AMD Radeon Instinct MI50 vs NVIDIA Tesla P40
Graphics Processor
| GPU model | Vega 20 | GP102 |
|---|---|---|
| GPU variant | Vega 20 GLXT | — |
| Architecture | GCN 5.1 | Pascal |
| Foundry | TSMC | TSMC |
| Manufacturing process | 7 nm | 16 nm |
| Number of transistors | 13,230 million | 11,800 million |
| Die size | 331 mm² | 471 mm² |
Graphics Card
| Release date | Nov 18th, 2018 | Sep 13th, 2016 |
|---|---|---|
| Generation | Radeon Instinct | Tesla |
| Production | Active | End-of-life |
| Interface | PCIe 4.0 x16 | PCIe 3.0 x16 |
| Launch price | — | 5,699 USD |
Clocks
| Base clock | 1200 MHz | 1303 MHz |
|---|---|---|
| Boost clock | 1746 MHz | 1531 MHz |
| Memory clock | 1000 MHz 2 Gbps effective | 1808 MHz 14.5 Gbps effective |
Memory Configuration
| Memory size | 16 GB | 24 GB |
|---|---|---|
| Memory type | HBM2 | GDDR5X |
| Memory bus width | 4096 bit | 384 bit |
| Bandwidth | 1,024 GB/s | 694.3 GB/s |
Render Configuration
| Shading units | 3840 | 3840 |
|---|---|---|
| TMUs | 240 | 240 |
| ROPs | 64 | 96 |
| Compute units | 60 | — |
| Cache L1 | 16 KB (per CU) | 48 KB (per SM) |
| Cache L2 | 4 MB | 3 MB |
| SM count | — | 30 |
Performance
| Pixel rate | 111.7 GPixel/s | 147.0 GPixel/s |
|---|---|---|
| Texture rate | 419.0 GTexel/s | 367.4 GTexel/s |
| FP16 (half) performance | 26.82 TFLOPS (2:1) | 183.7 GFLOPS (1:64) |
| FP32 (float) performance | 13.41 TFLOPS | 11.76 TFLOPS |
| FP64 (double) performance | 6.705 TFLOPS (1:2) | 367.4 GFLOPS (1:32) |
Dimensions & Outputs
| Slot width | Dual-slot | Dual-slot |
|---|---|---|
| Length | 267 mm 10.5 inches | 267 mm 10.5 inches |
| TDP | 300 W | 250 W |
| Suggested PSU | 700 W | 600 W |
| Outputs | 1x mini-DisplayPort | No outputs |
| Power connectors | 1x 6-pin + 1x 8-pin | 1x 6-pin + 1x 8-pin |
| Width | — | 111 mm 4.4 inches |
| Board number | — | PG610 SKU 200 |
API Support & Features
| DirectX | 12 (12_1) | 12 (12_1) |
|---|---|---|
| OpenGL | 4.6 | 4.6 |
| OpenCL | 2.1 | 3.0 |
| Vulkan | 1.2 | 1.2 |
| Shader model | 6.4 | 6.4 |
| CUDA | — | 6.1 |
