Seti@Home optimized science apps and information
Optimized Seti@Home apps => Discussion Forum => Topic started by: Cosmic_Ocean on 21 Jan 2012, 06:50:51 am
-
So back in October 2006, I put together a machine that I planned to make last five years and still be competitive to new stuff that comes along. I went with a 2P Opteron setup. Until the capacitors got tired of full-bore 24/7 for 4.5 years, it was doing a fantastic job of staying competitive. I replaced the bulged caps and that fixed the problem for a few months, but then it came back. Replaced more caps and it didn't change anything at all. I kind of limped along the past three months with anticipation of building a new rig, but the funds didn't cooperate.
Finally got the funds, and went with a nice new setup:
ASRock 970 EXTREME3 (http://www.newegg.com/Product/Product.aspx?Item=N82E16813157280)
AMD FX-6100 Zambezi 3.3GHz (http://www.newegg.com/Product/Product.aspx?Item=N82E16819103962)
G.SKILL Ripjaws Series 8GB (2 x 4GB) 240-Pin DDR3 1600 (PC3 12800) (http://www.newegg.com/Product/Product.aspx?Item=N82E16820231314) x 2
Seems like a pretty nice setup. I kept a lot of my old setup, with the exception of my OS drive (was a PATA 80gb WD Caviar). Boards these days don't have PATA at all, and I've been wanting to move over to SATA for my OS drive anyway. Dug around in my pile of computers and found an 80gb SATA drive. Ran a lot of tests on it and there are zero bad sectors, so that's all I needed. My plan is to use Acronis or something similar and copy it over to this SSD (http://www.newegg.com/Product/Product.aspx?Item=N82E16820226225) that I plan on getting sometime in the near future, when funds permit.
Then I'll upgrade my 8800GT to something like an HD 6970. I'll also need to go for a new power supply as this one is 5 years old and has been at 50-75% load 24/7. It seems fine, but the potential for failure is getting pretty high. Not to mention that the mobo reports the 12v rail as 11.3, 5v as 4.35, 3.3 as 3.1. Since they're all low, it may just be the board is reporting low, or it could just be that the PSU is getting weak.
So I got it all put together pretty easily. I was ecstatic to see that when you enter BIOS to change settings.. you can use the mouse, and it is 1024x768. That was pretty cool. Disabled Cool'n'Quiet because I'm going to be crunching and need full-speed. After I got Windows installed and pulled up CPU-Z, I noticed I was only running at 1.4GHz. Pulled up something to load the CPU and it jumped to 3.3. So I guess even when you disable cool'n'quiet.. it's still enabled. Might just have to go to the overclocking section and punch in what it should be.
[rant]
So I was hoping to get BOINC installed and start attempting to build an AP-only cache, but there's just too many problems. For starters, I'm going to RMA the board first-thing Monday. During the Windows installation, I had sound coming out of the speakers. Once I was at the desktop, I started installing drivers for everything. After installing the drivers for the onboard sound, I was getting constant balloon notifications about "you have just plugged an audio device in" and "you have just unplugged an audio device." On like half-second intervals. Constantly. Pulled up the Realtek HD Audio Manager and it shows you the ports either bold or faded for occupied/vacant. Well instead of the green port changing states, all six of them were doing all kinds of random state-changes. Unplugged the speakers, and it stopped. Plugged it back in, crazy crap.
My previous board had a screwy issue with the Realtek sound. The drivers from the board manufacturer didn't work right, so I ended up using the reference set from the Realtek website and that fixed it. Figured I'd give that a try this time. Didn't make any difference at all. Uninstalled the sound card in device manager and let it go back to the default windows drivers and that still didn't fix it. I think the board is just defective or something.
In addition to that, I think one of the USB2.0 stacks on the rear panel are defective. I plugged a cord in for my camera (nothing plugged in at the other end), and there was an electrical arc/pop and two of my HDDs spun down and came back up. So that's a bad socket if I can't handle having something plugged in without arcing and shorting out. Also, I have singing capacitors. They are almost silent if I disable the onboard NIC, but the volume and note they sing changes depending on CPU and general system load. I don't know if it is related to the possibly low voltage output from the PSU, but it just seems there's too many problems and I'm not going to just "make do" with it.
I would just contact ASRock about it and try to RMA through them, buuuut.. they just started a Chinese holiday for the next 10 days, so they're entirely useless until the 30th. I'm hoping that if I can RMA it with Newegg on Monday and ship it to them, I can hopefully get the replacement Friday, but that's probably very wishful thinking.
Then during all of my installations and configurations, the OS drive would go into this read/write cycle where the entire system would become unresponsive (even the mouse) for exactly 300 seconds and then be fine again. I disabled about 10 services that are basically entirely useless and it was still doing it. Finally figured out that it was Homegroup. Left the homegroup and disabled the services for that. Problem went away. Of course it didn't help that while I was trying to set up my email accounts in Outlook, Homegroup was doing whatever-the-hell it was doing, and at some point during updates, Windows decided to change my update setting from "check for updates and notify for download and install" to "automatic download and install, with automatic restart". So I'm trying to import my Outlook archive file which was about 1.3gb, .NET 4 is doing an install in the background, and Homegroup is hosing everything up. Damn that was an annoying hour and a half.
[/rant]
So I'm finishing up with installing everything but BOINC and getting everything configured the way I want it. Since it's going to be the same model mobo and all the same hardware, I won't have to worry about starting all over again when the replacement shows up. Also, since I haven't been able to play games for the past 3 months, I'm going to install them but not play them until I get a working replacement board, otherwise it'll just make that week (or so) just pure agony.
Just felt like sharing my woes.
-
Hi cosmic.
I have the 990 FX Extreme 3 with a FX 8150.
Did you update your Bios ?
Also check your settings in Bios.
I couldn´t run it with default settings.
Have set up everything to manual since nearly no value did match requirements.
Mike
-
Board came with the newest BIOS version already. Regarding the CPU being throttled even though that feature is disabled is one that I will have to manually overclock it to the stock value. However, BIOS settings won't fix the defective USB port and sound card.
-
Regarding CPU throttling you need to disable CPU Load Line Calibration.
Mike
-
Finally just went through and got the WEI scores. Not bad, but the HDD brings the whole score down.
Processor: 7.3
RAM: 7.6
Graphics: 6.9
Gaming graphics: 6.9
primary disk: 5.4
Haven't done the suggestion for fixing the throttling yet.
-
amd throttles the CPU and has turbo when its working hard. To stop the throttling just go into the Catalyst control center. Under performance click the CPU POWER. Then simply slide the lowest setting to the far right. This forces the CPU to stay at its max speed
-
amd throttles the CPU and has turbo when its working hard. To stop the throttling just go into the Catalyst control center. Under performance click the CPU POWER. Then simply slide the lowest setting to the far right. This forces the CPU to stay at its max speed
Yeah, I don't have Catalyst. According to AMD, there is no 900-series chipset. The drivers that are on the ASRock website (and on the CD) are labeled as "AMD all in 1 driver ver: 8.70a_WHQL", and there are many references to 8xx in the filenames and folder structures within. I'm thinking the 900-series is 800-based with a few new features or something, but isn't different enough to need different drivers? I don't know.
By the way, disabling spread spectrum in UEFI setup did not force it into full speed. I did also disable turbo, because as I read for how it operates, it will turn two or three cores completely off to turbo the remaining ones.
-
My Biostar TA 990 has a v8.71. For giggles I checked the 890 mobo from biostar and its chipset was v8.70
-
Regarding CPU throttling disable “CPU Thermal Throttle” in UEFI.
Please disable Cool’n’Quiet.
Please try different settings for “CPU Load-Line Calibration” and “AMD Turbo Core Technology” in UEFI.
Settings in CCC doesn´t prevent CPU throttling.
Mike
-
It sure doesn't I hadn't noticed it because I was running Seti full time so I never dropped down
-
Settings in CCC are similar to power saving settings which only take effect if some cores are idle.
That doesn´t happen while seti is running.
CPU Thermal Throttle and CPU Load Line callibration prevent overheating the CPU by reducing core voltage.
I tested it while overclocking.
4.3 GHZ was just to much for my heatsink, so i ordered a better one.
Mike
-
So. Since I'm going to be calling in the morning and getting an RMA set up, I was thinking "what am I going to do in the meantime? using the laptop is just weird. OO! I know.. set the laptop up ontop of my case, hook the USB keyboard and mouse up, hook the monitor up, and bam.
Then I got an even better idea. I hadn't nuked my old PATA C drive, and there really wasn't anything stopping me from just putting the old setup back together. Well, except for two bulged caps. Replaced those (again) and well.. here I am, back on the old setup. I can do a lot of stuff, except for write to the array at all, or play games. I can watch TV shows, listen to music, check email and frequent some forums until the replacement board comes back.
-
Sent the board off today. I'm thinking Tuesday (31st) to have the replacement board. After reading the reviews for it (like I should have done in the first place), it seems it's a real crapshoot as to whether there is going to be a problem with it or not. And the problems range all over the map; there's no one specific common issue. I've heard (from fellow nerds) of ASRock boards being pretty durable and reliable along with being affordable, so I figured I'd give it a shot. We'll see what happens with the replacement.
I do have a question about the opt apps though. If I install BOINC but don't attach to any projects, and then create the project folder in the data directory and put my app_info and EXE in there, THEN attach to the project, will that work right away? My goal is to keep my application details from having any references to stock apps, so I don't want to attach and get assigned work right away after the stock downloads. I want the new machine's app details to look like my single core machine (http://setiathome.berkeley.edu/host_app_versions.php?hostid=4082448)'s does. It has been running an app_info since before the app details came along and therefore has no references to stock apps.
[unrelated background information]
In the meantime, I'm doing a lot of data sorting/archiving to clear out enough space on the externals to hold the contents of the array for a few hours. Some time ago (I think it was ~2 years), one drive punted itself from the array randomly. Normally, that's not really that big of a problem (it's a 4x500gb raid-5). I just go in and delete the ghost raidset that thinks three disks are missing, then designate that disk as a hot spare, and the controller pulls it in and auto-rebuilds.
Well that one time two years ago, disk 2 punted itself, and for some stupid reason, I pulled the side panel off and touched one of the cables just right and two more removed themselves from the array as well. So then I had three ghost raidsets and the one correct one. Since each disk reported that it was a member of a raidset and lost contact with all the others, it was technically destroyed and not recoverable. I did a lot of searching around on the Internet and found that in the BIOS setup for the card, there's supposed to be an option for rebuilding where the initialize method is "None (Rescue)." Well mine didn't have that. After some more research, I found that particular option in the BIOS setup didn't come along until the next firmware revision after the original base release that my card has. However, the web interface (localhost:81) has the option in there.
So the fix for this scenario was to delete ALL of the volumesets and raidsets, create a new raidset with exactly the same parameters as before, but make sure initialize was set to Rescue. After an adrenaline explosion and many butterflies in my stomach, I hit the "confirm operation" button. It worked. The controller didn't automatically do it, but I went and told it to do a consistency check just to make sure everything was good. Ended up not having any problems once it was done. Fortunately there wasn't any write operations going on when I had my moment of stupidity.
But in the aftermath of that issue, the name for the raidset that appears is the same as it was before, except instead of #00, it is #R-00, R for "rescued." I had originally set it up for a 16KB stripe size, but I think I can get much better performance if I go to 64KB. I believe this because when I do read benchmarks, or just read large files from it, it tops out at 210MB/sec. Now the first time I had a disk punt itself, I thought for sure it was just a failed disk, so I panicked and ordered a new one from Newegg, and overnight shipping was only $6 more than the regular shipping that was like $2.49 or something like that. While I was waiting for the new disk to arrive, I put the "failed" disk into another machine and ran the WD DLG tools on it, specifically the long test that looks for errors. It found none. Did the short pass of zeroes (first and last million sectors), put it back into the main rig, booted up and told the controller it was a hot spare. Automatic rebuild.
Worked fine for a long time. But now I had a "spare" disk laying around. Of course I turned it into a storage disk that could be used as a raid spare (after swapping the cable from onboard to the card). Same model and everything, and that disk gets 115MB/sec on reads in the first 50GB, and doesn't drop below 100MB/sec until about 125-150GB. Even near the end, it's still pulling 65-70MB/sec. So since raid-5 on reads is basically a raid-0 stripe, I should be getting 300+ at the beginning, and well over 150 at the very end. But since I've only got 16KB for a stripe size, I max out at 210 at the beginning, and it drops down to 105 at the end, with a 175 average. That's still pretty good, but it should easily be twice what I'm getting. Of course I filled the array up too quickly and ran out of storage space elsewhere to rebuild it with a larger stripe size.
It sucks that HDD prices are sky-high right now, too. I was looking forward to picking up a 2TB drive to put in my external enclosure (it's a 5-bay trayless backplane on eSATA), so now my only option to fix my array is to burn stuff onto DVD+R's to clear out enough space to hold the array contents. I think I might just make enough room.. it'll be close. I've also wanted to upgrade the firmware on the raid card, but I'm not totally certain--nor confident--that it won't destroy the array in the process. Safest thing to do is have the data backed-up and then upgrade the firmware. Upgrading the firmware will give me a lot of new options.. such as "no alarm on fan fail." I've gone through five 40mm chipset fans in the 4 years I've had this card. I finally just said "enough!" and put the passive heatsink that came with the card on it, but with no fan, the alarm goes off. Yay old firmware.. ::)
[/unrelated background information]
Okay, so I don't really expect anyone to spend 5-10 whole minutes reading all of that, but I felt like explaining. If you read all of it.. I totally gave you an opt-out at the beginning. :P
-
I think your best bet would be to copy the old folder over, clean out all the files but leave all the folders in place.
-
I do have a question about the opt apps though. If I install BOINC but don't attach to any projects, and then create the project folder in the data directory and put my app_info and EXE in there, THEN attach to the project, will that work right away? My goal is to keep my application details from having any references to stock apps, so I don't want to attach and get assigned work right away after the stock downloads. I want the new machine's app details to look like my single core machine (http://setiathome.berkeley.edu/host_app_versions.php?hostid=4082448)'s does. It has been running an app_info since before the app details came along and therefore has no references to stock apps.
You can try attaching with an app_info and apps already popolated in the setiathome directory,
If i don't want to get Stock apps when i attach a host, i Deselect Use CPU & Use Nvidia GPU in my Default Project Preferences,
then when Boinc does the Prroject Initialisation it finds it can't ask for CPU or Nvidia GPU work, so it gets no Stock work, then you can do the app_info and apps,
Claggy
-
I think your best bet would be to copy the old folder over, clean out all the files but leave all the folders in place.
What about client_state? I want a new computer ID since it is a new computer.
Is it possible to set NNT on a global scale for BOINC before attaching?
-
I think your best bet would be to copy the old folder over, clean out all the files but leave all the folders in place.
What about client_state? I want a new computer ID since it is a new computer.
Is it possible to set NNT on a global scale for BOINC before attaching?
SETI@home preferences (http://setiathome.berkeley.edu/prefs.php?subset=project&updated=1)
As i said, deselecting 'Use CPU' and 'Use Nvidia GPU' will stop Boinc even being able to ask for work (from Seti) once it's done it's project initialisation,
But make sure you're set 'Default computer location' to same venue as you have deselected the preferences,
Claggy
-
SETI@home preferences (http://setiathome.berkeley.edu/prefs.php?subset=project&updated=1)
As i said, deselecting 'Use CPU' and 'Use Nvidia GPU' will stop Boinc even being able to ask for work (from Seti) once it's done it's project initialisation,
But make sure you're set 'Default computer location' to same venue as you have deselected the preferences,
Claggy
Buuuut.. those are enforced by 6.10+. I'm going to run 6.2.19.
-
SETI@home preferences (http://setiathome.berkeley.edu/prefs.php?subset=project&updated=1)
As i said, deselecting 'Use CPU' and 'Use Nvidia GPU' will stop Boinc even being able to ask for work (from Seti) once it's done it's project initialisation,
But make sure you're set 'Default computer location' to same venue as you have deselected the preferences,
Claggy
Buuuut.. those are enforced by 6.10+. I'm going to run 6.2.19.
How about installing Boinc 6.10.60, having the CPU and GPU options deselected, attach to Seti, install Optimised apps, then if you don't like 6.10.60, uninstall it and install 6.2.19?
(I think if you're going to run 6.2.19 you'll see more Credit anomalies since Boinc 6.12.19 doesn't report runtime like recent clients do)
Claggy
-
A new host will get the default venue, if that were set to only deliver "Astropulse v5" tasks the project couldn't send any.
Joe
-
I think your best bet would be to copy the old folder over, clean out all the files but leave all the folders in place.
What about client_state? I want a new computer ID since it is a new computer.
Is it possible to set NNT on a global scale for BOINC before attaching?
What I meant was to delete all the client_state, stderr, etc and just leave the 4 folders.
-
What I meant was to delete all the client_state, stderr, etc and just leave the 4 folders.
The four folders being:
projects\setiathome.berkeley.edu (app info and executable here)
slots\
and the base folder were cc_config is going to go?
I mean, it's sort of like migrating across installs. Install boinc, then it will open up and ask you to attach to something, shut it down completely, drop your backup into the right place, start it back up and it's fine. Except it won't have client_state and all those other xml files.. just cc_config, projects\setiathome.berkeley.edu and an app_info in there. Seems like it should work like that.
The suggestion about only allowing the obsolete _v5 is a good one, too.
-
So the RMA was approved and the replacement has shipped out today (Thursday). Just in time for UPS to hold it over the weekend and do a Monday delivery... yay! ::)
In the past, I would have had that nagging thought in the back of my mind that they sent me the same board that I sent them, but like all RMAs, you send them to Whittier, California, and the replacement is coming from Memphis, Tennessee. Still doesn't mean that they couldn't have just overnighted it from CA to TN, but that is unlikely. Guess I'll see what happens.
One thing that I did not test, which I thought of a day or two after I sent the board off, is that I should have put in a liveboot Linux CD and seen if the sound worked fine. That would have determined whether hardware or software was to blame. Like I said before, I heard windows sounds after the install was done and before I installed the drivers, and I don't recall exactly when the sound stopped working, but it was definitely after drivers were installed. Maybe it just went bad suddenly. It happens I guess.
And regarding the low voltage readings that I observed.. I just thought to pull up Everest on this old setup and take a gander at the voltages. They match what the new board saw. So my PSU is getting weak. I've got another one that is nearly identical (a 550 rather than a 600) and whilst it has been on and running for about 4 years, it has had much less of a load, so for the time being, I think it is going to be swapped in. That may account for the capacitor whine/squealing that I was hearing.
Also, I did go in and set my default settings to only the obsolete _v5, and do not accept from other applications, and told the default settings to assign the school venue, which has matching settings. I'm pretty sure the app_info trick will work before attaching, but just in case, that should be a reliable fail-safe.
edit: So I just swapped power supplies between the main rig case and the single-core machine's case. This board now reports 11.98, 5.03, and 3.31 for voltages. And then I had another thought: "I wonder if that will fix my inability to write to the array without the system locking up?" The answer: yes. I just wrote 40GB to the array and it is fine. So replacing the caps and fixing the voltages made the array usable again.
I guess the reason the new board didn't have any issues writing to the array is because the power draw on the new setup is about half of what this one is at idle. This old setup, according to my UPS, is sitting at 191 watts. I saw the new setup's load at idle sitting at the desktop was right around 120 watts. That difference made the array stable, I guess. You know, that whole voltage drop under load thing. Voltage was already low, so a load just brings it down farther.
So I do need a new power supply, but at least this swap to a better-condition one buys me some more time before it is necessary. I might go with a Corsair enthusiast series 750. I spec'ed out and built a rig for a friend of mine about a year ago and picked up one of those for him. I really like them. The cabling is nice, and they are basically silent. I used to swear by the Rosewill PSUs (what I currently have two of), but it seems the quality-control on them went to total crap after I bought my two.
-
Right, so I got my replacement board back. All is working well. I did the test of installing BOINC without attaching, and dropping the app_info in place. Still had to download all those PNGs that are useless, and got "project has no tasks available" for the first scheduler reply. Message log did not indicate that it detected an app_info, so I closed and re-opened BOINC and then it did. Waited for the next scheduler contact and got "no tasks available" again, changed the venue to the correct AP-only one, next interval, got one AP. Downloaded and is running.
Default ETA on it was 212 hours. Just in case it was either really high or really low, I set my cache preferences to 0.1 days to make sure I didn't get hundreds and have it be six months of work (yes, I know there are "limits" in place, but I have never once gotten a message about being at the limit even though I was over the limit). The present rate is showing that the ETA is dropping 15 seconds per second. Just passed 1.00% at 7:54 (m:ss). If that extrapolates out properly, it looks like it'll be about 13 and some change hours as opposed to my previous system running them in about 22. 300MHz and an architecture change makes a pretty big difference.
edit: Still only have one AP running, but I'm doing some x264 encoding that is using four more cores. 5 cores at full load and I'm running 64C with the stock cooler at stock speed and vcore. I wanted to use my 92mm Rosewill cooler, but my 8-year-old tube of Artic Silver 5 has seemingly dried up some. It doesn't spread very well, and the first couple of days I was using that, idle at 1400MHz was running 53C. However, in the box that the cooler came in years ago is a tube of the lesser-quality ceramic stuff. Might give that a try. I know it has to be possible to get under 50C at full load on all cores. Especially not even OC'ed.
edit2: So the stuff that was in the box for the heatsink seemed a little dry. When I was spreading it with a razor blade, it was clumping and peeling off in a few places. I just kept working at it until I got a smooth and mostly uniform very thin layer of it. Put the cooler on, booted up. Idles at 41C and ran HeavyLoad. Six cores at 100% for 15 minutes and it climbed pretty slowly to 55C and didn't go any higher. Watched the fan speed increase from 1100 up to 2200 (max) and once it hit 55C, the fan actually slowed down to ~2000 and stayed there and held 55C. I suppose that will work for the time being. Will definitely want to go with AS-5 though, when I get a fresh tube of it.
-
So after running on all cores overnight, I woke up to notice the durations were a bit longer than expected. This may be due to the shared FPU configuration after all. Of course the 20 or so "exited with zero status, if this happens frequently you may need to reset the project" instances may have something to do with it. For a test, I applied <ncpus>4</ncpus> to cc_config to see what difference that makes. I may need to kick it down to 3, but some more time will tell.