Info 11th BOINC Pentathlon 2020

[TennesseeTony]

..... I fixed a wall outlet — which doubled as junction box — which got hot and smelly when I had high load on the respective circuit. I still have several outlets which get warm now (but not actually hot), and should look into that after the Pentathlon.

I just moved everything into a workshop in the garage...I ran a 20amp circuit at 240v to that room, which is good for 4800 watts, although I want to keep it well under 3600w if possible. After firing everything back up, I touched the plug and outlet, both were cool to the touch. I touched the 20amp rated wiring and it was just a little warm, to my surprise. I checked the meter on my UPS and it said I was pulling 4200 watts! I move a few systems to 'standard' 120v outlets (about 850 watts worth).

I still am surprised that wiring will begin to heat BEFORE they hit their rated capacity. And yes, I hung a fire extinguisher nearby.

 

[StefanR5R]

Earlier this week, when v4.15 was current, I ran Rosetta only on 4 cores, therefore I don't have much data. But I recall seeing about the same pattern of RAM requirement as I got today from my 256 tasks sample in post #58.

I see the graph, but not sure what its telling me.

The most important datum is beneath the graph: Average ~0.9 GB per task.
The second most important datum is beneath the graph too: Peak almost 1.5 GB per task.

In the graph itself, each blue bar is one of the tasks which were running at that time.* The length of each bar is the RAM footprint of the task.
The numbering at the x axis is just the numbering of tasks after I had sorted them by RAM footprint in the spreadsheet.

*) I had and have more running, but took only a random sample of 256 of them. Furthermore, before I made this graph from tasks of several computers combined, I made graphs for two big computers and their distribution looked just like this combined picture. On computers with fewer tasks running at a time, the pattern may differ more, randomly into the favorable or unfavorable direction.

 

[Markfw]

The most important datum is beneath the graph: Average ~0.9 GB per task.
The second most important datum is beneath the graph too: Peak almost 1.5 GB per task.

In the graph itself, each blue bar is one of the tasks which were running at that time.* The length of each bar is the RAM footprint of the task.
The numbering at the x axis is just the numbering of tasks after I had sorted them by RAM footprint in the spreadsheet.

*) I had and have more running, but took only a random sample of 256 of them. Furthermore, before I made this graph from tasks of several computers combined, I made graphs for two big computers and their distribution looked just like this combined picture. On computers with fewer tasks running at a time, the pattern may differ more, randomly into the favorable or unfavorable direction.

OK, so ON AVERAGE below is what each machine should take. Number of cores, and below memory required
8________12______14_____16_____24______32_____64
14.4____21.6_____25.2__28.8____43.2____57.6___115.2

Edit: it took out all my spacing......

So those with a
3700x or the like, need a MINIMUM of 16 gig.
3900x or the like, need a MINIMUM of 24 gig. (32 is best)
Xeon 14 core or the like, need a MINIMUM of 32
3950x or the like, need a MINIMUM of 32 gig
2970wx or the like, need a MINIMUM of 64 gig
2990wx or the like, need a MINIMUM of 64 gig
Those with 64 cores or more (like your dual 32 core) 128 gig on up.

Since these are minimums, we have to see the real requirement once I get all my machines running 4.2 I may know better.

My 2970wx has about half 4.2 tasks and half 4.15 tasks , 47 tasks running and 30.4 gig used of 32. I could have a problem Tuesday,

 

[Markfw]

I just ordered a 64 gig kit for my 2970wx

 

[StefanR5R]

When I had put the 2x32c computer together (first build since I don't know how long), I realized that this one had as much memory as my previous four dual-socket computers combined.

I had vague plans to improve the four 2xE5v4's, such as doubling their RAM and equalizing their processor types. But after I ran the first few power efficiency comparisons between them and the new computer, I dropped these plans.

 

[StefanR5R]

Re what changed from Rosetta v4.15 to v4.20:

This update includes:

1. extraction of the Rosetta database into the project directory with all following jobs reading from the same database rather than extracting into the slot directory for every job. This significantly reduces the disk usage per job.
2. checkpointing in the Rosetta comparative modeling protocol. This should significantly reduce wasted cpu time if jobs are preempted often, particularly for jobs that take a long time to produce models.

I deduct from that:

• Changes to RAM requirements, if there were/ are any, are related to work unit batches, not to this change of application version.
• The bug with the i686 Linux application version when running it on x86-64 Linux may still be present.

 

[Assimilator1]

Markfw
For posting tables of numbers etc you need to put in dots or underscores, that's been the bane of many stats posters for decades .

So my 3600 is now running all 4.2 tasks (10 threads), Rosetta's RAM usage is, 7 tasks averaging 420 MB each, the remaining 3 720-800 MB.......... wait a sec, one of those was ~@1.2GB, but no task has just finished so that must mean their RAM usage varies wildly! Yep, now one is back to 1.2 GB! [edit] A few minutes later another task has now gone to 1.1GB. This one particular task seems to be fluctuating from 0.7-1.3 GB.

 

[biodoc]

Changes to RAM requirements, if there were/ are any, are related to work unit batches, not to this change of application version.

I agree. The task batch with "Junior_HalfRoid_design5_COVID-19" in the name take more than 1 GB per thread on my rigs. I'm running the 64-bit linux app exclusively though if that's relevant or not.

@Markfw , I did install the extra RAM on one of my 3700X's.

Attempt #1: The 2x 8 GB CL14 DDR4 3200 was installed in DIMMA2 & DIMMB2. XMP was turned on in the bios. I added the 2 x 8 GB CL16 DDR 3200 into the DIMMA1 & DIMMB1 slots but the computer did not post. I pulled the extra RAM modules and put the computer back into service. I decided to think about it.
Attempt #2: I shut down the computer and cleared the CMOS. I put the CL16 RAM modules into DIMMA2 & DIMMB2 and the CL14 RAM into DIMMA1 & DIMMB1. It now posted into bios with RAM running at 2133. I turned on XMP profile which pulled the timings from the CL16 RAM rebooted and it posted with all RAM running at 3200 with the slower CL16 timings. It's back up and running and so far so good.

 

[Assimilator1]

Yea you need to have all RAM modules on the same timings as well as speed, set to the slowest modules you have, as you found out .

Btw, is there anyway in Win10 to link PID (process ID) to a specific Rosetta task?

PS Updated my previous post, some WUs are varying from 0.7-1.3 GB RAM usage.

 

[StefanR5R]

is there anyway in Win10 to link PID (process ID) to a specific Rosetta task?

I don't have a Win10 box here to check. If taskmanager or another process monitor can be brought to show the command line of a process, then it will have something akin to the workunit name in it.

Vice versa, you can use boincmgr's advanced view, Tasks tab, click on a running task, [Properties], and there it shows Process ID (on Linux at least).

 

[Assimilator1]

No BOINC doesn't show PID in Win 10, as for task manager, well not that I could see, but I don't know it very well.

 

[TennesseeTony]

Are you sure?

 

[Markfw]

Markfw
For posting tables of numbers etc you need to put in dots or underscore, that's been the bane of many stats posters for decades .

Fixed, thanks !

 

[Assimilator1]

Are you sure?

View attachment 20556

I was referring to TM not showing the WU name (cmdline process), if you expand out that column all it shows is 'rosetta_4.20_windows_x86_64.exe', nothing about the WU (assuming it's the same as mine).

 

[Endgame124]

My Ryzen 2700X w/ 16GB of ram is 4GB into swapp with Rosetta processes. Think I'll give zram a try over there (and also order more ram)

 

[StefanR5R]

RAM starved Linux machines could be switched to Universe@home (City Run), unless of course if you prefer to go with medical work.

 

[Endgame124]

RAM starved Linux machines could be switched to Universe@home (City Run), unless of course if you prefer to go with medical work.

My wife, who is normally pretty anti computer upgrades, has given me the green light to order upgrades if it can be used for Covid research. Oh, the things I shall upgrade

 

[Markfw]

My wife, who is normally pretty anti computer upgrades, has given me the green light to order upgrades if it can be used for Covid research. Oh, the things I shall upgrade

Nobody calls be silly or wasteful, or any of the other names I have heard for years, since I facebooked all the things I am contributing to, and that virtually all are doing covid-19 research. That $10,000 a year in hardware, and about $9,000 in electricity !

Edit: and new signature !

 

[TennesseeTony]

My wife, who is normally pretty anti computer upgrades, has given me the green light to order upgrades if it can be used for Covid research. Oh, the things I shall upgrade

Tell your wife I love her.

 

[Endgame124]

Tell your wife I love her.

While Computer hardware isn’t her thing, she is a research scientist and works with super computers daily. She knows just enough to tell me not to waste money on stuff currently available as a new generation is just around the corner. So at the moment I’m building raspberry pis, buying Water cooling components, and limiting upgrades to things like ram for the 2700X. Come on Ryzen 4000 series or Nvidia 3000 series, one of the two needs to drop ahead of schedule.

 

[Assimilator1]

Any news on what Ryzen 4000 brings over the 3000 series?

 

[Endgame124]

Any news on what Ryzen 4000 brings over the 3000 series?

I don’t have any links off hand, but the rumors were pointing to Higher clocks, a big improvement to IPC (20%+) and improvements to the infinity fabric to improve memory latency. Overall, a pretty big upgrade.

 

[Endgame124]

I setup zram on my fedora box while I wait for ram to arrive. 4.5 GB into swap and you would never know you hit swap (unlike hitting my NVME swap, which you could 100% tell you were in swap).

Of course, Fedora is a different procedure to set it up than Raspbian / Debian, so if I’m not the only one using fedora, follow this guide to setup zram:

Using ZRAM as swap on Fedora – nullr0ute's blog

nullr0ute.com

 

[Endgame124]

7 of Rosetta's best 10 days were in the last 2 weeks. I'm guessing with this contest + covid, all of the 10 best days will be either April or May 2020

 

[StefanR5R]

They have corporate donors of computer time now, although certainly not at a scale like F@h.
Dutch Power Cows (their top user is an organization with a datacenter), Rackspace, HPE, CERN stick out.