Amd threadripper query

Hi everyone,

Not directly related to MRtrix, but I’m currently thinking about building a new PC for imaging analysis and the AMD threadripper cpus (3970/3990) look very attractive with their high number of cores. Can anyone comment on their compatability/stability running linux (Ubuntu) and mainstream tools (i.e. mrtrix, freesurfer, ANTS, fsl). I’ve read there are some common linux bugs associated with threadripper, but keen to hear from anyone who has experience with imaging.

Thanks,

Linux has always been ahead of the competition when it comes to supporting the latest processors on the market, so I would not really worry about that. As long as you run a recent Linux distribution that is well supported you should have no problem (Ubuntu 20.04 should play fine with the latest Threadripper CPUs out of the box).

Also, there is no reason to believe that a specific Intel or AMD CPU would be incompatible or unstable with any neuroimaging tools. Just make sure to include an adequate power supply and ample cooling if you are building a PC with a CPU that has a TDP of 280W.

As almost all MRtrix commands are multi-threaded, they will be blazing fast on machines with high core count.

I agree with everything @bjeurissen mentions here, and I’d add that if Linus Torvalds himself uses an AMD ThreadRipper system, I don’t think there’s going to be anything to worry about…

By the way, I’d be keen to hear about these bug reports you mention. Any chance you could link to them here so I can take a look?

As far as I can tell, the only linux issue with the latest Threadripper iteration was fixed 9 months ago: https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.4-4.19-4.14-MCE-Fix-TR.

1 Like

Sorry, by common bugs I was just referring to the boot issue with MCE @bjeurissen kindly linked above. It did lead me to read about some previous issues with AMD cpus that made me question whether to just stick with intel-xeon:
https://www.phoronix.com/scan.php?page=news_item&px=Ryzen-Segv-Response
https://arstechnica.com/gadgets/2019/10/how-a-months-old-amd-microcode-bug-destroyed-my-weekend/

Completely agree, my hesitation is more on stability as there are isolated forum posts of seemingly random crash/freeze under high workload. I was interested to hear whether anyone in this community runs a threadripper setup for their analyses, but I imagine 99% of setups use intel.

Thanks for your input,

1 Like

Thanks for confirming. These other issues you link to do sound worrying… But I’ve not encountered similar issues on my system so far – and I run a lot of heavy compile jobs. I’ll double-check the random number generator issue though, since we rely on this functionality in our code…

1 Like

for what it is worth, I use a ryzen 9 3900X (not quite threadripper but AMD) - I have not had any issues so far (about 6 month)

I’d forgotten to mention that I also run an AMD Ryzen 9 3950X 16-Core Processor, with no obvious problems for the last 6 months or so. Also not a ThreadRipper as such, but probably close enough in architecture.

I’ve also just double-checked the random number issue, and it doesn’t seem to be a problem on my system. That doesn’t necessarily mean the RDRAND implementation is working, given that Linux doesn’t rely on that alone as a source of entropy. But it’s certainly not causing trouble for applications on my system. It also seems this was more of a faulty motherboard / BIOS issue, rather than the CPU itself. I expect BIOS updates will be available for all affected motherboards by now…

That’s quite reassuring, thanks for the input. I look forward to building my first AMD system :slight_smile:. The value for money compared to Intel seems too good to pass up.