Collatz: Phasing out Multi-Size

[RSS]

Due to a variety of server scheduling problems that have not been able to be resolved, Collatz will be phasing out the multi-size workunits over the next week or two and going back to using multiple applications. You will once again control your own destiny and be able to choose the size WUs by the application chosen (micro, mini, standard, and large). The big addition (pun intended) that will be the first to be deployed is the large_collatz application which will be 32 times the size of the largest multi-size workunit. It is aimed at the fastest GPUs to reduce the upload/download and workunit startup overhead of running smaller workunits. That should result in a higher RAC as well as a reduced load on the Collatz server.

More... (http://boinc.thesonntags.com/collatz/forum_thread.php?id=1156)

[finalfugue]

The big addition (pun intended) that will be the first to be deployed is the large_collatz application which will be 32 times the size of the largest multi-size workunit. It is aimed at the fastest GPUs to reduce the upload/download and workunit startup overhead of running smaller workunits. That should result in a higher RAC as well as a reduced load on the Collatz server.

Sweeeeeeeet. What was EmSti saying about 295Xs now being in stock (http://www.setiusa.us/showthread.php?6745-Titan-Black-Now-Obselete&p=77741#post77741)?

[FourOh]

The project is transitioning now... one of my machines detached itself from the project, another is still crunching Solo but not getting any new work. Need to check Collatz account settings and keep an eye on these rigs to make sure they don't dry up.

In order to receive a Large unit, select "Test Applications" in account settings. I just got one for my 7790... we shall see how it goes!

[finalfugue]

Thanks for the heads up. Things are a little too hectic at work today to ensure a smooth changeover, but I just looked and under Collatz preferences you now have the option to select which application(s) you want to run. I have anything ATI Pitcairn and better set to solo and large, and lesser stuff on mini, micro, and solo. Just trying to keep it chugging. I'll check later to see how the numbers are shaping up.

[Al]

During the transition it might be a good idea to click the checkbox to accept work from other applications.

[EmSti]

The project is transitioning now... one of my machines detached itself from the project, another is still crunching Solo but not getting any new work. Need to check Collatz account settings and keep an eye on these rigs to make sure they don't dry up.

In order to receive a Large unit, select "Test Applications" in account settings. I just got one for my 7790... we shall see how it goes!


Test applications is not needed. I change my to solo and large only for the gpu and solo only for cpu. I started getting large on cuda and amd a short while later. Remember to edit the new config file that gets downloaded to match what you had in the solo config file.

Anyone know if a wingman will be needed on any of them?

[Bryan]

No Wingmen required on the large WU :D

[Al]

Hmmm, my first 2 large wus on my dual 7970 system estimates completion time @ 1550 hours.

[EmSti]

Hmmm, my first 2 large wus on my dual 7970 system estimates completion time @ 1550 hours.

The estimates are wrong. So far I am averaging 4.9 hours on the 7970, I have one cpu thread thread dedicated to each of the 7970 tasks. About 239,600 avg per large wu.

[Al]

LI'm 7 hours in and 20% done. Gpu load is 50% on one and 80% on the other 7970. The .config file is set. 2 cores are free.

[EmSti]

LI'm 7 hours in and 20% done. Gpu load is 50% on one and 80% on the other 7970. The .config file is set. 2 cores are free.

Where the large config files set before the large wus started? It not, I would think about dumping the running ones. You could try it on the 50%. The load should be the same as solo collatz.

[Al]

No they weren't. I dumped them and the loads came up to 99% on one and the other is pegged at 100% on the replacement wus. The 100% I find unusual and that one is running sloooow. The one at 99% is moving along at a pretty good clip.

[Shandia]

OMG! I just got the first GPU task of the new tasks. Since I'm currently only running GPU tasks for Collatz, I have no idea what the cpu tasks look like. :-t According to Boinc, this one ATI GPU task is going to take me 32 hours to do. YIKES!!! I hope the credits will reflect the long time needed to crunch this. I'm hoping the estimate is a mistake. If it is even remotely close, I can tell Collatz will be put totally away until I can get around to crunching it again with my cpu's. It's just too scarey for us little crunchers now. I'll probably move back to seti@home until I find another good gpu project to crunch on. :((

[finalfugue]

Really? I don't think they're that bad, about what I expected. I did what EmSti said about reconfiguring the new app files and dumped those that started with old settings (all before reading it, which made me feel smart for a n00b). Also, I'm not sure the new smaller apps/WUs have premiered yet (anyone running those have any input?). Those may better suit your hardware and you can adjust preferences soon. AFAIK old solo_collatz is still going strong and has reasonable sized WUs. I'm sure Slicker's credit granting scheme will remain much the same, so in the end you should end up about where you are now. Just with faster cards you don't have to deal with all the cycling. Also from what I've seen the first time BOINC sees something it pretty much freaks and takes a WAG. I'm sure someone here can help get it sorted- once the dust settles.

Sent from my HTC6500LVW using Tapatalk

[Shandia]

I'm sure someone here can help get it sorted- once the dust settles.

I am watching it closely. I went ahead and unchecked the Large Collatz Conjecture option to see if anything different rolls out of the pachinko machine in way of gpu tasks. Every change always needs a little time for the dust to settle so the cleanup can begin. :)

[Shandia]

I went ahead and re-check the large option due to me missing changing the app_config file. I knew something was up when I checked my card usage. Just in case, how should the app_config file look now? Not sure if the placement of <name>large_collatz</name> is right along with the <name>solo_collatz</name> and other info. My usage was at 74% on the video card and is showing 0.904 cpu usage. 8-|

After a few more changes, it is showing to be running at the sweet spot.Unfortunately, my gpu usage has now dropped to 25%. Somethings not right here. I think I broke it! :confused:

Maybe it broke itself?? It just errored out and loaded up a ton of solo tasks. The gpu is back to showing 99%. I guess things do work out on their own after all. :-bd

[Al]

Well, I've only got 1 7970 that will run the large wus well. The other is at 55% and 11 hours in, both in the same box. The 570s really aren't fast enough for my taste. Back to solo for me.

[Bryan]

Al, check in your projects folder and make sure that it didn't download the 32 bit app on you. The slower GPU "might" be running the 32 bit version and if so you need to put your settings into the "new" 32 bit config file.

Last night the same thing that happened to EmSti got me. BOINC in its infinite wisdom decided I needed to try the 32 bit app as a speed comparison. When I got up this morning my credits were about 1/2 of normal on the GTX 780 and that was the problem. I put the settings into the new config file and things are back to normal now.

PS I tried 2 of the large WU and had no problem with them. One granted credit at a 727k daily run rate and the 2nd at 777k. My card has been doing 805-830k per day so I went back to Solo.

[Shandia]

The solo tasks seem to be working for me. I almost can't wait for the ton of solo tasks to run out just to give the large another try. In so many ways, I think Boinc 7.2.42 is to blame for some of it. It seems to be a little laggy when it comes to changes on the fly. On my core 2 quad with an nvidia 9500gt, docking is still having issues and I lost a day due to it. I can also suspend a task and for about 10 seconds, despite showing that the task is suspended, the timer and % is still ticking away. So far I haven't touched it as far as Collatz is concerned. On my core I7 with the HD 5770, the time between me making changes to the app_config to try to get things working like they should, it seems to be taking close to a minute. (I have been changing the cpu thread count to see if I'm running 8 cpu tasks or 7cpu tasks after each change.) I would love to know if what I have for the app_config is correct or not. (kinda why I want another large file to come my way.) I tried a few different arrangements in the file and when the large was configured and working with the correct cpu usage.....then errored out, the solo loaded up and ran at the default settings. After making more changes in an attempt to "fix" it, it took a while for me know if the changes worked or not. Now, the solo seems to be working right. This is what I have the app_config as of currently. (well.....I actually have the cpu running at 0.98 instead of at 1.0. But I'm sure you get the idea with this.)

<app_config>
<app>
<name>solo_collatz</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
<app>
<name>large_collatz</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
</app_config>

[Al]

Nope, no 32 bit app installed. I'll let them finish, then go back to solo.

[Z-TAC]

* * *

<app_config>
<app>
<name>solo_collatz</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
<app>
<name>large_collatz</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
</app_config>

Isn't that just the default setting for the project without an app_config file?

[zombie67]

I hope the credits will reflect the long time needed to crunch this.

Never fear. Slicter always makes sure we get paid. Remember that april fools news post from collatz?

Edit: Slicker. Stupid autocorrect.

[Shandia]

Isn't that just the default setting for the project without an app_config file?

Not exactly. It is close. The default is to run 0.903 of a cpu thread...or at least that what it was showing me on my tasks that were running without any changes. I tend to try and micro-manage my tasks to do things exactly how I want it. If one can find some slight variance to improve their performance, I say go with it! I often will extend the number of threads far more than the default if I can get a better return. (I've actually run some gpu tasks using all 8 cpu threads on some projects including Collatz.) But if it wants to read as default instead of what I want it to use, it's pretty much useless to have. ;)

[Shandia]

Never fear. Slicter always makes sure we get paid. Remember that april fools news post from collatz?

Oh, I remember! And every time it popped up after April 1st, I was reminded. On April 1st I wasn't signed up for the project yet. After that scenario, I had to wonder if the message about phasing out multi-size tasks was a joke or not. Until I saw my first large task, I didn't know if it was for real.

[Fire$torm]

Well I'm trying a Large wu on my GTX 570 atm. BOINC currently says 86+ days but we all know that is just BOINC freaking out. My calcs estimate 11+ hours. And yes, I set up the Large config before starting the wu.

Anyone getting better times on their 570s?

[Al]

Mine were finished in 12.2 hours.

[finalfugue]

It certainly seems to be a big boy playpen, that's for sure. My 280Xs are wrapping up about like EmSti's 7970s (unsurprisingly), but the 7870 is sucking wind. Over 8, nearly 9 hours to complete a big one for about 221K. I think I'll take that back to solo after completing one-two more for comparison's sake, but I'm not used to CC taking anything Pitcairn or better out behind the woodshed. I know what I was on track to hit total credit wise, and if I don't see it in the next couple days- oh well, just have to adjust. It's still a useful tool for those 4 people who bought the 295X.

[Bryan]

As a footnote, the GTX 780 takes a little over 7 hours to finish one. I did 2 and moved back to Solo.

[Slicker]

The project is transitioning now... one of my machines detached itself from the project, another is still crunching Solo but not getting any new work. Need to check Collatz account settings and keep an eye on these rigs to make sure they don't dry up.

In order to receive a Large unit, select "Test Applications" in account settings. I just got one for my 7790... we shall see how it goes!

Test Applications is no longer required.

[Slicker]

OMG! I just got the first GPU task of the new tasks. Since I'm currently only running GPU tasks for Collatz, I have no idea what the cpu tasks look like. :-t According to Boinc, this one ATI GPU task is going to take me 32 hours to do. YIKES!!! I hope the credits will reflect the long time needed to crunch this. I'm hoping the estimate is a mistake. If it is even remotely close, I can tell Collatz will be put totally away until I can get around to crunching it again with my cpu's. It's just too scarey for us little crunchers now. I'll probably move back to seti@home until I find another good gpu project to crunch on. :((

Unfortunately, Dr. Anderson has ignore all requests to improve the estimated time remaining logic and as such, they SUCK!!!!!! Finish several WUs and it will start working better.

[Slicker]

Well, I've only got 1 7970 that will run the large wus well. The other is at 55% and 11 hours in, both in the same box. The 570s really aren't fast enough for my taste. Back to solo for me.

new app name means new config files. Just copy the settings from the solo*.config to the large*.config and you should be fine.

[Slicker]

Nope, no 32 bit app installed. I'll let them finish, then go back to solo.

See the latest News post in collatz and if an app is missing, let me know. I think I have them all now, but if I missed some, just let me know.

[Slicker]

As a footnote, the GTX 780 takes a little over 7 hours to finish one. I did 2 and moved back to Solo.

BOINC freaks out if I try and use the same config file name for multiple apps and keeps overwriting any changes the users make. So, for now, each app has to have its own config which is a bit of a pain. Everyone finally got the solo*.config files set up and now they need to deal with mini and large config files as well. Sorry guys, but I haven't found a way to have one config per GPU type regardless of app version or platform.

[Bryan]

BOINC freaks out if I try and use the same config file name for multiple apps and keeps overwriting any changes the users make. So, for now, each app has to have its own config which is a bit of a pain. Everyone finally got the solo*.config files set up and now they need to deal with mini and large config files as well. Sorry guys, but I haven't found a way to have one config per GPU type regardless of app version or platform.

Not a problem Slicker, I really like the new method of being able to optimize for an individual GPU type! I don't know what would happen if you had a mixed bag of GPUs in a single machine but I certainly don't have that problem ....I wish I did :D

The reason I moved back to Solo wasn't due to complaints on my cards speed since it was optimized. That is just the best that a 780 can do.

[Shandia]

BOINC freaks out if I try and use the same config file name for multiple apps and keeps overwriting any changes the users make. So, for now, each app has to have its own config which is a bit of a pain. Everyone finally got the solo*.config files set up and now they need to deal with mini and large config files as well. Sorry guys, but I haven't found a way to have one config per GPU type regardless of app version or platform.

I'm with Bryan on this one. It may take a little time to get each size optimized for individual systems. But I think it will be worth it since the resources I use between the solo and large tasks will be different. Waiting for the last 3 solo tasks to end so I can start playing with different configs for the large. I'll most likely do it to each size over the next week. (I've already added a couple of extra cpu threads for the large while using less on the solo)

[Shandia]

Ok....managed to work a large task. 235,798.74 for almost a full 29 hours with my best rig. It was a long day! I'll have to make a comparison of other tasks to know if it was worth it. May be great for bunkering....not so great at the end of a challenge.