GPU model | Picasso | TU104 |
---|---|---|
Architecture | GCN 5.0 | Turing |
Foundry | GlobalFoundries | TSMC |
Manufacturing process | 14 nm | 12 nm |
Number of transistors | 4,940 million | 13,600 million |
Die size | 210 mm² | 545 mm² |
GPU variant | — | TU104-895-A1 |
Release date | 2019 | — |
---|---|---|
Generation | Picasso (Vega) | — |
Production | Active | — |
Interface | IGP | — |
Predecessor | Raven Ridge | — |
Successor | Renoir | — |
Base clock | 300 MHz | 585 MHz |
---|---|---|
Boost clock | 1000 MHz | 1590 MHz |
Memory clock | System Shared | 1250 MHz 10 Gbps effective |
Memory size | System Shared | 16 GB |
---|---|---|
Memory type | System Shared | GDDR6 |
Memory bus width | System Shared | 256 bit |
Bandwidth | System Dependent | 320.0 GB/s |
Shading units | 192 | 2560 |
---|---|---|
TMUs | 12 | 160 |
ROPs | 4 | 64 |
Compute units | 3 | — |
SM count | — | 40 |
Tensor cores | — | 320 |
RT cores | — | 40 |
Cache L1 | — | 64 KB (per SM) |
Cache L2 | — | 4 MB |
Pixel rate | 4.000 GPixel/s | 101.8 GPixel/s |
---|---|---|
Texture rate | 12.00 GTexel/s | 254.4 GTexel/s |
FP16 (half) performance | 768.0 GFLOPS (2:1) | 65.13 TFLOPS (8:1) |
FP32 (float) performance | 384.0 GFLOPS | 8.141 TFLOPS |
FP64 (double) performance | 24.00 GFLOPS (1:16) | 254.4 GFLOPS (1:32) |
Slot width | IGP | Single-slot |
---|---|---|
TDP | 10 W | 70 W |
Outputs | No outputs | No outputs |
Power connectors | None | None |
Length | — | 168 mm 6.6 inches |
Suggested PSU | — | 250 W |
Board number | — | PG183 SKU 200 |
DirectX | 12 (12_1) | 12 Ultimate (12_2) |
---|---|---|
OpenGL | 4.6 | 4.6 |
OpenCL | 2.1 | 3.0 |
Vulkan | 1.2 | 1.2 |
Shader model | 6.4 | 6.6 |
CUDA | — | 7.5 |
Release date | — | Sep 13th, 2018 |
---|---|---|
Generation | — | Tesla |
Production | — | Active |
Interface | — | PCIe 3.0 x16 |