AMD Radeon Instinct MI8 vs NVIDIA Tesla T4
What is the difference between AMD Radeon Instinct MI8 and NVIDIA Tesla T4. Find out which graphics card has better performance.
Graphics Processor (GPU)
| Fiji | GPU Name | TU104 |
| Fiji XT | Model number | TU104-895-A1 |
| GCN 3.0 | Architecture | Turing |
| TSMC | Foundry | TSMC |
| 28 nm | Process Size | 12 nm |
| 8,900 million | Transistors | 13,600 million |
| 596 mm² | Die Size | 545 mm² |
Graphics Card
| Dec 12th, 2016 | Release Date | Sep 13th, 2018 |
| Radeon Instinct (MIx) | Family | Tesla (Txx) |
| Active | Production | Active |
| PCIe 3.0 x16 | Bus Interface | PCIe 3.0 x16 |
Memory
| 4 GB | Memory Size | 16 GB |
| HBM | Memory Type | GDDR6 |
| 4096 bit | Memory Bus | 256 bit |
| 512.0 GB/s | Bandwidth | 320.0 GB/s |
Performance
| 64.00 GPixel/s | Pixel fillrate | 101.8 GPixel/s |
| 256.0 GTexel/s | Texture fillrate | 254.4 GTexel/s |
| 8.192 TFLOPS (1:1) | FP16 (half) performance | 65.13 TFLOPS (8:1) |
| 8.192 TFLOPS | FP32 (float) performance | 8.141 TFLOPS |
| 512.0 GFLOPS (1:16) | FP64 (double) performance | 254.4 GFLOPS (1:32) |
Clock Speeds
| 500 MHz 1000 Mbps effective | Memory Clock | 1250 MHz 10 Gbps effective |
Render Config
| 4096 | Shading Units | 2560 |
| 256 | TMUs | 160 |
| 64 | ROPs | 64 |
| 16 KB (per CU) | L1 Cache | 64 KB (per SM) |
| 2 MB | L2 Cache | 4 MB |
Board Design
| Dual-slot | Slot Width | Single-slot |
| 152 mm 6 inches | Length | 168 mm 6.6 inches |
| 175 W | Thermal design power (TDP) | 70 W |
| 450 W | Suggested PSU | 250 W |
| No outputs | Display Connectors | No outputs |
| 1x 8-pin | Power Connectors | None |
API support
| 12 (12_0) | DirectX | 12 Ultimate (12_2) |
| 4.6 | OpenGL | 4.6 |
| 2.0 | OpenCL | 3.0 |
| 1.2 | Vulkan | 1.2 |
| 6.3 | Shader Model | 6.6 |