Page 1 of 1

The effects of HT on folding times

Posted: Thu May 29, 2014 2:25 pm
by Nathan_P
The debate on the benefits of hyper threading for F@H has raged back and forth as long as I have been on the forum (2009) and I saw it mentioned yet again recently but I have never seen anyone test it out. As I have just had to rebuild my 3rd rig I thought it would be the ideal time to do a couple of tests.

The rig consists of dual E5-2670 retail xeons, Asus Z9PE-D16 mobo, 18gb of ram, corsair AX760 psu, a usb2 flash drive running the [H] folding appliance v9 with the kraken. The WU used is P8823, R0, C3,G254

HT on.

Idle power 87w
Folding power 312w
tpf 2:31
PPD 139k

HT off.

Idle power:- 83w
Folding power:- 260w
tpf 3:46
PPD 73k

So that's a tpf that's 50% quicker, 90% more PPD and only 20% more power. Looks like using HT is definitely a must to max performance.
I know its only one wu but I'm not about to start dumping WU just to run tests, i'll run another test in a few days and see if a different WU shows any difference

Re: The effects of HT on folding times

Posted: Thu May 29, 2014 3:52 pm
by 7im
The debate ended with adding the QRB a few years ago. QRB was a tool to end that debate (along with SMP). Run one client on all cores, with HT. It's well documented here.

You should stop by more often. 5 years is a long time. ;)

Re: The effects of HT on folding times

Posted: Thu May 29, 2014 4:22 pm
by Jesse_V
HT is now highly recommended and I expect that you'll see similar results on all of your tests.
I take it you're using the V7 client, 7.4.4, yes?

Re: The effects of HT on folding times

Posted: Thu May 29, 2014 4:43 pm
by Nathan_P
7im, you know full well that I visit here often, and Bruce was only discussing this a few weeks ago

Jesse_V, nope I was using 6.34, for dedicated BA Linux folders v7 doesn't currently offer any benefits

Re: The effects of HT on folding times

Posted: Thu May 29, 2014 5:08 pm
by bruce
The answers to that question has not changed in 5 years, although the hardware has. Allocate all the resources you can to a single WU.

How many FPUs are available? What percentage of the operations depend on saturated FPUs or on other subsections of the CPU (i.e. which FahCore is being used). When you turned HT on/off, did you adjust the number of CPU threads? After making the changes did you give kraken an opportunity to re-optimize the threads? etc.

Using twice as many threads on the same WU is a good idea, though with HT, it's not as significant as having hardware where every thread has the exclusive use of a FPU.

Re: The effects of HT on folding times

Posted: Thu May 29, 2014 5:20 pm
by 7im
I'd like to see bruce's comments. He more likely gave an answer to a question from a newer folder that didn't know yet. Or an answer to solving a problem of system lag. Not a debate over it's use or not. Like we said, the debate is over.

The only question we can't always answer 100% is how to best configure your system when mixing CPU and GPU slots. Depends too much on the donor's use and the hardware types, and what the latest driver happens to want. We have good recommendations, as always, but YMMV.