Info 11th BOINC Pentathlon 2020

[Markfw]

OK, so the target is a mixture of

5x 16 GB DDR4-2666 and​

2x 32 GB DDR4-2993.​

Meaning the issue could be due to the mix of speeds, due to the mix of sizes, or due to both of these coinciding.

How about going into the BIOS while only workable RAM is installed, fixing the RAM speed from auto to 2666 (or even 2400) manually there, and then to try and install the additional RAM?

I just shipped it back. There is little to nothing to adjust in the bios that I can see. I will have to live with 110 gig.

 

[mopardude87]

I just shipped it back. There is little to nothing to adjust in the bios that I can see. I will have to live with 110 gig.

Oh man sorry to hear, so you and me are sorta in the same boat so to speak. Hurting on memory I can only hope my stimulus comes in so i could get mine faster then waiting till i get paid again.

 

[Endgame124]

Yes, linux mint 19.2

Mint is based off Ubuntu if I remember correctly. I’ve setup zram in fedora and red hat, and I’ve set to raspbian, but no Ubuntu yet. The process was very different between RHEL and raspbian, so I imagine Ubuntu would be a similar set of differences. If you’re interested, I could look into it, especially if setting up an AMD APU on mint isn’t an exercise in frustration.

 

[Markfw]

Mint is based off Ubuntu if I remember correctly. I’ve setup zram in fedora and red hat, and I’ve set to raspbian, but no Ubuntu yet. The process was very different between RHEL and raspbian, so I imagine Ubuntu would be a similar set of differences. If you’re interested, I could look into it, especially if setting up an AMD APU on mint isn’t an exercise in frustration.

In the past few days, I have only had one task of 113 that was waiting on memory., and it cleared in a couple of hours. I am OK, thanks.

 

[biodoc]

Won't lie when i have been so tempted to get a 3950x already to replace this thing then get 64gb of ram. I am a fan of air cooling and i heard the 3950x essentially demands water but idk if this is true?

Both my 3950X and 3900X are on air (Noctua NH-U14S) and both run cool on full load with Rosetta tasks IF I set the PPT at 105 W in the bios (Default is 142 W as @StefanR5R mentioned). Both are @ 3.75-3.85 GHz on Rosetta. The 3950X is in a closed case (side cover on) with a Radeon VII running Milky Way and the CPU temp is ~68C.

 

[mopardude87]

Both my 3950X and 3900X are on air (Noctua NH-U14S) and both run cool on full load with Rosetta tasks IF I set the PPT at 105 W in the bios (Default is 142 W as @StefanR5R mentioned). Both are @ 3.75-3.85 GHz on Rosetta. The 3950X is in a closed case (side cover on) with a Radeon VII running Milky Way and the CPU temp is ~68C.

That is pretty damn good for 32 threads. Fun note i am clocking about the same as you currently in eco mode on the 3900x but cause of lack of memory i can only work up 75% of the cpu and at a maximum of like 90% usage i am hitting 61cel. I may go into bios and play with those PPT settings but when i looked at them i saw settings below it for amp and stuff and was not sure where to set that from 0 so i went back to Eco.

I rarely tinker with voltage on stuff, i will downvolt as needed but that is about it. Most cpus minus a Q6600 never got overclocked.

 

[biodoc]

I may go into bios and play with those PPT settings but when i looked at them i saw settings below it for amp and stuff and was not sure where to set that from 0 so i went back to Eco.

On my gigabyte x570 boards, I go to 'settings', then 'advanced CPU settings' then enable PBO, set TDP to manual and enter 105 and PPT to manual and enter 105. There is another place in the bios where you can set other parameters but I haven't experimented with them yet. If you are running windows, you can experiment with these settings on the fly with the Ryzen Master software.

 

[mopardude87]

On my gigabyte x570 boards, I go to 'settings', then 'advanced CPU settings' then enable PBO, set TDP to manual and enter 105 and PPT to manual and enter 105. There is another place in the bios where you can set other parameters but I haven't experimented with them yet. If you are running windows, you can experiment with these settings on the fly with the Ryzen Master software.

Every time i get a bit comfortable i suddenly feel guilty i am dampering the things performance in Eco as honestly such a beast in my opinion needs to be fully opened. Shame i am waiting on the cooler cause i would be enabling PBO. I went back to default and its 81cel odd load temps with its cpu usage. Since i can't push this cpu any more till i get ram default works best.

Would have been like leaving the stock 2bbl intake manifold and carb on my old 1965 chrysler newport 383 big block, you simply slap on a 4bbl edelbrock set. That is how you open up a 383, that and well some open headers for starts. Sadly i sold that thing before i opened it up fully cause well a big C body Mopar with manual all drum brakes was quite scary in the rain and i was not in the mood for a full on disc conversion kit. Not for that car anyways, i let the next owner know they should consider such a thing and it was sold.

Sorry for the typical car analogy but you prob see what i mean by now

 

[mopardude87]

Edit: nvm

 

[mopardude87]

Decided to pause folding momentarily to see how BF1 handles on this 3900x with a 40% load in BOINC. I haven't played this game in a while and well its maybe the last good BF. I would do 50% BOINC but between that things ram usage and BF4 i was maxing out and it caused minor hitching as things loaded into the pagefile but it was still damn playable at 130 min and 180+ fps avg on the 1080ti.

I need more ram before i could fully play with this as intended Added pictures first of offline then did some online like Amiens 64p This is at 1080p.

Offline, the lowest fps i seen was like 156fps i think?

https://imgur.com/a/WWDJOSI

Amiens 64p Lowest i saw was 89fps and it was more or less constantly hanging over 100fps. Yeah CRONOS can take the beating! Yeah see about that gpu bottleneck? Yeah that is why i want a 2060 as a gamer while the 1080ti crunches. Would make better use of the 2060 then the 1080ti. My most demanding game till i upgrade to Ampere, i got backlog of titles i intend on getting with it.

https://imgur.com/a/vE0fUa5

Thought i had filters fixed, i had to join a euro server before finding out my filters excluded full servers. Tons of NA just full.

 

[StefanR5R]

Every time i get a bit comfortable i suddenly feel guilty i am dampering the things performance in Eco

Since you don't have particular cooling problems, nor at some limits with respect to a monthly or yearly budget for electric power bills for example, Eco mode is of limited use to you. As for the performance drop from 142 W to 88 W, I linked examples elsewhere before, which speak for themselves:

• 3900X performance, and performance/W
• 3950X performance (in these workloads, power usage should be quite near the PPT, therefore we can estimate perf/W)

If one could take Ryzens 3700X/3800X (the 8-core Ryzens) and squeeze 4 of them into a single package, one would get a 32-core Epyc. Performance and perf/W examples for 4x45 W PPT vs. 4x39 W PPT are here: relative, absolute.

Of course, all of these tests are solely about all-core throughput, not about response.

 

[mopardude87]

Since you don't have particular cooling problems, nor at some limits with respect to a monthly or yearly budget for electric power bills for example, Eco mode is of limited use to you. As for the performance drop from 142 W to 88 W, I linked examples elsewhere before, which speak for themselves:

• 3900X performance, and performance/W

Yeah i clicked on that link, then decided to try out "Hard Limit" set 105 PPT . Its loading about 3.9ghz to 4ghz with overall 70-72% cpu usage with 68-70cel temps so far but BOINC is being odd suddenly so its not pushing the cpu like earlier either.

I am curious if something was stuck in BOINC that wasn't working right then suddenly now its pulling much more resources then earlier? At default 75% its suddenly running out of memory on some of the tasks. I had to pull it back down to 65% and overall system usage still is hitting 77% currently. Not sure why the sudden large demand for more memory then its been asking for all day. All settings exactly as before the reboot, its just suddenly demanding tons more.

Found out according to Get My Payment my stimulus should be in on the luckiest of days the 13th I need this memory BADLY.

 

[StefanR5R]

Marathon — Rosetta@home

Not sure why the sudden large demand for more memory then its been asking for all day.

Earlier in this contest, there were two types of Rosetta tasks: One with a little less than 500 MB memory requirement, the other with 1,250 MB average memory requirement (up to 1,500 MB). If a computer ended up loading only tasks of the latter kind instead of a mix of both, it's obvious what happens.

I haven't checked the size distribution more recently.

----------------

City Run — Universe@Home
In the SETI.Germany forum, somebody pointed out that networking pressure towards the Universe@Home server may increase again when the second Javelin Throw at NumberFields starts, due to those users who have Universe + Numberfields combined by the basic bunkering method.

My impression at this point is that U@H's network connection will remain overused until after the end of the contest.

 

[StefanR5R]

Cross Country — Amicable Numbers
The next contest was announced just now.
Project URL: https://sech.me/boinc/Amicable/
Start: Tuesday May 12, 00:00 UTC (Monday May 11, 20:00 EDT / 17:00 PDT)
End: Sunday May 17, 00:00 UTC (Saturday May 16, 20:00 EDT / 17:00 PDT)

I have yet to update post #1. It may take me a while to collect technical info, as I am drafted to some gardening work before noon.

Edit,

SG's tech info, original, auto-translated​

We had a contest at Amicable Numbers immediately after yesteryear's Pentathlon. Edit 2, half a year later, Amicable Numbers switched to the now current 10^21 application, which is different from what we ran in May 2019. See post #241 below.

 

[mopardude87]

Marathon — Rosetta@home

Earlier in this contest, there were two types of Rosetta tasks: One with a little less than 500 MB memory requirement, the other with 1,250 MB average memory requirement (up to 1,500 MB). If a computer ended up loading only tasks of the latter kind instead of a mix of both, it's obvious what happens.

I haven't checked the size distribution more recently.

Makes sense, i dropped it down to 55% cpu usage cause memory usage was spiking to 84% and i didn't wanna chance it running out again. Running 13 tasks with memory usage hitting 77%. If the tasks are gonna be this unpredictable i will just keep it at this setting till i get my additional memory.

Shame i couldn't choose the task i want, would be nice to work 22 of the threads but its choking on memory here lol.

 

[StefanR5R]

Cross Country — Amicable Numbers
I am quoting these informative posts from our "Recent Changes in projects" thread, no further comment really necessary:

I wish somebody had mentioned this here:

You need 8.5 GB per each running GPU task.

That's system RAM, not GPU RAM - I think.

The requirement appears to be 8 GB of GPU VRAM (message 1204) and 8+ GB of (virtual) memory (message 1212) per task.

Edit: The current requirement is 2 GB VRAM and 8 GB RAM per task.

Amicable loves Turing. Great performance on that project with those GPUs. Ask me how I know.

No joke. An amicable user reports, "I noticed that my GTX 1650 appears to be at least as productive as my 1080 Ti, if not even a tad better (both 1.1-1.2 million cr/day) "

 

[StefanR5R]

Cross Country — Amicable Numbers

SG's tech info, original, auto-translated

The info points out that the "kernel size" of the GPU application can be configured.
When I visited Amicable Numbers preferences, I found a setting of 23 for NVidia GPUs there. Did I change this in the past? In any case, I am getting 100 % shader usage from it on GTX 1080Ti/ Linux. Power consumption is fluctuating but in the rage of 215 W (while the board power limit is configured to its default of 250 W).

 

[Endgame124]

What’s the best way to build the bunker for amicable? I’ve never run that project. Any use to running really old AMD GPUs on that project?

 

[StefanR5R]

What’s the best way to build the bunker for amicable?

You'd need to cope with Amicable's limit on (GPU-)tasks in progress of 20 per GPU. This limit is imposed by the server-side scheduler.

Any use to running really old AMD GPUs on that project?

I don't know; run times will certainly be substantial.
Further, I don't know whether the 8 GB 2 GB VRAM + ~8 GB RAM requirement applies to the AMD GPU application too. It's an OpenCL application on NVidia, hence maybe the same requirements are applicable to the AMD GPU version. (Maybe the info can be found at the Amicable Numbers message boards.)

 

[StefanR5R]

I'll retro-edit posts #244 and #241:
I am seeing 7.9+ GB RAM usage (resident) with "Amicable Numbers up to 10^21 v3.02 (opencl_nvidia) x86_64-pc-linux-gnu", and 1.8 GB VRAM usage. This is from several tasks so far. However, whether or not computers with less than 8 GB VRAM per GPU will be receiving tasks, this I don't know.

PS,
during the first 1/2 minute or 1 minute of the task (depends on CPU speed), the GPU is idle, while the is CPU busy preparing data tables.

PS 2:
VRAM requirement is 2 GB: Amicable forum thread

Current GPU version's requirements: any AMD (HD 5xxx or newer) or NVIDIA (GTX 4xx or newer) GPU with OpenCL support and at least 2 GB of video memory

If you run an AMD card on Windows, make sure you use driver version 14.12 or newer, otherwise it may fail at startup because of OpenCL code compilation error.

 

[StefanR5R]

Jeeper's bulletin #5:
"A strange silence lies over the Pentathlon. Veils of mist rise and cover the tactics of the teams. Here and there you can hear the deep breathing in the evening before the second day with the javelin. Is this tension, anxiety or sheer horror?"

 

[StefanR5R]

I collected the Cross Country dates and technical info in post #1 now.
Updated calendar:

                             Tu We Th Fr Sa Su Mo Tu We Th Fr Sa Su Mo
                    02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18
     Rosetta@home   __ __ __ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄   14 days Marathon
    Universe@home    _ __ __ __ __ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄                         5 days City Run
 Amicable Numbers                         _ __ __ ▄▄ ▄▄ ▄▄ ▄▄ ▄▄          5 days Cross Country
           t.b.a.                                                         3 days Sprint
NumberFields@home   __ __ __ ▄▄                                           1 day  Javelin Throw
                                   __ __ __ ▄▄                            1 day
                                                                          1 day
                                                                          1 day
                                                                          1 day
                    02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18
                             Tu We Th Fr Sa Su Mo Tu We Th Fr Sa Su Mo

 

[Endgame124]

Is it better to build points for Rosetta or Amicable? My ram upgrade doesn’t arrive until Tuesday so I would have to cut back on Rosetta on my host with the 1080ti if I added Amicable.

 

[StefanR5R]

@Endgame124, that's the big conundrum. I suppose you are already running zram on it, aren't you?

----------------

pschoefer noticed:

The computing performance at Amicable Numbers has increased today already to a degree that the ETA of the current project phase, according to server_status, advanced from 2028 to late 2026. Let's see how much further we will get this down.

 

[Endgame124]

@Endgame124, that's the big conundrum. I suppose you are already running zram on it, aren't you?

----------------

pschoefer noticed:

I am running a 8GB zram swap, which varies from being half full to completely full depending on the luck of the draw on Rosetta processes. I can increase the size to 16GB, but I’ve found Rosetta only reliably gets 50% compression, so that would only open up 4GB extra.