NVIDIA A100 SXM4 40 GB vs NVIDIA GeForce RTX 3080 Ti
Graphics Processor
| GPU model | GA100 | GA102 |
|---|---|---|
| Architecture | Ampere | Ampere |
| Foundry | TSMC | TSMC |
| Manufacturing process | 7 nm | 7 nm |
| Number of transistors | 54,200 million | 40,000 million |
| Die size | 826 mm² | 627 mm² |
| GPU variant | — | GA102-300-A1 |
Graphics Card
| Release date | May 14th, 2020 | 2020 |
|---|---|---|
| Generation | Tesla | GeForce 30 |
| Production | Active | Unreleased |
| Interface | PCIe 4.0 x16 | PCIe 4.0 x16 |
| Launch price | — | 1,399 USD |
Clocks
| GPU clock | 1410 MHz | — |
|---|---|---|
| Memory clock | 1215 MHz 2.4 Gbps effective | 2375 MHz 19000 MHz effective |
| Base clock | 1095 MHz | 1410 MHz |
| Boost clock | 1410 MHz | 1740 MHz |
Memory Configuration
| Memory size | 40 GB | 12 GB |
|---|---|---|
| Memory type | HBM2E | GDDR6X |
| Memory bus width | 5120 bit | 384 bit |
| Bandwidth | 1,555 GB/s | 912.0 GB/s |
Render Configuration
| Shading units | 6912 | 5120 |
|---|---|---|
| TMUs | 432 | 320 |
| ROPs | 160 | 96 |
| SM count | 108 | 80 |
| Tensor cores | 432 | 640 |
| Cache L1 | 192 KB (per SM) | 192 KB (per SM) |
| Cache L2 | 40 MB | 6 MB |
| RT cores | — | 160 |
Performance
| Pixel rate | 225.6 GPixel/s | 167.0 GPixel/s |
|---|---|---|
| Texture rate | 609.1 GTexel/s | 556.8 GTexel/s |
| FP16 (half) performance | 77.97 TFLOPS (4:1) | 35.64 TFLOPS (2:1) |
| FP32 (float) performance | 19.49 TFLOPS | 17.82 TFLOPS |
| FP64 (double) performance | 9.746 TFLOPS (1:2) | 556.8 GFLOPS (1:32) |
Dimensions & Outputs
| Slot width | IGP | Dual-slot |
|---|---|---|
| TDP | 400 W | 320 W |
| Suggested PSU | 800 W | 700 W |
| Outputs | No outputs | 1x HDMI3x DisplayPort1x USB Type-C |
| Power connectors | None | 1x 12-pin |
| Length | — | 267 mm 10.5 inches |
| Board number | — | PG132 SKU 20 |
API Support & Features
| DirectX | — | 12 Ultimate (12_2) |
|---|---|---|
| OpenGL | — | 4.6 |
| OpenCL | 3.0 | 2.0 |
| Vulkan | — | 1.2.140 |
| CUDA | 8.0 | 8.5 |
| Shader model | — | 6.5 |
