3 ms·
That is a sexy machine. Does anyone who understands GPUs know if all the NVIDIA do-linear-algebra-really-freaking-fast drivers that everyone uses for deep lear
by paultopia 8y ago
That is a sexy machine. Does anyone who understands GPUs know if all the NVIDIA do-linear-algebra-really-freaking-fast drivers that everyone uses for deep learning will run properly on the RTX 2070?
- TheAceOfHearts 8y agoI won't claim to understand GPUs, but I've read a LOT of comments from people complaining about NVIDIA cards on Linux. As I understand it, AMD is pushing more open standards, although I don't know how it compares in terms of performance. I actually have a tab open right now with the ROCm [0] docs, since I was looking to learn a bit more about what they were up to. They supports Tensorflow, although I don't think they're first-class citizens like NVIDIA. [0] https://rocm.github.io/ https://rocm.github.io/
- currymj 8y agoYes, definitely. You can see it among the benchmarks here. https://www.pugetsystems.com/labs/hpc/RTX-2080Ti-with-NVLINK---TensorFlow-Performance-Includes-Comparison-with-GTX-1080Ti-RTX-2070-2080-2080Ti-and-Titan-V-1267/#big-lstm-gtx-1080ti-rtx-2070-rtx-2080-rtx-2080ti-titan-v-tensorflow-training-performance-wordssecond https://www.pugetsystems.com/labs/hpc/RTX-2080Ti-with-NVLINK... It's the slowest tested obviously, mainly because of limited memory (so you have to put less on the device and compute less in parallel). But it would still be a huge speed up over CPU, like night and day. And "as fast as a 1080Ti" is really, really great. I will add, Pop!_OS comes with the latest NVIDIA drivers and lets you apt install the latest CUDA and even Tensorflow without any rigamarole.
- paultopia 8y agoCool, thanks!