NVIDIA A100 PCIe vs NVIDIA Tesla T4
Grafikprozessor
| GPU Name | GA100 | TU104 |
|---|---|---|
| Architektur | Ampere | Turing |
| Hersteller | TSMC | TSMC |
| Prozessgröße | 7 nm | 12 nm |
| Transistoren | 54,200 million | 13,600 million |
| Chipgröße | 826 mm² | 545 mm² |
| GPU Variante | — | TU104-895-A1 |
Grafikkarte
| Veröffentlichungsdatum | Jun 22nd, 2020 | Sep 13th, 2018 |
|---|---|---|
| Generation | Tesla | Tesla |
| Produktion | Active | Active |
| Businterface | PCIe 4.0 x16 | PCIe 3.0 x16 |
Takt-Geschwindigkeiten
| GPU-Taktung | 1410 MHz | — |
|---|---|---|
| Speicher-Taktung | 1215 MHz 2.4 Gbps effective | 1250 MHz 10 Gbps effective |
| Basis-Taktfrequenz | 765 MHz | 585 MHz |
| Boost-Takt | 1410 MHz | 1590 MHz |
Speicher
| Speichergröße | 40 GB | 16 GB |
|---|---|---|
| Speichertyp | HBM2E | GDDR6 |
| Speicherbus | 5120 bit | 256 bit |
| Bandbreite | 1,555 GB/s | 320.0 GB/s |
Render-Konfiguration
| Shading-Einheiten | 6912 | 2560 |
|---|---|---|
| TMUs | 432 | 160 |
| ROPs | 160 | 64 |
| Anzahl SM | 108 | 40 |
| Tensor Cores | 432 | 320 |
| Cache L1 | 192 KB (per SM) | 64 KB (per SM) |
| Cache L2 | 40 MB | 4 MB |
| RT-Kerne | — | 40 |
Theoretische Leistung
| Pixelrate | 225.6 GPixel/s | 101.8 GPixel/s |
|---|---|---|
| Texturrate | 609.1 GTexel/s | 254.4 GTexel/s |
| FP16 (half) Leistung | 77.97 TFLOPS (4:1) | 65.13 TFLOPS (8:1) |
| FP32 (float) Leistung | 19.49 TFLOPS | 8.141 TFLOPS |
| FP64 (double) Leistung | 9.746 TFLOPS (1:2) | 254.4 GFLOPS (1:32) |
Boarddesign
| Steckplatzbreite | Dual-slot | Single-slot |
|---|---|---|
| Länge | 267 mm 10.5 inches | 168 mm 6.6 inches |
| TDP | 250 W | 70 W |
| Vorgeschlagenes Netzteil | 600 W | 250 W |
| Ausgabe | No outputs | No outputs |
| Stromverbindung | 8-pin EPS | None |
| Board-Nummer | P1001 SKU 200 | PG183 SKU 200 |
Grafik-Features
| DirectX | — | 12 Ultimate (12_2) |
|---|---|---|
| OpenGL | — | 4.6 |
| OpenCL | 3.0 | 3.0 |
| Vulkan | — | 1.2 |
| CUDA | 8.0 | 7.5 |
| Shader-Modell | — | 6.6 |
