GPU model | Scrapper | GA100 |
---|---|---|
Architecture | TeraScale 3 | Ampere |
Foundry | TSMC | TSMC |
Manufacturing process | 32 nm | 7 nm |
Number of transistors | 1,303 million | 54,200 million |
Die size | 246 mm² | 826 mm² |
Release date | May 23rd, 2013 | — |
---|---|---|
Generation | Richland (HD 8000) | — |
Production | End-of-life | — |
Interface | IGP | — |
Predecessor | Trinity | — |
Successor | Kabini | — |
Base clock | 720 MHz | 900 MHz |
---|---|---|
Boost clock | 844 MHz | 1005 MHz |
Memory clock | System Shared | 1215 MHz 2.4 Gbps effective |
Memory size | System Shared | 48 GB |
---|---|---|
Memory type | System Shared | HBM2E |
Memory bus width | System Shared | 6144 bit |
Bandwidth | System Dependent | 1,866 GB/s |
Shading units | 256 | 6912 |
---|---|---|
TMUs | 16 | 432 |
ROPs | 8 | 192 |
Compute units | 4 | — |
SM count | — | 108 |
Tensor cores | — | 432 |
Cache L1 | — | 192 KB (per SM) |
Cache L2 | — | 48 MB |
Pixel rate | 6.752 GPixel/s | 193.0 GPixel/s |
---|---|---|
Texture rate | 13.50 GTexel/s | 434.2 GTexel/s |
FP32 (float) performance | 432.1 GFLOPS | 13.89 TFLOPS |
FP16 (half) performance | — | 55.57 TFLOPS (4:1) |
FP64 (double) performance | — | 6.947 TFLOPS (1:2) |
Slot width | IGP | IGP |
---|---|---|
TDP | 65 W | 400 W |
Outputs | No outputs | No outputs |
Suggested PSU | — | 800 W |
Power connectors | — | None |
DirectX | 11.2 (11_0) | — |
---|---|---|
OpenGL | 4.4 | — |
OpenCL | 1.2 | 3.0 |
Vulkan | — | — |
Shader model | 5.0 | — |
CUDA | — | 8.0 |
Release date | — | May 14th, 2020 |
---|---|---|
Generation | — | GRID |
Production | — | Active |
Interface | — | PCIe 4.0 x16 |