Recent Changes in projects

[StefanR5R]

Shortly after my report, damotbe responded and relaxed some server-side rules how resends are distributed. "Unreliable" hosts are receiving work again.

 

[Orange Kid]

Incase you haven't noticed the notice in the Boinc manager

GPUGRID: Suspending project Until we update the app. Sorry. 10/7/2020 6:51:31 PM · more...

 

[StefanR5R]

That was unusually fast.

Dears, thanks for your patience.

I updated the acemd3 apps.

Also, I verified that there were very few results by apps from the old CUDA version, so I won't be re-deploying them. In other words, apps now require CUDA 10 (Linux) and CUDA 10.1 (Windows). In terms of drivers versions:

CUDA 10.1 (10.1.105) >= 418.39 for Windows
CUDA 10.0 (10.0.130) >= 410.48 for Linux

https://docs.nvidia.com/deploy/cuda-compatibility/index.html

(I never wanted to update my Pascal based hosts from their 3xx driver. I am now glad I did so for SRBase after all; it'll now pay off for GPUGrid too.)

 

[StefanR5R]

PrimeGrid

Pierre de Fermat, look what you kicked off; couldn't you have just stuck to lawyering?

Within the zoo of different species of prime numbers, we all are of course just waiting for sightings of more (or any) of the following ones:

• Wall-Sun-Sun primes (also widely known as Fibonacci-Wieferich primes),
• near Wall-Sun-Sun primes,
• Wieferich primes,
• near Wieferich primes.

Apparently PrimeGrid looked for some of these already in the past, but didn't double-check. The search is planned to be restarted, and I presume the double-checking will come first.

PrimeGrid forum thread: Welcome to the Wieferich and Wall-Sun-Sun Prime Search!

Edit, old thread about the stop of the corresponding prior PRPNet searches: WSS and WFS are suspended (from 2017)

 

[StefanR5R]

pschoefer of SETI.Germany posted some background about the Wieferich and Wall-Sun-Sun Prime Search (WW) subproject. (In German; kindly use the translator of your choice.) Quite informative.

Application versions: OpenCL for AMD and NVidia GPUs, multithreaded CPU application; each one for x86-64 Windows, x86-64 Linux, x86-64 OS X.

 

[StefanR5R]

PrimeGrid Generalized Fermat Prime Search subprojects (GFN)

The CPU application versions have been changed as follows.
The GPU application versions remain unchanged.

cpuGFN22: removed

• This was an application which looked like a multithreaded one, but really only used a single thread. It was therefore not viable and has been pulled. GFN-22 is now a GPU-only subproject.

cpuGFN21: unchanged

• This is a multithreaded application which uses an FP64 transform, which is still optimum for the CPU application version of this subproject. My understanding is that a port to a different transform may be called for when the subproject progresses past some point.

cpuGFN20: updated from Genefer 3.3.4 to Genefer 3.3.5
cpuGFN19: updated from Genefer 3.3.4 to Genefer 3.3.5
cpuGFN18: updated from Genefer 3.3.4 to Genefer 3.3.5
cpuGFN17MEGA: new; based on Genefer 3.3.5
cpuGFN17LOW: updated from Genefer 3.3.4 to Genefer 3.3.5
cpuGFN16: new; based on Genefer 3.3.5

• GFN-16 and GFN-17-Mega have been GPU-only projects recently but can be run on CPUs again now.
• All of these applications are singlethreaded.
• Genefer 3.3.4 used classic 8087 style math. Genefer 3.3.5 is now using FMA, AVX, SSE4, or SSE2 depending on what the CPU supports. The x86-32 application version can fall back to FP64 on really old CPUs which don't even support SSE2.
• Genefer 3.3.5, when using FMA, is presumably five times faster than Genefer 3.3.4.
• That's nice, but cpuGFN20, cpuGFN19, cpuGFN18, being single-threaded, are not very viable to run on CPUs in practice. Their data allocations are 16 MB, 8 MB, and 4 MB respectively. Therefore many CPUs will have too small caches to properly support several concurrent tasks or even just one task of these applications, as they become bottle-necked by I/O to main memory.
• cpuGFN17MEGA, cpuGFN17LOW, and cpuGFN16 may perform reasonably though, even in comparison with GPU application versions, I presume. (I haven't tried.)

GFN15: unchanged

• There is no CPU-only application version. There are only GPU application versions available for this subproject.

Sources:

anouncement thread
prior testing thread​

 

[emoga]

Sidock@home has moved from a test project into production and is now exporting stats.
Another boinc project that aims to fight COVID-19.

CPU Workunits are available for linux, windows and mac.
Can we get TAC to create a team for the TeAm?

 

[Fardringle]

View attachment 35890
Sidock@home has moved from a test project into production and is now exporting stats.
Another boinc project that aims to fight COVID-19.

CPU Workunits are available for linux, windows and mac.
Can we get TAC to create a team for the TeAm?

The project does have applications for Windows, Linux, and MacOS, but currently does not have any work available for Windows. It also limits tasks in progress to 2X the number of CPU cores, so no bunkering. But the server seems to be responsive and the admin is posting regular status updates.

 

[emoga]

@Fardringle Make sure you update to the new url

SiDock@home

www.sidock.si

after detaching from the old test site.
I just got 16 CurieMarieDock tasks on a windows host after your post.

 

[Fardringle]

@Fardringle Make sure you update to the new url

SiDock@home

www.sidock.si

after detaching from the old test site.
I just got 16 CurieMarieDock tasks on a windows host after your post.

I am on the new/current URL. The server status page said that it had no work available for Windows when I posted that message. They must have added more since then.

 

[Fardringle]

I'm not TAC, but I went ahead and created a team for SiDock so those who want to run the project can get joined up. I'll transfer founder status to the TAC account once it is set up there..

TeAm AnandTech

www.sidock.si

 

[StefanR5R]

PrimeGrid – Wieferich and Wall-Sun-Sun search (ww)

The application was updated and is now ~3 times faster. This performance improvement was used to expand the search space. Therefore, new workunits with the new application version take about as long as old workunits with the old version did. But the new workunits give 3 times more credit: 12,000 per result. This puts ww's PPD quite near those of PPS-Sieve's.

(sources: testing thread, some PPD examples)

 

[StefanR5R]

NumberFields

I just released the new app version for Linux Nvidia Cuda.

I tested it on 100 WUs and it averaged about 33% faster over the previous Cuda version.

I am now making these same improvements to the openCL versions. If all goes well, I should have new apps for Windows and AMD cards within a week.

(source)

 

[TennesseeTony]

SiDock has tasks again. Looks to be a limit of 2 per logical (thread) CPU. I think it is COVID research, at the moment.

 

[Skivelitis2]

Van Der Waerden Numbers is showing signs of life. Web page is readable now at least but still can't create account. Server Status page wont load but a click of the Applications tab shows a Windows app created Jan 03 2021.
vdwnumbers.org

 

[Fardringle]

Van Der Waerden Numbers is showing signs of life. Web page is readable now at least but still can't create account. Server Status page wont load but a click of the Applications tab shows a Windows app created Jan 03 2021.
vdwnumbers.org

Not able to attach a client either, although I was able to log in to the site with my very old account. It would be interesting to see a project be revived after many years of inactivity. Especially if they actually keep it active this time.

Unable to handle request

edit: On 1/18/2021 I am now able to attach a computer to the project. It still is not fully up and running, but does seem to be making progress.

 

[waffleironhead]

Not sure if this should have a thread of its own or not, but anyway.
There is an opencl test App for the m queens project on yoyo.

"16 December 2020 GPU version for M Queens
We have now a first GPU version for the M Queens project for Windows. It requires an OpenCL 2.0 GPU. The version can be downloaded from github.com. The zip contains an installation instruction, the app and the app_info.xml. "

Release First release of OpenCL based m-queens solver · sudden6/m-queens · GitHub

looks pretty speedy from reading forums,

 

[Fardringle]

Private GFN Server is getting a GPU application:

New project - Search for small GFN primes

New GPU project, GFN-1x Small Primes, will be started January 21 2021 at 09:00 UTC

Please see PrimeGrid forum thread for full announcement. 19 Jan 2021, 12:30:52 UTC

edit: I haven't done a lot of testing yet, but with about a dozen GPU tasks completed on a Quadro RTX 3000 (roughly equivalent to a GTX 1080ti), the tasks averaged 995 seconds run time and 648 credits. So about 0.68 credits per second.

That same GPU has been averaging 541 seconds and 6187 credits per task on Moo! Wrapper. That's about 11.42 credits per second. Only just slightly better than the Private GFN tasks.

Looks like I won't be running much of the GPU application on Private GFN...

 

[Fardringle]

Not able to attach a client either, although I was able to log in to the site with my very old account. It would be interesting to see a project be revived after many years of inactivity. Especially if they actually keep it active this time.

Unable to handle request

edit: On 1/18/2021 I am now able to attach a computer to the project. It still is not fully up and running, but does seem to be making progress.

There are tasks available for Van Der Waerden Numbers now, but I suspect they are just for testing and not doing any actual meaningful work yet, since they only take 2-3 seconds to complete. They are rewarding credits, though...

12 hours later...well it was working for a while, but the server is completely offline at the moment. Looks like it might have crashed since the web page isn't even responding at the moment.

 

[StefanR5R]

TN-Grid
I believe last year's news about AMD's COVID-19 HPC Fund initiative was brought up somewhere in this forum, but I can't find any mention of it. One of the institutions which got a grant is University of Trento, Italy, who host the TN-Grid BOINC project.

The grant will give us access, for one year, to two AMD computational nodes, each one equipped with a AMD Epyc "Rome" and 8 AMD Radeon Instinct™ MI50 Accelerators.
This will not solve our well known server problem (it would be too complicated to transfer the server to the new machine and come back to the orginal one after one year) but it will give us a tremendous boost to our in-house computational power, also giving us the opportunity to explore Deep Learning algorithms.

AMD's own announcement sounded to me as if access to these resources was immediately, but:

AMD will grant us access in about a month.

Anyway, we will use them to try to develop a slightly modified version of our algorithm, suitable for gpu-style parallelization. Also we will use their specific deep learning power to help us better understand the whole thing (gene expansion, covid, causality etc).

However, like so many other plans and ventures in 2020/ 2021, this thing is facing a delay:

The granting of the aforementioned resources has been delayed, don't know the reasons, I guess that they were actually building up a new computing center from scratch and had some problems. Anyway, they told us to wait until the beginning of January.

(source)

I apologize for posting a "no changes really" post in the "recent changes" thread.

 

[Fardringle]

After a rather long "vacation", Asteroids@Home is back online and has new work available.

Edit: It was only a fairly small batch of work for testing. The admins posted that they hope to have everything up and running properly soon.

 

[StefanR5R]

Those of us who still have Cosmology@home result files: They can be uploaded now.
But it is not possible to report the results yet. (Feeder not running.)

(forum post)

 

[StefanR5R]

PrimeGrid Fermat Divisor Search PPS-DIV:
"Update. All WUs are sent to the hosts."
(forum post)

 

[StefanR5R]

Those of us who still have Cosmology@home result files: They can be uploaded now.
But it is not possible to report the results yet. (Feeder not running.)

All of the daemons are back up: server_status.php

 

[StefanR5R]

GPUGrid
There is now emerging support for Ampere based cards (RTX 3000):

I deployed the new app, which now requires cuda 11.2 and hopefully support all the latest cards. Touching the cuda versions is always a nightmare in boinc scheduler so expect problems.

The new application version was released for Linux, but not yet for Windows. (apps.php)

Driver 460.32 or later is required. Also, package libboost1.74 or later needs to be installed, since the application is currently dynamically linked against boost. (forum thread)