) | 867.8 GFLOPS (1:32) |
Taille Slot | Dual-slot | Dual-slot |
---|---|---|
Longueur | 267 mm 10.5 inches | 267 mm 10.5 inches |
Taille | 111 mm 4.4 inches | 112 mm 4.4 inches |
TDP | 260 W | 230 W |
Suggestions pour le PSU | 600 W | 550 W |
Sorties | 4x DisplayPort1x USB Type-C | 4x DisplayPort |
Connecteurs de puissance | 1x 6-pin + 1x 8-pin | 1x 8-pin |
Board number | PG150 SKU 510 | — |
DirectX | 12 Ultimate (12_2) | 12 Ultimate (12_2) |
---|---|---|
OpenGL | 4.6 | 4.6 |
OpenCL | 3.0 | 3.0 |
Vulkan | 1.2 | 1.2 |
CUDA | 7.5 | 8.6 |
Shader model | 6.6 | 6.6 |