PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise

Advanced search

Message boards : Number crunching : Hyper threading and LLR

Author Message
mackerelProject donor
Volunteer tester
Avatar
Send message
Joined: 2 Oct 08
Posts: 2396
ID: 29980
Credit: 411,470,142
RAC: 23,105
Discovered 2 mega primesEliminated 1 conjecture "k"Found 3 primes in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 5 primes in the 2019 Tour de PrimesFound 6 primes in the 2020 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (8,774,878)Cullen LLR Turquoise: Earned 5,000,000 credits (5,149,818)ESP LLR Turquoise: Earned 5,000,000 credits (6,454,573)Generalized Cullen/Woodall LLR Turquoise: Earned 5,000,000 credits (5,122,074)PPS LLR Emerald: Earned 50,000,000 credits (72,690,404)PSP LLR Jade: Earned 10,000,000 credits (15,223,714)SoB LLR Jade: Earned 10,000,000 credits (17,319,914)SR5 LLR Sapphire: Earned 20,000,000 credits (23,993,013)SGS LLR Turquoise: Earned 5,000,000 credits (7,342,780)TPS LLR (retired) Bronze: Earned 10,000 credits (34,130)TRP LLR Jade: Earned 10,000,000 credits (18,602,519)Woodall LLR Turquoise: Earned 5,000,000 credits (5,715,464)321 Sieve Jade: Earned 10,000,000 credits (18,014,408)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,383,853)Generalized Cullen/Woodall Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,626,419)PPS Sieve Emerald: Earned 50,000,000 credits (76,969,144)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,293,882)TRP Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,012,757)AP 26/27 Sapphire: Earned 20,000,000 credits (21,918,894)GFN Emerald: Earned 50,000,000 credits (71,887,807)PSA Ruby: Earned 2,000,000 credits (2,939,755)
Message 88645 - Posted: 2 Oct 2015 | 19:58:54 UTC

Note this is about running 1 task per core with and without HT. This is not about running one task per thread with HT.

This started off as an investigation into why a new Linux build seemed to be doing units faster than a near identical Windows system. I did some testing, including turning HT off on a box so that is the only difference I'm comparing. This confirms the performance difference I'm seeing is due to HT, not Linux vs. Windows. I guess I still have linux to thank, as without its funny scheduling I wouldn't have had to turn off HT to see this in practice. For whatever reason, it appears my previous understanding that HT didn't have any significant impact on LLR (when run 1 task per real core) doesn't apply now.

This is very clear in the case of SGS. I have fewer tasks done on TRP overall, but comparing the i3 systems the difference between SGS and TRP are very close, suggesting this does scale to bigger units also.

So good news for me is, I don't have to run around installing linux everywhere this weekend :) I just have to turn off HT on those systems which still have it on, at least while I'm concentrating on LLR.

Test details:
System 1a: Haswell i3-4150T, HT on, Windows
System 1b: Haswell i3-4150T, HT off, Windows (same physical system as above, differing only on HT setting)
System 2: Haswell i3-4150T, HT off, Linux (same mobo and CPU as system 1, but differnt GPU and OS)
System 3: Haswell i5-4570S, HT off, Windows
System 4: Skylake i7-6700K, HT on, Windows

Current SGS with 120k FFTs uses under 1MB L3 cache per unit, so not expected to be impacted by ram bandwidth and makes a good comparison of pure CPU performance. As the CPUs tested vary in clock, I've normalised the following results as if the CPUs were 3.0 GHz.

Systems 1a and 4, both with HT on, running as many tasks as physical cores, had SGS unit times of 884 and 820s respectively.
Systems 1b, 2, 3 either have HT disabled or not present in the first place. They averaged 735, 747, 737s a unit respectively.

This is a good indication that without HT the performance between them is pretty consistent. With HT on, the unit lengths increase 10 to 20%!
Caution: system 4 is my main desktop and was doing low stress background tasks at the time. I'll keep observing when not in use to refine the times accordingly. This Skylake system may behave differently than Haswell systems.

I have TRP units done on systems 1a and 2 also. System 1a took about 19% longer, which is close to matching the 18% longer in SGS between the two.

Profile Michael GoetzProject donor
Volunteer moderator
Project administrator
Avatar
Send message
Joined: 21 Jan 10
Posts: 13299
ID: 53948
Credit: 224,103,372
RAC: 33,158
The "Shut up already!" badge:  This loud mouth has mansplained on the forums over 10 thousand times!  Sheesh!!!Discovered the World's First GFN-19 prime!!!Discovered 1 mega primeFound 1 prime in the 2018 Tour de PrimesFound 1 prime in the 2019 Tour de PrimesFound 1 prime in the 2020 Tour de Primes321 LLR Ruby: Earned 2,000,000 credits (2,822,730)Cullen LLR Ruby: Earned 2,000,000 credits (2,005,249)ESP LLR Turquoise: Earned 5,000,000 credits (5,009,577)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,145,754)PPS LLR Jade: Earned 10,000,000 credits (13,316,582)PSP LLR Turquoise: Earned 5,000,000 credits (5,197,957)SoB LLR Sapphire: Earned 20,000,000 credits (34,291,181)SR5 LLR Jade: Earned 10,000,000 credits (10,007,110)SGS LLR Ruby: Earned 2,000,000 credits (2,276,330)TRP LLR Ruby: Earned 2,000,000 credits (4,414,012)Woodall LLR Ruby: Earned 2,000,000 credits (2,195,123)321 Sieve Jade: Earned 10,000,000 credits (10,061,196)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,170,256)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,059,304)PPS Sieve Sapphire: Earned 20,000,000 credits (22,885,121)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,035,522)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,051,121)AP 26/27 Jade: Earned 10,000,000 credits (10,118,303)GFN Emerald: Earned 50,000,000 credits (72,598,754)PSA Jade: Earned 10,000,000 credits (12,445,029)
Message 88648 - Posted: 2 Oct 2015 | 20:40:05 UTC - in response to Message 88645.

Usually, I advise NOT using real BOINC tasks for benchmarks because task lengths vary. With TRP, for example, there's at least a 33% difference in task lengths of tests with similar n values because of the difference in k values, which vary between 4 and 6 digits long. So you definitely shouldn't ever use TRP tasks as benchmarks because there's a very significant difference in the length of individual tasks.

Don't use any conjecture tasks as benchmarks. Don't use PPS either (or PPS-MEGA). PPSE isn't so bad, because k only varies from 1200 to 9998. It will still mess up benchmarks somewhat, but not nearly as much as most of the other LLR projects.

321, Woodall and Cullen probably are somewhat more consistent in run times, at least compared to most of the others. They'll vary in length, but they won't jump all over the place like the conjectures and PPS. Genefer will behave similarly.

SGS is probably the only LLR or Genefer sub-project where you can safely use live BOINC tests for benchmarks. SGS tasks vary only a tiny amount (virtually unmeasurable), so your benchmark results will be valid. Anything else and the results are unreliable because no two tasks have the same length.

If you want to benchmark using tasks that are longer than SGS, you really should run static tests from the command line rather than using live BOINC tasks.
____________
My lucky number is 75898524288+1

mackerelProject donor
Volunteer tester
Avatar
Send message
Joined: 2 Oct 08
Posts: 2396
ID: 29980
Credit: 411,470,142
RAC: 23,105
Discovered 2 mega primesEliminated 1 conjecture "k"Found 3 primes in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 5 primes in the 2019 Tour de PrimesFound 6 primes in the 2020 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (8,774,878)Cullen LLR Turquoise: Earned 5,000,000 credits (5,149,818)ESP LLR Turquoise: Earned 5,000,000 credits (6,454,573)Generalized Cullen/Woodall LLR Turquoise: Earned 5,000,000 credits (5,122,074)PPS LLR Emerald: Earned 50,000,000 credits (72,690,404)PSP LLR Jade: Earned 10,000,000 credits (15,223,714)SoB LLR Jade: Earned 10,000,000 credits (17,319,914)SR5 LLR Sapphire: Earned 20,000,000 credits (23,993,013)SGS LLR Turquoise: Earned 5,000,000 credits (7,342,780)TPS LLR (retired) Bronze: Earned 10,000 credits (34,130)TRP LLR Jade: Earned 10,000,000 credits (18,602,519)Woodall LLR Turquoise: Earned 5,000,000 credits (5,715,464)321 Sieve Jade: Earned 10,000,000 credits (18,014,408)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,383,853)Generalized Cullen/Woodall Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,626,419)PPS Sieve Emerald: Earned 50,000,000 credits (76,969,144)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,293,882)TRP Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,012,757)AP 26/27 Sapphire: Earned 20,000,000 credits (21,918,894)GFN Emerald: Earned 50,000,000 credits (71,887,807)PSA Ruby: Earned 2,000,000 credits (2,939,755)
Message 88650 - Posted: 2 Oct 2015 | 20:49:45 UTC - in response to Message 88648.

I chose SGS mainly as I observed PPSE varying quiet a bit. I assumed this was down to the two FFT sizes, but didn't narrow it down further than that. SGS times were observed to be relatively consistent.

On TRP, I find providing you compare the same FFT size units, the times are in a similar ball park. Looking at the handful of units I've done, they're only varying up to 2%. Even if there is variation, as long as that variation is small compared to the effect being observed, I think it is a good enough indicator.

stream
Volunteer moderator
Project administrator
Volunteer developer
Volunteer tester
Send message
Joined: 1 Mar 14
Posts: 766
ID: 301928
Credit: 481,118,463
RAC: 76,661
Discovered 1 mega primeFound 1 prime in the 2018 Tour de PrimesFound 1 prime in the 2019 Tour de PrimesFound 1 prime in the 2020 Tour de Primes321 LLR Jade: Earned 10,000,000 credits (10,011,570)Cullen LLR Jade: Earned 10,000,000 credits (10,009,374)ESP LLR Jade: Earned 10,000,000 credits (10,009,221)Generalized Cullen/Woodall LLR Jade: Earned 10,000,000 credits (10,012,217)PPS LLR Jade: Earned 10,000,000 credits (11,055,307)PSP LLR Jade: Earned 10,000,000 credits (10,044,081)SoB LLR Turquoise: Earned 5,000,000 credits (8,798,554)SR5 LLR Turquoise: Earned 5,000,000 credits (9,904,596)SGS LLR Turquoise: Earned 5,000,000 credits (6,530,572)TRP LLR Turquoise: Earned 5,000,000 credits (9,911,706)Woodall LLR Turquoise: Earned 5,000,000 credits (9,963,318)321 Sieve Sapphire: Earned 20,000,000 credits (20,004,228)Generalized Cullen/Woodall Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,047,667)PPS Sieve Sapphire: Earned 20,000,000 credits (20,866,490)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,043,271)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,015,177)AP 26/27 Sapphire: Earned 20,000,000 credits (20,045,194)GFN Emerald: Earned 50,000,000 credits (53,544,479)PSA Double Silver: Earned 200,000,000 credits (200,301,443)
Message 88656 - Posted: 2 Oct 2015 | 21:28:21 UTC

You should have Windows Boinc process pinned to two correct physical cores ("Set affinity" option in task manager), and check that LLR processed will inherit this setting, before doing these tests. Otherwise, you'll end up with same scheduler problem when two LLR tasks will be _sometimes_ run on same physical core, causing performance loss.

On Linux, I've never noticed any significant performance losses when HT is turned on. Correct affinity settings (one core per cruncher) using a script which I've posted in another thread is the key.

mackerelProject donor
Volunteer tester
Avatar
Send message
Joined: 2 Oct 08
Posts: 2396
ID: 29980
Credit: 411,470,142
RAC: 23,105
Discovered 2 mega primesEliminated 1 conjecture "k"Found 3 primes in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 5 primes in the 2019 Tour de PrimesFound 6 primes in the 2020 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (8,774,878)Cullen LLR Turquoise: Earned 5,000,000 credits (5,149,818)ESP LLR Turquoise: Earned 5,000,000 credits (6,454,573)Generalized Cullen/Woodall LLR Turquoise: Earned 5,000,000 credits (5,122,074)PPS LLR Emerald: Earned 50,000,000 credits (72,690,404)PSP LLR Jade: Earned 10,000,000 credits (15,223,714)SoB LLR Jade: Earned 10,000,000 credits (17,319,914)SR5 LLR Sapphire: Earned 20,000,000 credits (23,993,013)SGS LLR Turquoise: Earned 5,000,000 credits (7,342,780)TPS LLR (retired) Bronze: Earned 10,000 credits (34,130)TRP LLR Jade: Earned 10,000,000 credits (18,602,519)Woodall LLR Turquoise: Earned 5,000,000 credits (5,715,464)321 Sieve Jade: Earned 10,000,000 credits (18,014,408)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,383,853)Generalized Cullen/Woodall Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,626,419)PPS Sieve Emerald: Earned 50,000,000 credits (76,969,144)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,293,882)TRP Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,012,757)AP 26/27 Sapphire: Earned 20,000,000 credits (21,918,894)GFN Emerald: Earned 50,000,000 credits (71,887,807)PSA Ruby: Earned 2,000,000 credits (2,939,755)
Message 88664 - Posted: 3 Oct 2015 | 11:07:28 UTC - in response to Message 88656.

Ok, on system 1a, I just put some units through after setting the affinity of boinc client to logical processors 0 and 2. Unit times are now practically identical to HT off case.

I haven't tested it yet, but I presume the affinity setting will be lost on reboot?

I guess this opens a new question: which is better? HT off, or HT on with affinity? With HT off, the CPU time wont change but elapsed time will increase according to other CPU usage. With HT and affinity set, CPU time may increase with other activity, but elapsed time might drop slightly if there is any time saving through HT. So I think the latter is better overall, but any gains will be tiny anyway so probably not worth worrying about.

stream
Volunteer moderator
Project administrator
Volunteer developer
Volunteer tester
Send message
Joined: 1 Mar 14
Posts: 766
ID: 301928
Credit: 481,118,463
RAC: 76,661
Discovered 1 mega primeFound 1 prime in the 2018 Tour de PrimesFound 1 prime in the 2019 Tour de PrimesFound 1 prime in the 2020 Tour de Primes321 LLR Jade: Earned 10,000,000 credits (10,011,570)Cullen LLR Jade: Earned 10,000,000 credits (10,009,374)ESP LLR Jade: Earned 10,000,000 credits (10,009,221)Generalized Cullen/Woodall LLR Jade: Earned 10,000,000 credits (10,012,217)PPS LLR Jade: Earned 10,000,000 credits (11,055,307)PSP LLR Jade: Earned 10,000,000 credits (10,044,081)SoB LLR Turquoise: Earned 5,000,000 credits (8,798,554)SR5 LLR Turquoise: Earned 5,000,000 credits (9,904,596)SGS LLR Turquoise: Earned 5,000,000 credits (6,530,572)TRP LLR Turquoise: Earned 5,000,000 credits (9,911,706)Woodall LLR Turquoise: Earned 5,000,000 credits (9,963,318)321 Sieve Sapphire: Earned 20,000,000 credits (20,004,228)Generalized Cullen/Woodall Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,047,667)PPS Sieve Sapphire: Earned 20,000,000 credits (20,866,490)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,043,271)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,015,177)AP 26/27 Sapphire: Earned 20,000,000 credits (20,045,194)GFN Emerald: Earned 50,000,000 credits (53,544,479)PSA Double Silver: Earned 200,000,000 credits (200,301,443)
Message 88666 - Posted: 3 Oct 2015 | 12:14:26 UTC - in response to Message 88664.

I haven't tested it yet, but I presume the affinity setting will be lost on reboot?

Of course. You must start process with necessary affinity. When starting from command line, it's simple: type "START /?" and you'll find a option to set affinity. For services, I don't know.

I guess this opens a new question: which is better? HT off, or HT on with affinity? With HT off, the CPU time wont change but elapsed time will increase according to other CPU usage. With HT and affinity set, CPU time may increase with other activity, but elapsed time might drop slightly if there is any time saving through HT. So I think the latter is better overall, but any gains will be tiny anyway so probably not worth worrying about.

I came to exactly the same thoughts. I have one Linux server running in HT mode with Boinc process affinity pinned to correct cores, so some unavoidable server activity have chance to be hyperthreaded, but it's almost impossible to say how big this theoretical gain is.

From my experience, you really must disable HT in case of CPU overclocked "to the edge". With HT, more circuits seems to be active on die, so temperatures are rising. If you'll occasionally get full load on all cores (esp. on Windows which runs tons of trash when computer is idle), it raises chances of bad calculation, system crash, or overheating.

Profile SteveRCProject donor
Avatar
Send message
Joined: 22 Mar 10
Posts: 146
ID: 57364
Credit: 453,715,676
RAC: 9,422
Discovered 1 mega primeEliminated 2 conjecture "k"s321 LLR Turquoise: Earned 5,000,000 credits (5,306,247)Cullen LLR Turquoise: Earned 5,000,000 credits (5,756,791)ESP LLR Turquoise: Earned 5,000,000 credits (5,917,101)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (4,688,089)PPS LLR Jade: Earned 10,000,000 credits (16,359,448)PSP LLR Turquoise: Earned 5,000,000 credits (6,488,430)SoB LLR Jade: Earned 10,000,000 credits (12,275,142)SR5 LLR Jade: Earned 10,000,000 credits (11,517,371)SGS LLR Jade: Earned 10,000,000 credits (13,536,986)TRP LLR Jade: Earned 10,000,000 credits (10,407,980)Woodall LLR Turquoise: Earned 5,000,000 credits (6,452,373)321 Sieve Silver: Earned 100,000 credits (246,534)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (6,474,263)Generalized Cullen/Woodall Sieve (suspended) Sapphire: Earned 20,000,000 credits (25,971,807)PPS Sieve Double Bronze: Earned 100,000,000 credits (120,681,691)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,557,924)TRP Sieve (suspended) Jade: Earned 10,000,000 credits (10,896,883)AP 26/27 Jade: Earned 10,000,000 credits (13,596,609)GFN Double Bronze: Earned 100,000,000 credits (160,216,125)PSA Turquoise: Earned 5,000,000 credits (6,367,890)
Message 88679 - Posted: 4 Oct 2015 | 9:05:45 UTC - in response to Message 88666.

... With HT, more circuits seems to be active on die, so temperatures are rising. If you'll occasionally get full load on all cores (esp. on Windows which runs tons of trash when computer is idle), it raises chances of bad calculation, system crash, or overheating.


While this may be true, my gut-feel is that running with HT on, and 'locking' LLR tasks to correct cores with high-priority, will ALWAYS show an improvement (however slight) over simply turning HT off and filling all cores with tasks. Mainly because those 'tons of trash' will force continual swapping of the LLR tasks in and out of the cores (with consequent loss of cache 'continuity'), which have to be shared with HT off, but there will be virtually no swapping of 'locked' LLR tasks with 'free' (HT) cores for all the trash (or sieve tasks!) to play around in.

Clearly there will be a rise in temperatures, but that's merely an indication that the cores are doing more work in less time, and therefore being more efficient!

____________

Profile Gary Craig
Volunteer tester
Avatar
Send message
Joined: 30 Dec 09
Posts: 3213
ID: 52890
Credit: 1,005,618,748
RAC: 7
Discovered 1 mega prime321 LLR Ruby: Earned 2,000,000 credits (2,893,273)Cullen LLR Ruby: Earned 2,000,000 credits (2,440,687)ESP LLR Turquoise: Earned 5,000,000 credits (5,738,876)Generalized Cullen/Woodall LLR Turquoise: Earned 5,000,000 credits (6,292,626)PPS LLR Turquoise: Earned 5,000,000 credits (9,648,951)PSP LLR Turquoise: Earned 5,000,000 credits (5,653,927)SoB LLR Jade: Earned 10,000,000 credits (10,558,341)SR5 LLR Turquoise: Earned 5,000,000 credits (5,748,705)SGS LLR Ruby: Earned 2,000,000 credits (3,335,713)TRP LLR Jade: Earned 10,000,000 credits (12,602,818)Woodall LLR Ruby: Earned 2,000,000 credits (2,282,622)321 Sieve Gold: Earned 500,000 credits (740,566)Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (59,788,598)Generalized Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (2,143,068)PPS Sieve Double Gold: Earned 500,000,000 credits (524,673,938)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,130,821)TRP Sieve (suspended) Jade: Earned 10,000,000 credits (10,074,710)AP 26/27 Sapphire: Earned 20,000,000 credits (43,842,888)GFN Double Silver: Earned 200,000,000 credits (224,648,943)PSA Emerald: Earned 50,000,000 credits (62,378,755)
Message 88684 - Posted: 4 Oct 2015 | 17:41:22 UTC
Last modified: 4 Oct 2015 | 17:50:57 UTC

With some (even relatively recent) versions of Linux, leaving HT "on" for LLR is a bad idea. The scheduler is not good at spreading around work among the physical cores even when running a "half load" (e.g. 4 tasks on a 4-core, 8-thread i7). The tool "turbostat" bears this out... sometimes I'd get all 4 running on two physical cores. I'd say it is always better to turn off HT in BIOS if you can, for LLR or GFN (leave it on for sieving).

Keep in mind this advice is specific to Linux, and I'm not perfectly up-to-date, but reasonably so (ubuntu 14.04)

--Gary

rjs5
Send message
Joined: 20 Feb 11
Posts: 34
ID: 87238
Credit: 638,334,010
RAC: 286,715
Discovered 3 mega primes321 LLR Jade: Earned 10,000,000 credits (10,244,140)Cullen LLR Jade: Earned 10,000,000 credits (10,235,118)ESP LLR Jade: Earned 10,000,000 credits (10,191,519)Generalized Cullen/Woodall LLR Jade: Earned 10,000,000 credits (10,253,932)PPS LLR Jade: Earned 10,000,000 credits (10,025,002)PSP LLR Jade: Earned 10,000,000 credits (10,188,140)SoB LLR Jade: Earned 10,000,000 credits (10,297,183)SR5 LLR Jade: Earned 10,000,000 credits (10,048,145)SGS LLR Jade: Earned 10,000,000 credits (10,012,220)TRP LLR Jade: Earned 10,000,000 credits (10,179,749)Woodall LLR Jade: Earned 10,000,000 credits (10,178,319)321 Sieve Jade: Earned 10,000,000 credits (10,090,924)Cullen/Woodall Sieve (suspended) Gold: Earned 500,000 credits (879,677)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,273,914)PPS Sieve Emerald: Earned 50,000,000 credits (60,310,763)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,437,484)TRP Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,509,929)AP 26/27 Sapphire: Earned 20,000,000 credits (23,000,627)GFN Double Silver: Earned 200,000,000 credits (380,446,709)PSA Sapphire: Earned 20,000,000 credits (35,533,950)
Message 88695 - Posted: 4 Oct 2015 | 20:20:53 UTC - in response to Message 88679.

... With HT, more circuits seems to be active on die, so temperatures are rising. If you'll occasionally get full load on all cores (esp. on Windows which runs tons of trash when computer is idle), it raises chances of bad calculation, system crash, or overheating.


While this may be true, my gut-feel is that running with HT on, and 'locking' LLR tasks to correct cores with high-priority, will ALWAYS show an improvement (however slight) over simply turning HT off and filling all cores with tasks. Mainly because those 'tons of trash' will force continual swapping of the LLR tasks in and out of the cores (with consequent loss of cache 'continuity'), which have to be shared with HT off, but there will be virtually no swapping of 'locked' LLR tasks with 'free' (HT) cores for all the trash (or sieve tasks!) to play around in.

Clearly there will be a rise in temperatures, but that's merely an indication that the cores are doing more work in less time, and therefore being more efficient!



It seems like there are some workload/OS combinations that are not well behaved with HT.

A Study of Hyper-Threading in High-Performance Computing Clusters

I suspect that it might have something to do with software prefetching in the application that chokes the memory bus, but that is my pure speculation. GWP used by the LLR programs had substantial software prefetching the last time I looked. Determining a correlation is a lot easier than determining the "cause".


Yegor001
Send message
Joined: 17 Sep 11
Posts: 21
ID: 112215
Credit: 426,461,391
RAC: 180,528
321 LLR Turquoise: Earned 5,000,000 credits (5,062,970)Cullen LLR Turquoise: Earned 5,000,000 credits (5,473,149)ESP LLR Turquoise: Earned 5,000,000 credits (7,222,843)Generalized Cullen/Woodall LLR Jade: Earned 10,000,000 credits (12,077,058)PPS LLR Turquoise: Earned 5,000,000 credits (5,346,865)PSP LLR Sapphire: Earned 20,000,000 credits (30,214,038)SoB LLR Double Bronze: Earned 100,000,000 credits (173,905,777)SR5 LLR Turquoise: Earned 5,000,000 credits (8,431,433)SGS LLR Sapphire: Earned 20,000,000 credits (36,411,512)TRP LLR Turquoise: Earned 5,000,000 credits (5,803,910)Woodall LLR Turquoise: Earned 5,000,000 credits (5,483,649)321 Sieve Turquoise: Earned 5,000,000 credits (5,713,403)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (6,174,701)Generalized Cullen/Woodall Sieve (suspended) Jade: Earned 10,000,000 credits (13,062,265)PPS Sieve Jade: Earned 10,000,000 credits (16,683,745)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,734,831)TRP Sieve (suspended) Jade: Earned 10,000,000 credits (18,451,367)AP 26/27 Gold: Earned 500,000 credits (792,428)GFN Sapphire: Earned 20,000,000 credits (49,403,039)PSA Bronze: Earned 10,000 credits (12,770)
Message 88937 - Posted: 11 Oct 2015 | 10:51:25 UTC - in response to Message 88666.

Of course. You must start process with necessary affinity. When starting from command line, it's simple: type "START /?" and you'll find a option to set affinity. For services, I don't know.


For services in Windows you may use Process lasso. I use it for PG crunching, and I'm satisfied.

Message boards : Number crunching : Hyper threading and LLR

[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2020 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 3.98, 2.90, 2.49
Generated 19 Sep 2020 | 13:05:44 UTC