Posts by Eric Driver

21) Message boards : Number crunching : Too much credit ? (Message 2343)
Posted 24 days ago by Profile Eric Driver
Post:
Jean-Luc:
Thank you for the well thought out and detailed post. Your understanding of the project is quite good for an "amateur". I should hire you to handle our social media and marketing!

But yes, I should do a better job of explaining the fundamentals behind this project, something which I find very hard to do. I will rework the home page and add a few words as you suggest. Maybe I could also start a thread on the message boards where people could ask questions. Anyways, this will take a few weeks to get to, as I am focusing right now on getting the rest of the gpu app versions done. But please remind me again in a few weeks, in case I forget to come back too it.

To answer your other question, the WUs are exactly the same for both gpu and cpu.
22) Message boards : Number crunching : Too much credit ? (Message 2339)
Posted 24 days ago by Profile Eric Driver
Post:

Eric, would it be possible for you to make notes on your credit experiences during this transition? I really think that BOINC should - belatedly - address the real effectiveness of these various credit schemes (especially at times of transition), and your cool-headed observations would be most helpful.


Sure, I can send you my observations via PM if that works for you. Or would the BOINC projects mailing list be better?
23) Message boards : Number crunching : Too much credit ? (Message 2336)
Posted 25 days ago by Profile Eric Driver
Post:
I agree, the credit is too high. As a temporary bandaid, I have reduced it. And I can further reduce it if need be.

Credits are currently set to a fixed value per WU (based on average runtime). It looks like this was set too high. I am not into the credit thing, so I have not been paying attention to it and the message boards were quiet about it (until now).

It seems like a fixed credit per WU is fair. If a GPU is 20x faster it will get 20x the credit per hour. From what people have said, it seems like other projects pay a disproportionately higher number of credits for GPUs. Is there a good reason for this or is it just to attract more users?
24) Message boards : Number crunching : GPU Error (Message 2325)
Posted 26 days ago by Profile Eric Driver
Post:
I'm also seeing a few tasks with memory access issues.

Task with error: https://numberfields.asu.edu/NumberFields/result.php?resultid=40204463

This is Nvidia 418.54 with 980m GPU and application version: Get Decic Fields v3.01 (cuda30)
x86_64-pc-linux-gnu

There are successful tasks too on the same system: https://numberfields.asu.edu/NumberFields/result.php?resultid=40234951


Thanks for reporting. It looks like you have a couple tasks with that same error.

I have seen some of these before on other hosts. When I rerun them with cuda-memcheck they finish successfully with no errors. But of course I have a different GPU, so maybe it's something with that particular compute capability. I'm still trying to figure it out. At least it doesn't happen too often.
25) Message boards : News : GPU app - beta version for linux nvidia (Message 2322)
Posted 26 days ago by Profile Eric Driver
Post:

Great to hear that you have one more thing to try. Much appreciated.

Tasks were comming, but would sit for hrs with no tasks..
I was set for 0.10 work, I have now adjusted to 0.50 and see if that makes a difference..

Will let you know if that changes anything


I created a new thread for this issue.
In a nutshell, I think its fixed now.
26) Message boards : News : GPU app - beta version for linux nvidia (Message 2320)
Posted 27 days ago by Profile Eric Driver
Post:
If anyone cares, I have calculated average task times for some of my cards (sample size 50):

GTX 1660 Ti: 239 secs
GTX 1080 Ti: 306 secs
GTX 1070 Ti: 337 secs
GTX 980: 333 secs

These are with two tasks running in tandem on each card. Hooray for Turing!


Interesting! Thanks for sharing.
27) Message boards : Number crunching : scheduler problem (Message 2319)
Posted 27 days ago by Profile Eric Driver
Post:
You may have noticed the server scheduler has been very fickle of late.

I think part of the problem is related to how WUs have been segregated into batches (1 for normal app and 1 for GPU app). Since the GPU app has been relatively stable and has agreed with the CPU version on every WU tested (over 400), I have decided to remove the batch descriminator (this will open up the DS14x12 WUs to the GPU app).

I now have over 100 GPU WUs. Interestingly enough they are all of the DS14x12 variety (batch 0) and all have a higher priority (due to accelerated retry mechanism). Ironically, I think this points to the root cause of the scheduler problem...

I think this may have fixed the problem, at least for me it has. Please let me know if you still have problems receiving tasks.
28) Message boards : News : GPU app - beta version for linux nvidia (Message 2317)
Posted 27 days ago by Profile Eric Driver
Post:
Any estimate on when we'll have an app that will run under windows?


Hard to say. I need to get he OpenCL version working first before I can answer that (my hope is that porting openCL to windows will be easier than cuda).
29) Message boards : News : GPU app - beta version for linux nvidia (Message 2316)
Posted 27 days ago by Profile Eric Driver
Post:
Excellent stuff for the GPU app,

Is there a daily quota?
I ask as I can not get any tasks.

Thanks


The daily quota is high. I too have noticed the scheduler is a little fickle at times. I get the uninformative message "got 0 new tasks" after the client clearly requested new tasks. I turned on the most verbose debug setting in the scheduler and looked at the log after clicking update in the client. Still nothing to say why it wont give work. And then 30 minutes later, for some reason, it downloads a bunch of work. I've been over all the possible server config options and see nothing that could explain this. I've also looked at the feeder and it's definitely interleaving work for both batches, so I'm pretty sure the scheduler is the problem.

There is one other thing I am going to try...
30) Message boards : News : GPU app - beta version for linux nvidia (Message 2312)
Posted 28 days ago by Profile Eric Driver
Post:
Out of tasks again?


No, there's enough.

I too have noticed the scheduler being stingy with GPU jobs. My client had no GPU jobs this morning and after clicking update, it only gave me 5 new tasks.

This might be related to the "daily result quota" on the server. The day before I had hundreds of jobs in my queue that all failed after an auto update that changed my Nvidia driver. Ever since that, the server does not want to give me tasks. If you aborted a bunch of jobs, you might be in the same boat as me.
31) Message boards : News : GPU app - beta version for linux nvidia (Message 2306)
Posted 29 days ago by Profile Eric Driver
Post:

Regardless of the cards I am running, or the power you think I use, the card is compute 3.0 capable AND is running 418.56 drivers. It should work.


Yes, it successfully built for compute capability 3.0 and higher, so it "should" work. But unfortunately it does not.
32) Message boards : News : GPU app - beta version for linux nvidia (Message 2305)
Posted 29 days ago by Profile Eric Driver
Post:
There is only one GPU in the Top Computers list older than the Maxwell series successfully crunching the GPU WUs. It is a GTX 780. My GTX 1050 Ti has an average run time 20% faster than the GTX 780 and it uses less than half the electricity. Time to consider upgrading those Kepler and Fermi GPUs.


So it works for compute capability 3.5 or higher. I will change the plan_class to reflect this.
33) Message boards : News : GPU app - beta version for linux nvidia (Message 2299)
Posted 29 days ago by Profile Eric Driver
Post:
You could try upgrading to the 10.1 version of cuda. Other than that, I wouldn't waste your time, this exe may just not work on "older cards". Maybe the OpenCL version will fare better.
34) Message boards : News : GPU app - beta version for linux nvidia (Message 2296)
Posted 29 days ago by Profile Eric Driver
Post:
Soooo is there a time frame before I can throw my Radeon cards at this?
It's a bit of a shame to see my hours and hours of CPU crunching now being smashed by NVIDIA GPU work :(

Oh...and you're going to need to rework your badge system with all these credits. 100Mil is nothing now.


The OpenCL stuff is turning out to be more work than I thought, but maybe a few more days.

And yes, the badge system has needed an overhaul for a while now. If only I had more time...
35) Message boards : News : GPU app - beta version for linux nvidia (Message 2295)
Posted 29 days ago by Profile Eric Driver
Post:

Tasks vary in length so fixed credit is unfair as well. The only other way is if the FLOPs calculated during the task was known for each task. The prior method was open to abuse since it was a function of the GPUs FLOPs and time.


It may be unfair, but I would argue it's more fair than the previous methods. Credit from runtime is open to abuse and CreditNew pays next to nothing. I have computed the average FLOPS over 200 WUs for the current dataset and based the credit on that. In the long run this should be fair and accurate based on the law of large numbers (isn't this the principle behind CredtNew anyways?). The only abuse could be cherry picking, but with GPU tasks only taking a few minutes, someone would have to be insane to sit there and try to distinguish the fast from the slow.
36) Message boards : News : GPU app - beta version for linux nvidia (Message 2293)
Posted 29 days ago by Profile Eric Driver
Post:
It was paying around 20 per run sec on my GTX 1080 which is better than many GPU apps but not crazy. Now it's only paying 0.08 per run sec, 1/250 of what it was which is crazy low. It's worse paying than Asteroids on a GPU. Einstein pays 5-7 per run sec.


OK. I changed credit to a fixed credit based on average runtime. Hopefully this will be a good compromise. I tried to target what Einstein pays (actually a tad more). I can adjust later as necessary. Now we wait and see...
37) Message boards : Number crunching : GPU Error (Message 2292)
Posted 29 days ago by Profile Eric Driver
Post:
fully misstake.

Tried to install the cuda*.run package, guesses ?

Even i marked the install of the cuda toolkit,
the setup doesn't install an path the "lib64" folder for
the needed libraries and config files.

The manually setup don't work for me,
Keyboard and mouse were uninstalled
for unknown reason. I've had it all.


Sorry for all your troubles... sounds frustrating.
38) Message boards : News : GPU app - beta version for linux nvidia (Message 2283)
Posted 29 days ago by Profile Eric Driver
Post:
I added the min_cuda_version to the plan_class, to see if that's the root cause of the problem with the 6xx cards (if it is, I would expect them to stop receiving tasks)

On the flip side, let me know if properly functioning hosts stop receiving tasks, and I can lower the min version.
39) Message boards : News : GPU app - beta version for linux nvidia (Message 2282)
Posted 29 days ago by Profile Eric Driver
Post:
So it would appear 6xx cards have a problem with the gpu app. Most everyone with issues has that in common.
40) Message boards : News : GPU app - beta version for linux nvidia (Message 2279)
Posted 29 days ago by Profile Eric Driver
Post:
lol, at least you guys are getting tasks. I just keep getting "got 0 new tasks".

I disabled cpu tasks, just to test the GPU versions. I am running driver 418.56 AND I have a compute 3.0 card. Are there any tasks or am I configured incorrectly?


There should be plenty of tasks available. And I see no reason why the scheduler would not give you work.

Is the 3.01 executable in your project directory? I had the same problem last night, where I had the old 3.00 executable and it wouldn't download the new exe or send new tasks...but it eventually started working.


Previous 20 · Next 20


Main page · Your account · Message boards


Copyright © 2019 Arizona State University