Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
![]() |
World Community Grid Forums
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
No member browsing this thread |
Thread Status: Active Total posts in this thread: 31
|
![]() |
Author |
|
BobbyB
Veteran Cruncher Canada Joined: Apr 25, 2020 Post Count: 609 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() |
This is a follow up to my thread: Which cooler are you using?
https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,45409 I finished the project and fired up the 5950x with a Noctua NH-D15 (double fan) using a Gigabyte B550M-K 32GiB. Just idling I get about 24 Celsius but no work. Eventually I get about 8 WCG tasks and I see about 60 C. I can hear the fans speeding up. Wow this is going to cook when I get 32 WUs I am thinking. At some point I get 32 tasks and check the temperatures: 46 Celsius. WTF. Quoi? Those were WCG tasks. Now I am running 32 Rosetta tasks and see the same temps. I use the command sensors which is on my Linux So how does this happen? I did no tweaking in the Bios like I did with my 3900x to lower the temp. Is there something wrong here? Now I am wondering if I need the second fan. The memory is too high and the second fan has to be higher so the case does not close. The other 2 Ryzens don't have cases either so.... It's a tight fit everywhere. |
||
|
MJH333
Senior Cruncher England Joined: Apr 3, 2021 Post Count: 266 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() |
Is there something wrong here? Bobby,I think this is expected behaviour. When running single core the clock speed will ramp to its maximum. When running all core, the clock speed of all cores will reduce to keep the machine within its power limits. The CPU temperature will go up with clock speed. My 5950X is running in Eco mode with an Arctic Liquid Freezer II 280 CLC. At idle it runs at about 30C. If crunching, say, 2 or 3 work units it runs at about 60C. If crunching 28 workunits, it runs at about 42C. (This is my personal machine, so I run BOINC on a maximum of 28 threads to save some threads for my use.) Cheers, Mark |
||
|
BobbyB
Veteran Cruncher Canada Joined: Apr 25, 2020 Post Count: 609 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() |
Good to know.
In doing this, is the throughput being reduced or still at max. I'm guessing at max. The estimated time of a WU is ~7.5 hours but it is done in 3 hours. So I'll try removing the second fan. Maybe I should have gone with a liquid fan solution. Thanks |
||
|
MJH333
Senior Cruncher England Joined: Apr 3, 2021 Post Count: 266 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() |
In doing this, is the throughput being reduced or still at max. Sorry, I'm not quite sure what you're asking here.WUs complete faster when the clock speed is higher. But I think overall throughput is likely to be greater by running all-core (or nearly all-core, as in my case). It might be an interesting experiment to turn off simultaneous multi-threading and see whether overall throughput increased. I expect it would go down, but it's not something I've tried. (I'm running in Eco mode mainly to keep my electricity bills down. Electricity has been expensive here in the UK for the last couple of years!) Your NH-D15 is a great cooler. I have an NH-U12A in one of my other machines. No chance of getting filled with gunk like a liquid cooler! Cheers, Mark |
||
|
MStenholm
Advanced Cruncher Denmark Joined: Jan 7, 2010 Post Count: 97 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Reply to BobbyB posting The estimated time of a WU is ~7.5 hours but it is done in 3 hours.
----------------------------------------This is for the current Rosetta jobs at default settings wich for some reason is 3 hours, It used to be 8 hr hence the estimated time of 8 hours, not 7.5. Anyway if you were working on some or many rb units which uses a lot of memory it slows down your CPU and the temperture drops. Back to WCG - if I run X MCM1 units on my 5950X the temperature will be 1-3 degree C higher than the current, non rb jobs when running the same X amount jobs. I can see that you didn't run any rb jobs so you should see what I see - a slightly lower temperature compared to WCG MCM1. ![]() [Edit 1 times, last edit by MStenholm at Mar 30, 2024 5:10:49 PM] |
||
|
BobbyB
Veteran Cruncher Canada Joined: Apr 25, 2020 Post Count: 609 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() |
MJH333:
WUs complete faster when the clock speed is higher. But I think overall throughput is likely to be greater by running all-core Overall throughput: yes, this is what I meant.It might be an interesting experiment to turn off simultaneous multi-threading I'll pass also ![]() I've heard about electricity bills in the UK on BBC news. There was a thread here about electricity some time ago and my area of the world was amongst the lowest and it's 100% hydro. MStenholm: Yes, it does show 8 hours but sometimes 7.5 and I am running all Rosetta now but the temps were the same with all WCG (first post). Have not seen a mix yet. I assume rb units = Rosetta Some stats: k10temp-pci-00c3 Adapter: PCI adapter Tctl: +46.4°C Tccd1: +30.8°C Tccd2: +32.5°C MiB Mem : 32009.5 total, 21477.0 free, 8547.9 used, 1984.6 buff/cache MiB Swap: 4094.0 total, 4094.0 free, 0.0 used. 23026.9 avail Mem |
||
|
MStenholm
Advanced Cruncher Denmark Joined: Jan 7, 2010 Post Count: 97 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
MStenholm: Yes, it does show 8 hours but sometimes 7.5 and I am running all Rosetta now but the temps were the same with all WCG (first post). Have not seen a mix yet. I assume rb units = Rosetta Well yes and no. The normal Rosetta jobs are small, 300-400 MB but the Robetta job can be as large as 1.3 GB so if you run all 32 threads you run out of memory. They are named like this example rb_03_18_603343_597603_ab_t000__robetta _IGNORE_THE_REST_04_08_2977637_6_1 This is one that I completed and it had a Peak working set size of 654.79 MB. Rosetta have many nice features like data of the past week or so on completed job and you can see what your fellow chrunchers use and how the CPU benches. Your combination of Ubuntu 22.04 and the 5950X benches very poorly and your points are very low. You should as minimum expect 150 points for a 3 hour job even in Eco mode. Just saying.Take a look at your 3900X with Measured floating point speed 7049.91 million ops/sec which nets around 60 points/hour. That is more like it, 30 as you get on the 5950X is low. ![]() [Edit 1 times, last edit by MStenholm at Mar 30, 2024 9:07:12 PM] |
||
|
BobbyB
Veteran Cruncher Canada Joined: Apr 25, 2020 Post Count: 609 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() |
I have not done any rb units yet but some are in the queue and none are more than 115K. Will keep my eye on them.
----------------------------------------I see some database files 400-495 MB I will have to look at my account and browse their forums at bit. Later: I see what you mean about the numbers. The 3900x is getting better credits. Is it because of the 5950x vs 3900x or something I did? That unit is using the HD which was running a I3770 CPU before. I built the new unit with the same case and just booted the same drive. Something it "remembered"? Does the BIOS need some tweaking? Even later: I think I may have solved it. I got back into the BIOS and reset to optimum defaults. I remembered a question on the first boot which asked a question something to do with a change and reset. I ignored it Anyway, I reran the benchmarks and it changed FROM: 5808 floating point MIPS (Whetstone) per CPU (This is lower than my 3900x) 81490 integer MIPS (Dhrystone) per CPU TO: 8036 floating point MIPS (Whetstone) per CPU 113149 integer MIPS (Dhrystone) per CPU [Edit 3 times, last edit by BobbyB at Mar 31, 2024 3:51:55 AM] |
||
|
MStenholm
Advanced Cruncher Denmark Joined: Jan 7, 2010 Post Count: 97 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
That is more like it. Now all jobs downloaded before the new bench will still use the old number 5808 so the points most likely will still be low but give it time. The performance, the average number of decoys done per hour was low and a new bench will not change that so there might be some underlaying reason for the sub-par point/performance. My 5950X only runs 19 jobs but does around 3 times as many decoys per hour for each job, +/- a large uncertainty since the jobs differs.
----------------------------------------So your 5950X is now a dual-boot on a HD. How many jobs to your run concurrent, 32 or less. You might want to dial it down a bit when the rb-units starts. Sorry for some reason my questionmark have left my keyboard on this laptop. ![]() |
||
|
BobbyB
Veteran Cruncher Canada Joined: Apr 25, 2020 Post Count: 609 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() |
Now all jobs downloaded before the new bench will still use the old number 5808 I kind of figured that. I checked the stats on their site for jobs which ran after my reset and boot and they showed the same ratio of points to hours.I'm still processing yesterday's downloads. Rosetta is a backup to WCG at 0 resource share (weight) but I cheated. I set it to 100, suspended everything else, and did an update to let it fill my cache because I was getting no WUs from WCG, then set it back to 0 where it is now. No my 5950x is not dual boot. I rebuilt the PC with a new motherboard, CPU, memory, and fan. I just used the old HD and Ubuntu which was on it. It's 24/7 running 32 threads. I am now running some rb units and "top" shows nothing out of the ordinary. You seem to know the stats for my units. Are you telling me anyone can see other people's stats on Rosetta? Here are a few spare question marks you can use for that laptop: ???????????? ![]() |
||
|
|
![]() |