GPU Name | Fiji | TU104 |
---|---|---|
GPU Variante | Fiji XT | TU104-895-A1 |
Architektur | GCN 3.0 | Turing |
Hersteller | TSMC | TSMC |
Prozessgröße | 28 nm | 12 nm |
Transistoren | 8,900 million | 13,600 million |
Chipgröße | 596 mm² | 545 mm² |
Veröffentlichungsdatum | Dec 12th, 2016 | Sep 13th, 2018 |
---|---|---|
Generation | Radeon Instinct | Tesla |
Produktion | End-of-life | Active |
Businterface | PCIe 3.0 x16 | PCIe 3.0 x16 |
GPU-Taktung | 1000 MHz | — |
---|---|---|
Speicher-Taktung | 500 MHz 1000 Mbps effective | 1250 MHz 10 Gbps effective |
Basis-Taktfrequenz | — | 585 MHz |
Boost-Takt | — | 1590 MHz |
Speichergröße | 4 GB | 16 GB |
---|---|---|
Speichertyp | HBM | GDDR6 |
Speicherbus | 4096 bit | 256 bit |
Bandbreite | 512.0 GB/s | 320.0 GB/s |
Shading-Einheiten | 4096 | 2560 |
---|---|---|
TMUs | 256 | 160 |
ROPs | 64 | 64 |
Compute Units | 64 | — |
Cache L1 | 16 KB (per CU) | 64 KB (per SM) |
Cache L2 | 2 MB | 4 MB |
Anzahl SM | — | 40 |
Tensor Cores | — | 320 |
RT-Kerne | — | 40 |
Pixelrate | 64.00 GPixel/s | 101.8 GPixel/s |
---|---|---|
Texturrate | 256.0 GTexel/s | 254.4 GTexel/s |
FP16 (half) Leistung | 8.192 TFLOPS (1:1) | 65.13 TFLOPS (8:1) |
FP32 (float) Leistung | 8.192 TFLOPS | 8.141 TFLOPS |
FP64 (double) Leistung | 512.0 GFLOPS (1:16) | 254.4 GFLOPS (1:32) |
Steckplatzbreite | Dual-slot | Single-slot |
---|---|---|
Länge | 152 mm 6 inches | 168 mm 6.6 inches |
Breite | 111 mm 4.4 inches | — |
TDP | 175 W | 70 W |
Vorgeschlagenes Netzteil | 450 W | 250 W |
Ausgabe | No outputs | No outputs |
Stromverbindung | 1x 8-pin | None |
Board-Nummer | — | PG183 SKU 200 |
DirectX | 12 (12_0) | 12 Ultimate (12_2) |
---|---|---|
OpenGL | 4.6 | 4.6 |
OpenCL | 2.0 | 3.0 |
Vulkan | 1.2 | 1.2 |
Shader-Modell | 6.3 | 6.6 |
CUDA | — | 7.5 |