Main Page
|
|
Lab Hardware
|
|
Lab Software
|
In the instructions below, substitute your UW Net ID wherever you see "uwnetid".
Last updated: 30 Jan 2023
GPU Overview
Several School of Engineering and Technology (SET) computers contain Graphics Processing Unit cards, or GPUs.
As of 2019, there are two major GPU vendors: Nvidia and AMD. Nvidia developed and promoted the use of GPUs for areas other than graphics processing. In the past, fast rendering of computer-generated images, especially in games, was the mainstay of GPUs. Some people decided to try general-purpose computing on GPUs (GPGPU), but a lack of standards hampered its adoption. OpenCL from Khronos Group and CUDA from Nvidia are the two most popular GPGPU interfaces in the industry.
Other people realized that the massive parallel-processing of graphics programmable or stream processors and the fast graphics RAM that they access could be used for physics engines simulations (again primarily for games) and later for machine learning. Nvidia's CUDA libraries and tools remain very popular, and have been integrated into machine learning tools such as tensorflow. Nvidia GPUs are the only official "CUDA-enabled" products, but other tools can use the CUDA application programming interfaces (APIs).
AMD returned to the central processing unit (CPU) and GPU market in 2017 with their Ryzen and Epyc CPUs as well as Radeon GPUs. Instead of re-inventing the wheel, AMD chose to implement OpenCL as Radeon Open Compute, or ROCm, and port CUDA source code to more portable source code. That way, they can offer CUDA tools such as tensorflow, recompiled to run on AMD GPUs (and AMD and Intel CPUs) instead of just Nvidia GPUs.
The UW also offers a web page on GPUs for Machine Learning which describes GPU use and references GPU capabilities from UW Hyak, Commercial Cloud providers, and Google Colab.
SET GPUs
| Company | Model | Cores | RAM | Qty | Location | GPU Activity | Notes |
|---|---|---|---|---|---|---|---|
| Nvidia | RTX 4090 | 16384 | 24GB | 1 | g4090a.d.insttech.washington.edu | nvidia-smi | Available by 1 Mar 2023.Use Husky OnNet for off-campus access. |
| RTX 2080 Ti | 4352 | 11GB | 1 | g2080tic.d.insttech.washington.edu | Use Husky OnNet for off-campus access. | ||
| 1 | g2080tia.d.insttech.washington.edu | ||||||
| 1 | g2080tib.d.insttech.washington.edu | ||||||
| RTX 2080 | 2944 | 8GB | 1 | g2080b.d.insttech.washington.edu | |||
| 1 | g2080c.d.insttech.washington.edu | ||||||
| 1 | g2080d.d.insttech.washington.edu | ||||||
| 1 | g2080e.d.insttech.washington.edu | ||||||
| 1 | g2080f.d.insttech.washington.edu | ||||||
| 1 | g2080g.d.insttech.washington.edu | ||||||
| RTX 2070 | 2304 | 8GB | 1 | g2070a.d.insttech.washington.edu | |||
| 1 | g2070b.d.insttech.washington.edu | ||||||
| 1 | g2070c.d.insttech.washington.edu | ||||||
| 1 | g2070d.d.insttech.washington.edu | ||||||
| 1 | g2070e.d.insttech.washington.edu | ||||||
| 1 | g2070f.d.insttech.washington.edu | ||||||
| 1 | g2070g.d.insttech.washington.edu | ||||||
| 1 | g2070h.d.insttech.washington.edu | ||||||
| GeForce GTX Titan Xp | 3840 | 12GB | 1 | mcs2.insttech.washington.edu | |||
| 1 | mcs3.d.insttech.washington.edu | ||||||
| AMD | Radeon RX Vega 56 | 3584 | 8GB | 1 | mcs1.insttech.washington.edu | Use
gpu command args For example, if you use python3 and the script you use is convolutional.py: gpu python3 convolutional.py |
Troubleshooting Problems
- Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
If this is on mcs1, use the instructions found in the notes section.
Change Log
| 30 Jan 2023 | Added g4090a's RTX 4090 entry for computer available by 1 Mar 2023. |
| 20 Jan 2022 | Removed dead g2080a and gvega64 GPUs and dead computer hosting cesgpu. |
| 20 Feb 2020 | Added g2080tic. |
| 30 Jul 2019 | Added many more Nvidia GPUs. |
| 17 Jul 2019 | Added gvega64 |
| 15 Apr 2019 | Original file |
Hours | Support Information | News |
Policies | Emergencies