Simultaneous Multi-Threading in World Community Grid (mixed projects, July, 2020).

If you read this, you might be wondering what the results are. It has been 8 days and I'm going to take a look at how the machine without SMT enabled on its Ryzen 7 2700 has performed. That said, there is a snag: Windows decided to update the OS on all four of my servers about two days ago, so they've been sat doing nothing for a whole day, before I checked the wattage meter to determine if they were still doing work.


So, uh, thanks Windows. And another reason why I should port these servers over to Ubuntu: on top of the existing reason of better scheduling and less overheads. Maybe soon. Maybe.


Anyway, I will wait another week before I draw any conclusions, but here is some preliminary results:



Ryzen 7 2700, 3025 MHz all core, 2666 C16-18-18, SMT On

Ryzen 7 2700, 3025 MHz all core, 2400 C16-16-16, SMT OFF

Ignoring today's date on the above graph, it hasn't even got started on that yet, and also if you ignore the 16th of July: that was when Windows had decided to restart the servers. I disabled on the 12th, so the figures before that are with SMT.


But Sash, the RAM configurations are different!

Yes, I know. And actually, come to think of it: I am going to run a week without SMT on the same system, then re-enable it for a week and do the test that way, but it just takes ages and I procrastinate, don't judge me! Anyway: the RAM speed isn't that different so it wouldn't make a huge difference, but you can see from the results that SMT enabled is getting a bit more work done, keep in mind the 266 MHz higher data rate on the RAM at the same primary CAS latency, though. Make of that what you will.


But Sash, projects are not all the same!

Ssssh. Jeese. Stop with your legitimate criticism, I'm trying to wing SCIENCE! here. Okay, look, you're right. If I want to do this properly, I'm going to run the same project on the machine I stated above, and test all the different projects individually. It'll take ages, but it will provide the most conclusive data.


So far...

I think, overally, by 'eyeballing' it: SMT is an improvement for these mixed projects. For reference, as of typing this, my machines are set to all available projects. They are:


With all that in mind, I must now wait another week for a stable, non-SMT average result from the Mixed Workload phase of the test!


Or maybe five days?


Three days?


24 hours?


( ͡◉ ͜ʖ ͡◉)

4 views

©2020 by Sashleycat (Read the Legal stuff)