| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: Africa Rainfall Project
Thread: Work Available |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 3171
|
|
| Author |
|
|
adriverhoef
Master Cruncher The Netherlands Joined: Apr 3, 2009 Post Count: 2150 Status: Offline Project Badges: 
|
I believe your computer is fast enough. Didn't you mention that you were considering to buy a Ryzen 9 7950X? Yes I believe I have a fast machine however I don't recall considering getting the CPU you mentioned above. I was using a 7900X then I switched to Intel now have a 14900KF There was a little fairy who whispered in my ear recently, reminding me of talking about Ryzens at the time when you were considering to buy something new, Speedy, so I've been searching the forums, and found this post in a thread that spoke about a possible upgrade of your machinery.  Hope you like my search skills, Speedy. Adri PS Saw an interesting documentary about the Whanganui river (in your very beautiful country) last night on Dutch TV. It seems that it's the world's first river (as a natural resource) to be given its own legal identity, with the rights, duties and liabilities of a legal person, owing to its importance to the region's Māori people. |
||
|
|
Speedy51
Veteran Cruncher New Zealand Joined: Nov 4, 2005 Post Count: 1283 Status: Offline Project Badges: 
|
Adri, you are correct, thanks for reminding :-) Your research skills are fabulous. I would have been referring to when I planned to upgrade.
----------------------------------------Glad you liked it. Yes we do have a beautiful country  |
||
|
|
spRocket
Senior Cruncher Joined: Mar 25, 2020 Post Count: 274 Status: Offline Project Badges: 
|
Yes. It's pretty amazing that we thought we hit the power wall years ago and here we are with lower-power processors that are four times faster than 10 years ago. I'm fast becoming a fan of today's mini-PCs. You can get lots of crunching power, in a small package, with low power consumption. Downside: no dedicated GPU unless you have Thunderbolt and an external GPU. It would be interesting to see what my i9-12900HK would do if I ran just 12 threads, since it has six dual-threaded P-cores and eight single-threaded E-cores. Currently I run 18 threads, and work unit performance is consistent across projects. I'm seeing about 12-14 hours for an ARP unit in this configuration. |
||
|
|
Boca Raton Community HS
Advanced Cruncher Joined: Aug 27, 2021 Post Count: 125 Status: Offline Project Badges: 
|
Yes. It's pretty amazing that we thought we hit the power wall years ago and here we are with lower-power processors that are four times faster than 10 years ago. I'm fast becoming a fan of today's mini-PCs. You can get lots of crunching power, in a small package, with low power consumption. Downside: no dedicated GPU unless you have Thunderbolt and an external GPU. It would be interesting to see what my i9-12900HK would do if I ran just 12 threads, since it has six dual-threaded P-cores and eight single-threaded E-cores. Currently I run 18 threads, and work unit performance is consistent across projects. I'm seeing about 12-14 hours for an ARP unit in this configuration. It would be interesting to see. It can be tough to dictate that the P-cores are being used without software intervention, but you can always see what happens. Either way, you might see a boost in speed running less threads but I am not sure what the thresholds/temps are for that CPU. EDIT: you could always disable the E-cores completely via bios, but most people are not a fan of that idea. [Edit 1 times, last edit by Boca Raton Community HS at Jan 31, 2025 5:19:56 PM] |
||
|
|
spRocket
Senior Cruncher Joined: Mar 25, 2020 Post Count: 274 Status: Offline Project Badges:
|
A quick look indicates that when I set it up to run 12 tasks, six get distributed to the P-cores and six end up on the E-cores. I have yet to see how this all affects execution times (that will really take a few days), but I'm leaning towards the idea that an E-core ends up at least equivalent to a P-core's extra thread.
You can use taskset on Linux to pin processes to cores, but that is a manual process. |
||
|
|
Boca Raton Community HS
Advanced Cruncher Joined: Aug 27, 2021 Post Count: 125 Status: Offline Project Badges:
|
A quick look indicates that when I set it up to run 12 tasks, six get distributed to the P-cores and six end up on the E-cores. I have yet to see how this all affects execution times (that will really take a few days), but I'm leaning towards the idea that an E-core ends up at least equivalent to a P-core's extra thread. You can use taskset on Linux to pin processes to cores, but that is a manual process. Is your DRAM pushed to the reasonable limit? I feel (and others are welcome to disagree) that DRAM for ARP is going to play a HUGE role in the processing speed. We are running a stable system on KLEVV DRAM at 8400MT/s and it dramatically reduced the time these tasks take. Shout out to KLEVV for allowing us to try their DRAM, it's awesome. Also, to stay on topic of the thread, I feel I have seen a reduction in ARP going out today, but no shortage of MCM1. |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12338 Status: Offline Project Badges:
|
The amount and speed of RAM is a significant factor in ARP completion times, especially at checkpoints. Keeping checkpoints apart can improve completion times. I do this by suspending one for a minute or so if they get close.
Mike |
||
|
|
spRocket
Senior Cruncher Joined: Mar 25, 2020 Post Count: 274 Status: Offline Project Badges:
|
Is your DRAM pushed to the reasonable limit? I feel (and others are welcome to disagree) that DRAM for ARP is going to play a HUGE role in the processing speed. We are running a stable system on KLEVV DRAM at 8400MT/s and it dramatically reduced the time these tasks take. 8400 MT/s... eep. This little mini maxes out at 3200 (and I'm running 64 GB of 3200 MT/s Crucial). I don't like to play the overclock game. Looking further, the CPU is on a fairly short leash (45W continuous TDP, 64W burst). I'm thinking I might get my best crunch for the watt at 14 threads, but it's going to take a long while to determine that. First I want to see what 12 does. Now that a few units have been turned in, I'm seeing a slight speed boost (maybe .1 to .2 hours) on MCM. From the credits/day standpoint, it might be a wash. EDIT: I'm ending the experiment - after seeing more MCM work units, it looks like my speed boost was really an illusion. [Edit 1 times, last edit by spRocket at Jan 31, 2025 10:02:08 PM] |
||
|
|
gj82854
Advanced Cruncher Joined: Sep 26, 2022 Post Count: 99 Status: Offline Project Badges:
|
Out of 223 executing ARP1 work units across 6 machines, not a single one is an extreme and only abour 10% are accelerated. Seems like the gap is only going to grow. Someone needs to "unstick" the 104-109 generations
|
||
|
|
catchercradle
Advanced Cruncher Joined: Jan 16, 2009 Post Count: 125 Status: Offline Project Badges: 
|
Out of 223 executing ARP1 work units across 6 machines, not a single one is an extreme and only abour 10% are accelerated. Seems like the gap is only going to grow. Someone needs to "unstick" the 104-109 generations Similar here. No extremes, just two out of 12 accelerated, one of which will finish in less than half an hour. On the plus side, I can keep 12 out of 16 real cores filled and keep a queue of 4 or 5 tasks and most are finishing with run times of about 4.5 hours. A week ago, I could only keep that many cores populated by having a VM running Windows on the same box. Now all my tasks are running natively in Linux. In the VM they took between one and a half and two and a half hours longer. |
||
|
|
|