NVIDIA GRID K100 vs NVIDIA Tesla T4
Graphics Processor
| GPU model | GK107 | TU104 |
|---|---|---|
| Architecture | Kepler | Turing |
| Foundry | TSMC | TSMC |
| Manufacturing process | 28 nm | 12 nm |
| Number of transistors | 1,270 million | 13,600 million |
| Die size | 118 mm² | 545 mm² |
| GPU variant | — | TU104-895-A1 |
Graphics Card
| Release date | Jun 28th, 2013 | Sep 13th, 2018 |
|---|---|---|
| Generation | GRID | Tesla |
| Production | End-of-life | Active |
| Launch price | 63 USD | — |
| Interface | PCIe 3.0 x16 | PCIe 3.0 x16 |
Clocks
| GPU clock | 850 MHz | — |
|---|---|---|
| Memory clock | 891 MHz 1782 Mbps effective | 1250 MHz 10 Gbps effective |
| Base clock | — | 585 MHz |
| Boost clock | — | 1590 MHz |
Memory Configuration
| Memory size | 256 MB | 16 GB |
|---|---|---|
| Memory type | DDR3 | GDDR6 |
| Memory bus width | 128 bit | 256 bit |
| Bandwidth | 28.51 GB/s | 320.0 GB/s |
Render Configuration
| Shading units | 192 | 2560 |
|---|---|---|
| TMUs | 16 | 160 |
| ROPs | 16 | 64 |
| SMX count | 1 | — |
| Cache L1 | 16 KB (per SMX) | 64 KB (per SM) |
| Cache L2 | 256 KB | 4 MB |
| SM count | — | 40 |
| Tensor cores | — | 320 |
| RT cores | — | 40 |
Performance
| Pixel rate | 3.400 GPixel/s | 101.8 GPixel/s |
|---|---|---|
| Texture rate | 13.60 GTexel/s | 254.4 GTexel/s |
| FP32 (float) performance | 326.4 GFLOPS | 8.141 TFLOPS |
| FP64 (double) performance | 13.60 GFLOPS (1:24) | 254.4 GFLOPS (1:32) |
| FP16 (half) performance | — | 65.13 TFLOPS (8:1) |
Dimensions & Outputs
| Slot width | Dual-slot | Single-slot |
|---|---|---|
| TDP | 130 W | 70 W |
| Suggested PSU | 300 W | 250 W |
| Outputs | No outputs | No outputs |
| Length | — | 168 mm 6.6 inches |
| Power connectors | — | None |
| Board number | — | PG183 SKU 200 |
API Support & Features
| DirectX | 12 (11_0) | 12 Ultimate (12_2) |
|---|---|---|
| OpenGL | 4.6 | 4.6 |
| OpenCL | 3.0 | 3.0 |
| Vulkan | 1.1 | 1.2 |
| CUDA | 3.0 | 7.5 |
| Shader model | 5.1 | 6.6 |
