Posts by spRocket

1) Message boards : Number crunching : x86 vs Arm Performance (Message 99878)
Posted 3 Dec 2020 by spRocket
I've certainly been impressed by the Pi 4. Seeing it get a RAC comparable to an admittedly heavily-throttled i7-640LM (ThinkPad X201 Tablet) kind of blows my mind. Still, it's nothing compared to my "old" Ryzen 7 1700. And, strictly speaking, you're going to want active cooling with a Pi 4, though a 400 has enough of a heat sink to use passive cooling.

One potential cluster idea: something that would hold multiple Pi 4 Compute Modules. It might be kind of awkward with the new module design, though, unless you're going to use a pizza box form factor.
2) Message boards : Number crunching : ARM Compute Raspberry Pi4 with Ubuntu Mate 20.04 beta1 (Message 98889)
Posted 8 Sep 2020 by spRocket
I've since switched my Pi 4 from the vanilla Debian userland back to the official Raspberry Pi version, and put it to work on WCG to compare it to the Pi 3 systems chugging away. One thing I've noticed is the CPU running at about 65-70°C instead of 50-55°, suggesting that vanilla Debian isn't taking full advantage of the Pi 4's CPU. That being said, "vcgencmd get_throttled" on the Pi 4 is showing zero after about 18 hours of crunching WCG,, so there's no throttling or near-throttling going on. Active cooling is never a bad idea on a Pi 3 or 4.

Also, before I've even turned in a full day's worth of WCG work units, it's clear that the Pi 4 is roughly twice as fast as a Pi 3 at processing an OpenPandemics WU, far beyond what the 300 MHz speed boost would suggest on its own. I guess the Cortex-A72 has much better IPC and/or a much beefier FPU than the A53.
3) Message boards : Number crunching : ARM Compute Raspberry Pi4 with Ubuntu Mate 20.04 beta1 (Message 98764)
Posted 6 Sep 2020 by spRocket
Here's a very nice little script that can set up zram for you:

Follow the instructions, and it will seamlessly configure it to start on boot.

Now that I've had this 4 GB Pi 4 running for a while, I'm now seeing its RAC hovering around 430. Not a huge amount in the grand scheme of things, but still pretty surprising for such a tiny, low-power system. I'm running right at the ragged edge of available memory - I recently saw this in the log:

Message from server: Rosetta needs 3814.70 MB RAM but only 3809.64 MB is available for use.

The box is still crunching away happily, though. Once in a while, if I have boinctui running and catch this system finishing up a work unit, I'll see one other unit go into "waiting for memory" status for a few minutes and then resume.

I've also set zram up on a Pi 3B and a 3B+ as well, for use on WCG. I have all four cores on the 3B+ running, and three on the older 3B along with a couple of software-defined radio tasks (which still leaves plenty of room for overhead). I've noticed that the SDR box can crunch work units faster: about 20 hours per work unit vs. 22 on the box with all four cores crunching. The long work units and random validation delays mean it will be a while before I get a handle on what RAC I'll really get with these two.

Also, if you run any ARM aarch64 systems on WCG, they will always get COVID work units - that seems to be the only ARM binary on WCG.
4) Message boards : Number crunching : ARM Compute Raspberry Pi4 with Ubuntu Mate 20.04 beta1 (Message 98602)
Posted 20 Aug 2020 by spRocket
I can vouch for zram. I now have four threads running on a 4 GB Pi4, under Debian 10 aarch64 (running with the Raspios kernel). Of course, I also had to tweak my CPU and memory usage settings, with 100% for the number of CPUs and CPU time, and not-in-use memory usage set to 90%.

With basic stick-on aluminum heat sinks on the CPU, memory, and USB controller, and a tiny fan running off the 3.3V supply, I'm getting about 51-54°C for my CPU temperature.

I think my next project will be to try to get this to boot via PXE from my homebrew ZFS NAS box, and if I can get that to work, get the PoE hat and power it from the spare port on my router. No SD card, no power supply, just plug it in and go.
5) Message boards : Number crunching : no new tasks? (Message 97560)
Posted 24 Jun 2020 by spRocket
As of this morning (US Central time), looks like the well has run dry again.

I'd like to figure out a work share setup that would allow WCG tasks to run to completion when the Rosetta tasks come back. I've never liked the "suspend one to do another" approach - I'd rather have what's in the queue finish up regardless of which project it belongs to.
6) Message boards : Number crunching : no new tasks? (Message 97553)
Posted 24 Jun 2020 by spRocket
I've been seeing a steady ~30000 for a while now since the tasks returned, which is typical when there isn't a task shortage.

I went and set up WCG as a backup project, so if the tasks disappear, my Rosetta boxes can smoothly pick up WCG units and crunch them. It's kind of annoying when the Rosetta work units come back, though, since BOINC will randomly suspend the WCG tasks once Rosetta ones return.
7) Message boards : Number crunching : Quite a few signal 11 errors on a Linux host - what does it mean? (Message 96139)
Posted 5 May 2020 by spRocket
Both my homebrew NAS and an old machine I unretired turned out to be bitten by this bug as well. I've put those two onto WCG, and they've been happily crunching away.
8) Message boards : Cafe Rosetta : A great stress tester :) (Message 95576)
Posted 30 Apr 2020 by spRocket
So far, so good on isolating the bad RAM. The box has been cranking away for 10 days now without a hiccup. There was one reboot for a kernel update, but that was intentional. I haven't noticed any problems with losing that last gigabyte.

Still, it's time for me to save up for a new Ryzen board - and ECC RAM. I have no regrets about picking ECC on my last Ryzen build (no reported memory errors according to edac-util), but the A10 APU in the system with the bad RAM doesn't support ECC.
9) Message boards : Cafe Rosetta : Ubuntu guide to install Rosetta? (Message 95401)
Posted 26 Apr 2020 by spRocket
Just follow the steps here:
10) Message boards : Number crunching : Tells us your thoughts on granting credit for large protein, long-running tasks (Message 95135)
Posted 22 Apr 2020 by spRocket
I think I've picked up a 4 GB work unit on one of my systems - it has 8 GB RAM, but at the moment, only a single task is running, and I haven't touched its settings. The "top" command shows a resident size of 2.874GB.
11) Message boards : Cafe Rosetta : A great stress tester :) (Message 94927)
Posted 19 Apr 2020 by spRocket
And, the hilarity continues. I un-mothballed another old box and fired up BOINC. Things were going smoothly when I went to bed last night, but when I woke up this morning I discovered that the system was going haywire, with 160+ GB (yes, gigabytes) of goodness in /var/log and the system running at a crawl. I fired up memtest86+ and, lo and behold, discovered that I had a bad stick of RAM. I told it to report the error in badram format, set GRUB_BADRAM in /etc/default/grub accordingly, ran update-grub, and rebooted. Now it has about 7 GB of RAM instead of 8, but I'm not spending money on what is now an obsolete system, unless it's to replace the motherboard and CPU along with the RAM.

We'll see if I have to drop a task or not, but I think 7 GB should be enough to run four Rosettas.
12) Message boards : Cafe Rosetta : Record performances with EPYC (Message 94851)
Posted 19 Apr 2020 by spRocket

I wish I could afford even one of those. Even a Ryzen 9 is a bit too rich for my finances, though.
13) Message boards : Cafe Rosetta : A great stress tester :) (Message 94821)
Posted 19 Apr 2020 by spRocket
Yup, years ago at my office, we got a bunch of white-box PCs that had the infamous Deer power supplies. Those didn't hold up well - I remember powering up one of the Slot 1 Pentium II boxes to hear a loud fizzing sound and see a cloud of smoke come out from the power supply. Not only did the PSU fry to a crisp, it took the motherboard and HDD with it.

I've had a Seasonic in a Mini-ITX system that I use for Asterisk, and it's been soldiering on for quite a while now. My Corsair-equipped boxes have also held up well.
14) Message boards : Cafe Rosetta : A great stress tester :) (Message 94813)
Posted 19 Apr 2020 by spRocket
A couple of days ago, while I was SSH'ing into my main desktop, I suddenly heard less noise coming from the room and my connection froze up. I walked over, clicked the power button, and... nothing.

Several weeks of running full-tilt with Rosetta@Home and GPUGRID were just too much for the cruddy "500W" power supply that I had never gotten around to replacing when I upgraded the motherboard to a Ryzen 7/1700 a few years ago. I did an autopsy on the power supply to find it chock full of Jun Fu capacitors and some the wimpiest heat sinks I've seen; I suspect that PSU would have difficulty keeping up with a 250W load. I had to run over to one of the few remaining freestanding computer shops in the area (fortunately, open for business since people need work-from-home supplies!), and wound up getting an 850W Seasonic, which ought to have enough headroom for a beefier GPU should I upgrade that as well. Bonus: my UPS reports about a 3-4 percentage point drop in electrical load with BOINC cranking away.

The moral of the story: Make sure your power supply is up to snuff!
15) Message boards : News : Rosetta's role in fighting coronavirus (Message 94529)
Posted 15 Apr 2020 by spRocket
This needs to be in the FAQ thread.

To use a Raspberry Pi, you would first need a Pi 4 with at least 2 GB of RAM. Also, Raspbian won't do, since it's a 32-bit distro and the ARM version of Rosetta is 64-bit only. You need to use an aarch64 distro such as Fedora or Arch.

On a 2 GB Pi 4, you'll only have enough RAM to run a single Rosetta task at a time, since it can use up about 1.5 GB per task.

None of my Pi systems are Pi 4, so I can't put those to work on Rosetta.
16) Message boards : Number crunching : Problems and Technical Issues with Rosetta@home (Message 93981)
Posted 9 Apr 2020 by spRocket
Seems to me, at least when you're dealing with 8+ cores and SMT, that leaving at least one thread open for overhead is a good idea, particularly if you're planning to use the machine as a normal desktop. I have a Ryzen 7 1700 cranking away (8 cores/16 threads) running Linux, and I find that it's best to use 14 cores for Rosetta, one core and the GPU for GPUGRID, and one core for overhead. Otherwise, the system bogs down noticeably.
17) Message boards : Number crunching : Newcomers don't accept work (Message 93497)
Posted 5 Apr 2020 by spRocket
The project has temporarily run out of tasks to distribute. We just need to be patient.

In the meantime, I've set those of my systems that are capable of handling Rosetta@Home to crunch on World Community Grid when no R@H work units are available. I also have a couple of old Callisto-era AMD chips that can't handle R@H that are full-time on WCG.
18) Message boards : Cafe Rosetta : I'm new here and I notice that we're almost out of WU, is that normal? (Message 92921)
Posted 1 Apr 2020 by spRocket
One thing I've done (but it's a bit wonky to set up) is to make World Community Grid my backup project by setting its resource share (which they call "project weight") to zero and then joining my Rosetta systems to WCG as well. That way, if no Rosetta units are forthcoming, it can seamlessly switch to WCG.

Some pointers:
19) Message boards : Number crunching : Problems and Technical Issues with Rosetta@home (Message 92640)
Posted 30 Mar 2020 by spRocket
Looks like I found my answer as to why one more machine I managed to add only got a single task. :)
20) Message boards : Number crunching : Problems on old AMD processors (pre-Bulldozer) (Message 92399)
Posted 27 Mar 2020 by spRocket
I think I'll just give another one of my other older machines a cleaning. I tried it earlier and I started hearing a thermal warning tone from its speaker - but on the other hand, it has 8 GB of RAM installed.

Next 20

©2023 University of Washington