NVIDIA vs AMD for tensorflow

As mentioned in the z440 post, the workstation comes with a NVIDIA Quadro K5200. I've been a happy user of AMD hardware since Radeon HD 4850 (upgraded 5870 and R9 390 later). Unfortunately, tensorflow only supports Cuda - possibly due to missing OpenCL support in Eigen. Is it worth switching just for that? I did a few experiments:

CPU (8core Xeon E5-1680v3)
real 8m24.991s
user 104m13.048s
sys  0m38.772s

All CPUs ~80% usage

GPU (NVIDIA Quadro K5200)
real 2m12.990s
user 2m47.136s
sys  0m30.500s
All CPUs ~20% usage

I think it's worth it - about 4x improvement by using the GPU vs a high end 8 core Xeon. And this GPU is 2 generations back - a GTX 1080 or newer will probably give an even higher benefit.
So for now, I'll stick with NVIDIA.

Comments

Post a Comment

Popular posts from this blog

Slow ethernet on RT-AC68U

 Upgrading my home internet connection from 100 to 600Mbit didn't make as much of a difference as I expected. speedtest.net reported barely over 300Mbit on ethernet or WiFi. Connecting directly through the modem (some Huawei) gave the expected speeds. Turns out, the Asus RT-AC68U has a setting in Advanced Settings / LAN: "NAT Acceleration" was disabled and setting it to "Auto" (the only other option) enabled "CTF (Cut Through Forwarding)". This was all that was needed to get the same speeds through the RT-AC68U as directly from the modem.
Read more

Debian Testing installation notes

Main installation went smoothly, I installed gnome 3 and xfce, at boot it defaults to gnome. No sudo The installation process prompts for a root and user password and without thinking much, I set both. This resulted in my default user not being allowed to run commands with sudo. Add the default user to the sudoers group: # log as root, as the user is not allowed to use sudo yet $ su # add user to the sudo group $ /sbin/usermod -a -G sudo <username> It is necessary to log out for changes to take effect. No sound My Asus Xonar is supported on Linux (as of early 2019), however the installation default sets the card to stereo and probably uses a different output plug. In my case, this is easily solved by running alsamixer and changing the analog out settings to multichannel. Then store the settings:  $ /sbin/alsactl --file ~/.config/asound.state store and I make it the system default:  $ sudo cp ~/.config/asound.state /var/lib/alsa/asound.state After reboot, so
Read more