Nom du GPU | Lexa | GA102 |
---|---|---|
GPU variant | Lexa PRO (215-0904018) | GA102-225-A1 |
Architecture | GCN 4.0 | Ampere |
Fonderie | GlobalFoundries | Samsung |
Taille du processus | 14 nm | 8 nm |
Transistors | 2,200 million | 28,300 million |
Taille des matrices | 103 mm² | 628 mm² |
Date de publication | Apr 20th, 2017 | Jan 2022 |
---|---|---|
Génération | Polaris | GeForce 30 |
Production | End-of-life | Active |
Prix de lancement | 79 USD | 1,199 USD |
Bus interface | PCIe 3.0 x8 | PCIe 4.0 x16 |
Prédécesseur | Arctic Islands | GeForce 20 |
Successeur | Vega | — |
gpu.details.availability | — | 2022 |
Critiques | — | 50 in our database |
Base clock | 1100 MHz | 1335 MHz |
---|---|---|
Boost clock | 1183 MHz | 1665 MHz |
Mémoire clock | 1750 MHz 7 Gbps effective | 1188 MHz 19 Gbps effective |
Taille de la mémoire | 2 GB | 20 GB |
---|---|---|
Type de mémoire | GDDR5 | GDDR6X |
Bus mémoire | 128 bit | 320 bit |
Bande passante | 112.0 GB/s | 760.3 GB/s |
Shading units | 512 | 10240 |
---|---|---|
TMUs | 32 | 320 |
ROPs | 16 | 112 |
Compute units | 8 | — |
Cache L1 | 16 KB (per CU) | 128 KB (per SM) |
Cache L2 | 512 KB | 6 MB |
Comptage SM | — | 80 |
Tensor cores | — | 320 |
RT cores | — | 80 |
Taux de pixel | 18.93 GPixel/s | 186.5 GPixel/s |
---|---|---|
Taux de texture | 37.86 GTexel/s | 532.8 GTexel/s |
FP16 (half) performance | 1,211 GFLOPS (1:1) | 34.10 TFLOPS (1:1) |
FP32 (float) performance | 1,211 GFLOPS | 34.10 TFLOPS |
FP64 (double) performance | 75.71 GFLOPS (1:16) | 532.8 GFLOPS (1:64) |
Taille Slot | Dual-slot | Dual-slot |
---|---|---|
Longueur | 145 mm 5.7 inches | 285 mm 11.2 inches |
TDP | 50 W | 350 W |
Suggestions pour le PSU | 250 W | 750 W |
Sorties | 1x DVI1x HDMI1x DisplayPort | 1x HDMI3x DisplayPort |
Connecteurs de puissance | None | 1x 12-pin |
Board number | D090-01 | — |
Taille | — | 112 mm 4.4 inches |
Taille | — | 40 mm 1.6 inches |
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 | — | 8.6 |