I am currently on the latest nvidia driver APPROVED for my 64-bit PC. Together they have CUDA 9.2 and CC 6.1.
It seems that I do not require a newer driver - can you be more accurate ?

Hi John,

First of all I can't see any cuda application running by your account at all. But if you are planning to change this I have a few notes.

I don't know what you mean by "latest nvidia driver APPROVED for my 64-bit PC". Have you checked this page: NVIDIA > Download Drivers? It states, that there is most up-to-date driver for tour GPU:
Version: 445.75 WHQL
Release Date: 2020.3.23

Having MX150 with such outdated driver version (398.35) will prevent you from running the latest cuda102 application. You still can use the older cuda55 but you may experience some issues. And you will definitely miss some improvements that has been introduced by NVIDIA with their latest CUDA SDK (Release notes ). The most important idea behind having updated applications or drivers is that in newer versions there are bugfixes applied, leaving aside improvements. Unless you have very important reason to keep an outdated driver I strongly recommend to upgrade yours. Even if you are not planning to use your CUDA enabled GPU for Asterodis@home.

Regards
Georgi

Some years ago I was advised by another project to do the same. So I upgraded to the latest WHQL version. It would crash from time to time. So I went to the PC manufacturer's page and downloaded their latest certified nvidia driver - and the problems disappeared. I guess I will just have to try it.

Here is Nvidia's warning on their download page -
As part of the NVIDIA Notebook Driver Program, this is a reference driver that can be installed on supported NVIDIA notebook GPUs. However, please note that your notebook original equipment manufacturer (OEM) provides certified drivers for your specific notebook on their website. NVIDIA recommends that you check with your notebook OEM about recommended software updates for your notebook. OEMs may not provide technical support for issues that arise from the use of this driver.

Your wording gives the impression that 10.2 is more efficient for the application - whereas it is actually a requirement. If you had made this clear, I would not have posted.

Good luck with the new app.

Version: 445.75 WHQL
Release Date: 2020.3.23

I updated 3 machines to this driver so that I could get the new WUs (cuda102win).

So far it's not looking good. These machines ran fine with the cuda55 WUs. Not only are they now using an entire CPU core but the new WUs make the machine nearly unresonsive. They also have a much higher failure rate than the cuda55 WUs. I've had to stop Asteroids work on these 3.

Can you suggest a solution other than reverting to an old NVidia driver?

All 3 updated machines are running: NVIDIA GeForce GTX 1060 3GB (3072MB) driver: 445.75 OpenCL: 1.2

Hi Beyond,

The code inside the new CUDA102WIN application is basically the same as in old CUDA55 application. Still there a some necessary bugfixes which does not affect the execution plan though.
Of course you can always switch to the oldest version, but without having those bugfixes applied your error rate will be significantly higher. Second major change comes from the new Nvidia CUDA libraries. Every one could check Nvidia Release notes and ChangeLog if interested in what was improved and fixed.

When the new application was finished it was deeply analyzed with the latest tools provided by Nvidia Insight Package before it has been released.
Since then CUDA102WIN did a lot of work, great work actually, on many configurations. Still there could be exceptions which could be caused by many external factors - Overclocking, bad heat dissipation etc.

One other thing that I can suggest is to downgrade form latest 445.75 Game Driver to latest 442.19 Studio driver and try.

Can you provide some more information about the CPU + GPU usage and how did you measure it - anything - screenshots, logs are welcome by your convenience for both CUDA55 & CUDA102WIN?

Thanks in advance.
Georgi

Hi Beyond,

The code inside the new CUDA102WIN application is basically the same as in old CUDA55 application. Still there a some necessary bugfixes which does not affect the execution plan though.
Of course you can always switch to the oldest version, but without having those bugfixes applied your error rate will be significantly higher. Second major change comes from the new Nvidia CUDA libraries. Every one could check Nvidia Release notes and ChangeLog if interested in what was improved and fixed.

My error rate was higher with the new version.

When the new application was finished it was deeply analyzed with the latest tools provided by Nvidia Insight Package before it has been released.
Since then CUDA102WIN did a lot of work, great work actually, on many configurations. Still there could be exceptions which could be caused by many external factors - Overclocking, bad heat dissipation etc.

These machines were built for crunching, have fantastic cooling and are not overclocked.

One other thing that I can suggest is to downgrade form latest 445.75 Game Driver to latest 442.19 Studio driver and try.

They're switched over to GPUGrid now (which is a far more demanding application) and are running perfectly there, as they do everywhere else.

Can you provide some more information about the CPU + GPU usage and how did you measure it - anything - screenshots, logs are welcome by your convenience for both CUDA55 & CUDA102WIN?

The CPU usage can be seen in BoincTasks and in Task Manager. GPU usage is monitored in MSI Afterburner.

Here's the 3 machines that I upgraded to new NVidia version that you suggested. You can see the specs yourself:

http://asteroidsathome.net/boinc/show_host_detail.php?hostid=610458

http://asteroidsathome.net/boinc/show_host_detail.php?hostid=610456

http://asteroidsathome.net/boinc/show_host_detail.php?hostid=656984

They ran fine with the CUDA55 app, as did all my other machines.

Hi Beyond,

You can give it a try now as we've made some changes over the project.
It will be just interesting to know if it is working or not.

I will not going to argue with you according error rate. I'll not quote your words either. I find such approach to be very useless.
Anyway.
Yes, we had some issued for last few days, and hopefully they will be resolved one after another. Some times in order to achieve better results or there is the need to implement support for new hardware or new data structure, there comes the issues as well. And we are trying to fix them as fast as possible. Thus any help from our contributors is more than welcome.
I asked you if you could give us some more detailed info of what you measured. But instead you gave us patronizing speech.
Correct me if I'm wrong but I'm getting mixed feelings here. If you feel better with some other project, please, nobody is forced to stay.

Here we all are one big team - Scientists, Administrators, Developers, Contributors and we are helping each other.

Now to the problem. Every application has its own specifics. Period_search is not an exception and in order to work correctly it needs to be controlled by the Host system. That could lead to some CPU usage - usually a single logic processor - which is totally normal and which depends on the CPU or GPU vendor, architecture, generation, software libraries and last but not least it depends of the specifics inside the code base.

Have a good day.
Georgi

I posted the problems I found with your new app simply to report the bugs and to allow you to track them down. I quoted you so that I could answer your questions individually. Isn't that the purpose of quoting? I'll try the new version but am now reticent to give a report. You should be encouraging bug reports. I've run a lot of WUs here and at many many other projects for decades. I don't need to be talked down to.

Regards/Beyond

BTW, just what was my patronizing speech?

The new fixes helped as the Windows 7 interface is now normally responsive.
The WUs took about 60% longer to complete than with CUDA55.

Hi Beyond,

Thank you for your feedback.

On one hand complete times of WUs is not a constant value. There are many factors which has impact on that. Different amount of Light curves or amount of data per LC is one of them. Actual size of the WU is not determinant. Large WU can be done with small number of iterations while small one could take twice or more number of iterations. There is no rule. It takes what it takes.

On other hand there are rules on the server. Very complicated algorithm works behind which may decide to send more heavily WUs to proven faster host while in the same time keeps sending lighter ones to slower machines. It is constantly changing strategy based on many many factors along with the changes in hosts behavior. If you are interesting BOINC has detailed documentation available online.

And last but not least it is all about precision. This is not a race.
Old applications suffer from bugs, so for now we well keep them only for back compatibility until users update their software and for some reasonable time.
Keep on mind that old applications could now generate some faulty output results.

This is why we encourage our contributors to upgrade their platform to the latest software platforms and driver packages if possible.

Thanks.
Georgi

This is not a race.

It is now. You've been selected for the next Formula-BOINC Sprint.

http://formula-boinc.org/sprint.py?sprint=4&lang=&year=2020

---

Thank you Georgi for you explanation. I wish other projects, like for example my beloved Gaia@Home had this level of communication with the volunteers.

It is now. You've been selected for the next Formula-BOINC Sprint.

Oh man, I hate those. I've seen what happened to Universe@Home when it was selected for a race. First people hoarding tasks and later, when the race started, overloading the servers to the point no real work could be done. I think in the long run those races do more harm than good, but that's just my opinion.

---

my blog about raspberry pis and astrophotography

This is why we encourage our contributors to upgrade their platform to the latest software platforms and driver packages if possible.

What in particular should we make sure is up to date? Since this is not a GPU project, that leaves that complication out. AFAIK there are no drivers to help CPUs behave better.

Actually since there are Cuda application tasks this is a GPU project if I am correct ?

Bill F

---

In October 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.

This is why we encourage our contributors to upgrade their platform to the latest software platforms and driver packages if possible.

What in particular should we make sure is up to date? Since this is not a GPU project, that leaves that complication out. AFAIK there are no drivers to help CPUs behave better.

Actually since there are Cuda application tasks this is a GPU project if I am correct ?

Bill F

It is not a GPU project. As you can see here Applications there is a lot of hardware supported.
_______________________________

Actually since there are Cuda application tasks this is a GPU project if I am correct?

I hadn't noticed them (I only have AMD cards). What's strange is the total Gflops shown for Cuda is much lower than for CPU. Are those tasks rare?

Bill, how many do you run concurrently on your nvidia card? So I can get an idea of speed.

And Georgi, any plans for OpenCL? My Tahitis are hungry. So little physics projects for GPUs.

Well, I can only repeat myself from here: https://asteroidsathome.net/boinc/forum_thread.php?id=624&postid=6561

Hi guys,

As I said in one of my posts here:

Creating OpenCL application to support AMD GPUs was always at our RoadMap for years. Even now there is ongoing development on that. No mater there were some turbulence during the years, postponing and getting back on, we are doing our best to make it happen. Once we have solid PoC application we will be glad to provide it to our contributors.

Cool things takes time.

Thanks.
Georgi

Happy holidays!

---

“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson

Actually since there are Cuda application tasks this is a GPU project if I am correct?

I hadn't noticed them (I only have AMD cards). What's strange is the total Gflops shown for Cuda is much lower than for CPU. Are those tasks rare?

Bill, how many do you run concurrently on your nvidia card? So I can get an idea of speed.

And Georgi, any plans for OpenCL? My Tahitis are hungry. So little physics projects for GPUs.

Bill here I believe that my NVIDA GX 1060 runs 4 concurrently and the Run and CPU times are a fraction of the time on my CPU only systems.

---

In October 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.