With the upcoming challenge the age-old question poses itself again: Does HT help with throughput for this one or not?

First of all, I normally avoid running subprojects with GPU apps on CPUs, but for the challenge I'm going to throw everything I have at this (it does add up, and the 1950X can contribute almost GTX 980 levels of tasks). Anyway, during last year's challenge I was running the CPU app as well and had HT enabled on every system capable of it. I don't remember explicitly, but I think I did some testing beforehand and found that it made a tiny improving difference.

Yesterday I tested it on an i7-3770 and extrapolated runtimes after the tasks had reached 1%. With only 4 tasks one came in at ~65,000 seconds, with 8 it was ~128,000. Today I clocked those same tasks again (8 running) and times are between 129,000 and 130,000. This would still suggest a miniscule benefit, but there seems to be quite some variation between units as well as speeding up and slowing down throughout single tasks. Bottomline, I'm not so sure anymore that it does help. Worst case scenario it's just twice as long and no harm done (apart from some tasks not making it before the challenge ends, in which case turn if off)?

Anyone have any findings of their own?
____________
Long live the sievers.

+ Encyclopaedia Metallum: The Metal Archives +

If anyone cares, I let some tasks run to completion and here are the results:

HT on, 8 tasks in tandem: av. 129,200 seconds (5.35 tasks per day, 37.45 per week)
HT on, 4 tasks in tandem: av. 65,000 seconds (5.31 tasks per day, 37.20 per week)

All things averaged, there seems to be about a 15-minute difference in favour of HT on/100% loaded, which considering the task lengths is quite negligible and perhaps not even significant and just a result of the small sample size. Maybe it'd be even less of a thing if I disabled HT for real (too lazy to do so, sorry). Anyway, I think I'll go for 50% of threads, I just prefer the shorter tasks.
____________
Long live the sievers.

+ Encyclopaedia Metallum: The Metal Archives +

I have 58 physical CPU cores I'll be loading up for the challenge. Unless may preliminary calculations are totally off (based on testing with 100+ threads instead, but as we've seen it should be more or less the same), I should be able to do 600-700 tasks for that particular timeframe, the 1950X and 9900k together accounting for more than half of that number. That's about a fifth of what my GPUs can do. Definitely worth it, even though the RTX 2080 alone will manage over 100 tasks more than all the CPUs. But hey, the BOINC spirit and all.
____________
Long live the sievers.

+ Encyclopaedia Metallum: The Metal Archives +

More testing done on different machines and I got totally different results...

i9-9900k
16 tasks: 21.43 t/day (64,500 secs)
8 tasks: 16.74 t/day (41,300 secs)

1950X
29: 37.85 (66,200)
16: 26.55 (52,100)

Ryzen 2600
12: 13.12 (79,000)
6: 9.95 (52,100)

i7-2600
8: 5.67 (122,000)
4: 4.96 (69,700)

I have no idea why these consistently show HT to help quite a bit while it didn't with the 3770...
____________
Long live the sievers.

+ Encyclopaedia Metallum: The Metal Archives +