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
1) Message boards : Number crunching : Solar Eclipse Challenge (Message 119920)
Posted 1457 days ago by DwightHoward2

I have no idea how far your laptop differs from my desktop. First of all, computational performance of laptops is nowadays limited by the heat dissipation capacity of the cooling system. Second, do you have dual channel RAM in it? Many laptops are sold with single channel RAM per default. LLR benefits from RAM bandwidth. With single channel RAM, it is again possible that the optimum is 1 thread per task, HT disabled. But this is just a wild guess on my part.


Thank you all for the information... Probably my first time deep diving into this program so I'm very new into all of it.

But I actually do have adequate cooling for a laptop even at full load but the RAM bandwidth would explain why my i5-4570 is outperforming it. I'm going to play around with setting to hopefully be more ready for the next challenge
2) Message boards : Number crunching : Solar Eclipse Challenge (Message 119892)
Posted 1457 days ago by DwightHoward2
So a couple days ago I decided to switch to PPS LLR in preparation for the challenge. The throughput of PPS was rather uninspiring compared to the other projects I had been running. So I decided to look at how I was running PPS and did some quick testing on both my i5-4670 (DDR3-1600) and my i7-4790 (DDR3-1600). I started out with one task on four cores and was getting a lot of 1st's, but not a lot of total units done. So I tried two tasks on two cores each and found that while each unit was slower, overall throughput was higher. At that point I knew I had to try the traditional one task per core and see what happens. Then it only seemed prudent to run all three conditions on both CPUs to see if there was a difference.

These tests were all run using live units and the times are visually taken from the BOINC manager. So please take these results with more than a few grains of salt. haha

i7-4790 DDR3-1600 PPS LLR runtimes

Per Unit:
4c 1t = 10 minutes
2c 2t = 17 minutes
1c 4t = 37 minutes

Per two units:
4c 1t = 20 minutes
2c 2t = 17 minutes
1c 4t = 37 minutes

Per four units:
4c 1t = 40 minutes
2c 2t = 34 minutes
1c 4t = 37 minutes

i5-4670 DDR3-1600 PPS LLR runtimes

Per unit:
4c 1t = 10 minutes
2c 2t = 17 minutes
1c 4t = 40 minutes

Per two units:
4c 1t = 20 minutes
2c 2t = 17 minutes
1c 4t = 40 minutes

Per four units:
4c 1t = 40 minutes
2c 2t = 34 minutes
1c 4t = 40 minutes

PPS runs remarkably similarly on the two CPUs when running one or two tasks at a time. Bump up to four tasks at a time and the i5 starts to fall off.

If you have a Haswell CPU it looks like the best for speed is one task on four cores and if you want the best throughput it looks like the best is two tasks on two cores each.

Running four tasks at once was all around bad on both. It had the slowest speed per unit, lowest or second lowest throughput, and had the highest CPU temperature.

Different CPU families and different RAM speeds will drastically alter these results.

May whoever reads this find it informative or inspiring.
Happy prime finding!


How do you go about changing how many tasks per core or how many cores per task (however its setup)?
I'm interested because I never knew you could change those values and I have a laptop with an i7-7700hq that runs extremely slow on task when it is useing all 8 cores.
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2022 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 0.47, 0.58, 0.67
Generated 9 Aug 2022 | 8:27:31 UTC