GPU model | Cape Verde | GA100 |
---|---|---|
GPU variant | Cape Verde PRX | — |
Architecture | GCN 1.0 | Ampere |
Foundry | TSMC | TSMC |
Manufacturing process | 28 nm | 7 nm |
Number of transistors | 1,500 million | 54,200 million |
Die size | 123 mm² | 826 mm² |
Release date | Jun 30th, 2016 | May 14th, 2020 |
---|---|---|
Generation | Arctic Islands | GRID |
Production | End-of-life | Active |
Interface | PCIe 3.0 x16 | PCIe 4.0 x16 |
Predecessor | Pirate Islands | — |
Successor | Polaris | — |
GPU clock | 925 MHz | — |
---|---|---|
Memory clock | 1125 MHz 4.5 Gbps effective | 1215 MHz 2.4 Gbps effective |
Base clock | — | 900 MHz |
Boost clock | — | 1005 MHz |
Memory size | 2 GB | 48 GB |
---|---|---|
Memory type | GDDR5 | HBM2E |
Memory bus width | 128 bit | 6144 bit |
Bandwidth | 72.00 GB/s | 1,866 GB/s |
Shading units | 512 | 6912 |
---|---|---|
TMUs | 32 | 432 |
ROPs | 16 | 192 |
Compute units | 8 | — |
Cache L1 | 16 KB (per CU) | 192 KB (per SM) |
Cache L2 | 256 KB | 48 MB |
SM count | — | 108 |
Tensor cores | — | 432 |
Pixel rate | 14.80 GPixel/s | 193.0 GPixel/s |
---|---|---|
Texture rate | 29.60 GTexel/s | 434.2 GTexel/s |
FP32 (float) performance | 947.2 GFLOPS | 13.89 TFLOPS |
FP64 (double) performance | 59.20 GFLOPS (1:16) | 6.947 TFLOPS (1:2) |
FP16 (half) performance | — | 55.57 TFLOPS (4:1) |
Slot width | Dual-slot | IGP |
---|---|---|
TDP | 65 W | 400 W |
Suggested PSU | 250 W | 800 W |
Outputs | No outputs | No outputs |
Power connectors | 1x 6-pin | None |
Board number | C750 C906 | — |
DirectX | 12 (11_1) | — |
---|---|---|
OpenGL | 4.6 | — |
OpenCL | 1.2 | 3.0 |
Vulkan | 1.2 | — |
Shader model | 5.1 | — |
CUDA | — | 8.0 |