GPU model | Aldebaran | TU104 |
---|---|---|
GPU variant | Aldebaran | TU104-895-A1 |
Architecture | CDNA 2.0 | Turing |
Foundry | TSMC | TSMC |
Manufacturing process | 6 nm | 12 nm |
Number of transistors | 58,200 million | 13,600 million |
Die size | unknown | 545 mm² |
Release date | Dec 2021 | Sep 13th, 2018 |
---|---|---|
Generation | Radeon Instinct | Tesla |
Production | Active | Active |
Interface | PCIe 4.0 x16 | PCIe 3.0 x16 |
Base clock | 1000 MHz | 585 MHz |
---|---|---|
Boost clock | 1700 MHz | 1590 MHz |
Memory clock | 1600 MHz 3.2 Gbps effective | 1250 MHz 10 Gbps effective |
Memory size | 64 GB | 16 GB |
---|---|---|
Memory type | HBM2e | GDDR6 |
Memory bus width | 4096 bit | 256 bit |
Bandwidth | 1,638 GB/s | 320.0 GB/s |
Shading units | 6656 | 2560 |
---|---|---|
TMUs | 416 | 160 |
ROPs | 0 | 64 |
Compute units | 104 | — |
gpu.details.mcm | 1 | — |
Cache L1 | 16 KB (per CU) | 64 KB (per SM) |
Cache L2 | 16 MB | 4 MB |
SM count | — | 40 |
Tensor cores | — | 320 |
RT cores | — | 40 |
Pixel rate | 0 MPixel/s | 101.8 GPixel/s |
---|---|---|
Texture rate | 707.2 GTexel/s | 254.4 GTexel/s |
FP16 (half) performance | 181.0 TFLOPS (8:1) | 65.13 TFLOPS (8:1) |
FP32 (float) performance | 22.63 TFLOPS | 8.141 TFLOPS |
FP64 (double) performance | 22.63 TFLOPS (1:1) | 254.4 GFLOPS (1:32) |
Slot width | Dual-slot | Single-slot |
---|---|---|
Length | 267 mm 10.5 inches | 168 mm 6.6 inches |
Width | 111 mm 4.4 inches | — |
TDP | 300 W | 70 W |
Suggested PSU | 700 W | 250 W |
Outputs | No outputs | No outputs |
Power connectors | 2x 8-pin | None |
Board number | — | PG183 SKU 200 |
DirectX | — | 12 Ultimate (12_2) |
---|---|---|
OpenGL | — | 4.6 |
OpenCL | 3.0 | 3.0 |
Vulkan | — | 1.2 |
Shader model | — | 6.6 |
CUDA | — | 7.5 |