GPU model | Navi 21 | TU102 |
---|---|---|
Architecture | RDNA 2.0 | Turing |
Foundry | TSMC | TSMC |
Manufacturing process | 7 nm | 12 nm |
Number of transistors | 26,800 million | 18,600 million |
Die size | 520 mm² | 754 mm² |
GPU variant | — | TU102-875-A1 |
Release date | Jun 8th, 2021 | Aug 13th, 2018 |
---|---|---|
Generation | Radeon Pro | Quadro |
Production | Active | Active |
Launch price | 2,249 USD | 9,999 USD |
Interface | PCIe 4.0 x16 | PCIe 3.0 x16 |
Base clock | 2075 MHz | 1395 MHz |
---|---|---|
Boost clock | 2320 MHz | 1770 MHz |
Memory clock | 2000 MHz 16 Gbps effective | 1750 MHz 14 Gbps effective |
Memory size | 32 GB | 48 GB |
---|---|---|
Memory type | GDDR6 | GDDR6 |
Memory bus width | 256 bit | 384 bit |
Bandwidth | 512.0 GB/s | 672.0 GB/s |
Shading units | 3840 | 4608 |
---|---|---|
TMUs | 240 | 288 |
ROPs | 96 | 96 |
Compute units | 60 | — |
RT cores | 60 | 72 |
gpu.details.l0-cache | 32 KB per WGP | — |
Cache L1 | 128 KB per Array | 64 KB (per SM) |
Cache L2 | 4 MB | 6 MB |
gpu.details.l3-cache | 128 MB | — |
SM count | — | 72 |
Tensor cores | — | 576 |
Pixel rate | 222.7 GPixel/s | 169.9 GPixel/s |
---|---|---|
Texture rate | 556.8 GTexel/s | 509.8 GTexel/s |
FP16 (half) performance | 35.64 TFLOPS (2:1) | 32.62 TFLOPS (2:1) |
FP32 (float) performance | 17.82 TFLOPS | 16.31 TFLOPS |
FP64 (double) performance | 1,114 GFLOPS (1:16) | 509.8 GFLOPS (1:32) |
Slot width | Dual-slot | Dual-slot |
---|---|---|
Length | 267 mm 10.5 inches | 267 mm 10.5 inches |
Width | 120 mm 4.7 inches | 111 mm 4.4 inches |
Height | 50 mm 2 inches | — |
TDP | 250 W | 260 W |
Suggested PSU | 600 W | 600 W |
Outputs | 6x mini-DisplayPort | 4x DisplayPort1x USB Type-C |
Power connectors | 1x 6-pin + 1x 8-pin | 1x 6-pin + 1x 8-pin |
Part number | 102-D16487 | — |
Board number | D164 | — |
DirectX | 12 Ultimate (12_2) | 12 Ultimate (12_2) |
---|---|---|
OpenGL | 4.6 | 4.6 |
OpenCL | 2.1 | 3.0 |
Vulkan | 1.2 | 1.2 |
Shader model | 6.5 | 6.6 |
CUDA | — | 7.5 |