High payout projects

Gilthanis

Gilthanis

[H]ard|DCer of the Year - 2014
Joined
Jan 29, 2006
Messages
8,731
I used to keep this via a PM, but should better suffice here.

Today I am noticing Citizen Science Grid paying out rather high. My borgs have been running the current EXACT MNIST Batch Norm CNN Trainer v0.30 work units and seem to be earning around .219 points per second per thread. It may be a fluke, but figured I would give the heads up to anyone else wanting to push our position or just wants to monopolize on the scoring opportunity.

For those not seeing the significance, that would equate to 18,921.6 PPD/thread
 
Amicable Numbers still seems to be ~.022 to ~.028 PPS/thread (CPU)
POGS is similar in points with ~.022 PPS/thread 1440 PPD on x86 and .005 PPS/thread 432 PPD ARM Theoretical
 
In the applications list:

EXACT MNIST Batch Norm CNN Trainer?

After adding CSG through BAM! I get: This project is using an old URL. When convenient, remove the project, then add https://csgrid.org/csg
 
Last edited:
Gilthanis said:
Today I am noticing Citizen Science Grid paying out rather high
Click to expand...
I started this project two days ago. Phoenicis appears to be on board yesterday.

I run limited tasks since I'm focusing on Denis@home (which yields horrible PPD). I'm trying to put our team in a respectable FB position and also a personal goal since I haven't run this project before.
 
Yeah... most Bio/Medical projects suck for points. GPUGrid is one of the few that doesn't. Also, keep in mind that if the project uses the Credit New system, there is a reason people call it Credit Random. It is supposed to even out over time but every time there is a slight change, credit goes ape shit all over again. Starts a bit high and eventually goes to shit. I did a lot of numbers with it and multi threaded apps for YAFU a few years back. It never leveled out. Every day it changed. And it seemed to score best with quad core setups. But that was just at YAFU and again, a couple years ago. I'm sure things have changed there since then.
 
I'm finishing up last few Dennis tasks and moving to CSG. Hope to rack up some points in CSG before they revert to low credit scoring (if it does).
 
Is there a master list of other BOINC projects that [H] has a team for? I didn't see any in the guides - perhaps I'm missing. I would want to associate all of my work with the team [H] as much as possible.

Thanks :)

Nevermind - I found the sticky - Adding in all projects that I can for now.
 
Last edited:
Glad you found my lists. It is a lot of info and there are still a few projects out there with rogue H teams, but we have it mostly lined out. If only we could get the WCG one fixed.... grrrrr
 
Gilthanis said:
Glad you found my lists. It is a lot of info and there are still a few projects out there with rogue H teams, but we have it mostly lined out. If only we could get the WCG one fixed.... grrrrr
Click to expand...
Thank you though - I'm sure that was boring as hell to create, seems tedious, and I've got some issues with some projects not associating with the [H] team even if the projects' page / account shows that I am. I'm assuming it's an issue with syncing or with the BAM. I'll look into this once I get back home after work.

Is the BOINCStats site the go-to site for stats or is there another site that's better?

On topic - The highest PPD still appears to be the GPU based projects and it looks like it would be Collatz still?
 
Yes... GPU's always outscore CPU's in BOINC projects. And Collatz is still the highest payout. I prefer BOINCStats generally but www.free-dc.org is pretty good too. However, it doesn't have an account manager.
 
I prefer free-dc for stats, but BOINCStats for the manager.
 
Welcome back Jon855!

If you do decide to go for Collatz, there's a small band of us already trying to build on our FB points there in-between sprints/challenges.
 
And make sure you install both OpenCL drivers and the recommend c++ listed on the home page of the project
 
Credits for the old application (search up to 2^64) doubled
It's sad to see everyone leave the old (and unfinished) application "Amicable Numbers up to 2^64" so I decided to double credit points for it.

Everyone can switch their CPUs back to gain some more credits.

Note that you can run "Amicable Numbers up to 2^64" on CPU and "Amicable Numbers up to 10^20" on GPUon the same PC at the same time!

Here's what you need to set in project preferences:
Use CPU, AMD/ATI GPU, NVIDIA GPU = yes
Run application 2^64 = yes
Run application 10^20 = no
Accept work from other applications = yes
settings_2_64_and_10_20.png

With these settings your CPU will get tasks (and more credits in general) from "2^64" only, and your GPU will get tasks from "10^20" thanks to "Accept work from other applications = yes" selected.
3 Jul 2017, 16:09:14 UTC · Discuss
 
I just stumbled across a thread at CSG pointing out excessive credits - https://csgrid.org/csg/forum_thread.php?id=2435#7127
however, nothing has been done about it yet. I still say it is crediting 6+ times the normal project rate. Anyone point whoring should allocate spare CPU cycles to the Wildlife sub project.

I still say that GCC over pays for its NCI monkeys applicaiton. Running multiple clients generates absurd points per day. Though we have found Windows x64 to only truly allow for up to 250 clients without major failure rates, you can run multiple VM's on the same host to run even more. I have one server that has 250 clients on the host OS and now has two VM's running 250 clients each for a total 750 clients on one machine. This server is doing other stuff and isn't a high end server so it isn't maximized for this project. However, if it were we would be talking 4 apps/client (10points each app) * 750 clients = 30,000 PPH. 30k * 24 = 720k PPD. So, clearly my server has some issues that I need to find the bottle neck on because I am not hitting 720k PPD with everything total yet. I do know the clients run much better on the host OS and also the more cores the better. A single core managing 250 clients will work but really isn't optimal. I'm also running off of spinner HDD's. One is running straight off of a RAM drive. I've been tempted to up the CPU's in the VM's to 2 each to see if output would improve much.
 
I think PrimeGrid could become high payout with selectively choosing projects 500K PPD plus at up to 8pps. Here is a Google Drive link to a spreadsheet I"m working on: Prime Grid Projects - PPM

To me it looks like the PrimeGrid application to run is PPS (sieve) for Nvidia GPUs. I'm not sure if this payout translates over to AMD cards or not. The Genefer applications payout well also, but higher the GFN-## higher they tax the GPUs, longer the task, and higher chance for an invalid result. I expect the Genefer 22 task to take 30ish hours for a payout of around 850k, which still doesn't reach the PPD for the PPS sieve tasks.

I'm crunching a GFN-21 task right now, and have a GFN-22 task for tonight and through tomorrow.

Here is an image of its current form as of 8/16:
upload_2017-8-16_12-50-29.png


On 8/18 I'm going to switch over to on PPS Sieve tasks and see what PPD I reach in actuality.
 
The larger GFN (formerly WR work units) definitely credit more. That is because PrimeGrid gives a bonus for work from harder sub projects or sub projects they feel at the time are priority. So, there is a tiered bonus on their site for work. This bonus can change at any moment. PrimeGrid.JPG

GFN is double Precision so you will find AMD cards out performing nVidia because of that.

PrimeGrid is a high paying project for GPU's. Collatz way outscores it. GPUGrid can score pretty high too if you return your work within 24-48 hours. As you can get up to a 50% bonus on those work units as well.
 
Yep.

I guess it is interesting that the PPS (sieve) tasks outscores them all in PPD on my card and that tasks doesn't talk about bonuses.

With so many of us having Nvidia cards and faster GPUs that we could do some serious damage if we pick our tasks appropriately in PrimeGrid.


I apologize for not focusing on Seti tasks with my card at the moment. I figured out that the Windows client has extremely unoptimized GPU tasks for Seti, and the Lunatics build is the same way. There is some unoptimized "bug" lets call it that makes the task take 100% longer than needed. There is a linux build that gets times in the 1 to 3 minute range, where the Windows builds get in the 6.5 to 10 minute range. Too much of a waste in power and time. I think I can get a linux boot stick made up on Tuesday night.
 
The problem is that the team in general typically prefer to focus on Bio/Medical research. With regards to GPU's, that limits you to FAH and GPUGrid. FAH seems to have an endless supply and better funding. GPUGrid isn't Pande Group but has a set of headaches to add to the learning curve. Out team is split between FAH loyalists and FAH-scorned. GPUGrid has a lot of headaches users don't wish to tolerate in a project and so most tend to leave their systems idle most of the time. Sad affairs. I personally just run GPU's until a project hits 1 million points. Then I focus them on challenges and badge upgrades. Typically my boxes with cards are set to pull all badge projects and let the client choose which to work on from there. I'm not as much of a point whore as most, so I really am not all that concerned. ;)
 
I can understand the desire to run Bio/Medical projects. I'm just having fun seeing what a fast card can do. Wanted to pass my findings a long, incase anyone else was interested in the possibility of taking first in a project.
 
Citizen Science Grid is the project. Just like WCG it has sub projects that you can choose from. However, right now only Wildlife sub project has work. So, not changes need made other than attaching to the Citizen Science Grid project.
 
Projects = Application ?

DNA@Home Gibbs Sampler: no
Convolutional Neural Network with Seam Carving: yes
EXACT MNIST Batch CNN Trainer 2.0: yes
EXACT MNIST Batch Norm CNN Trainer: yes
EXACT Batch Norm With Pooling CNN Trainer: yes

Those are the only choice I see; but might look on the wrong location

Update: found it in the writing of the project ... have a 36-Core EC2 instance on it.
 
Last edited:
Only got 472k a day for the Genefer 22 task in Prime Grid. A lot less than I thought it would be. Maybe due to running at 70% power target. Not sure I want to bother running another one of those tasks at 100% power target, since it would still take a day and still get less points than the PPS Sieve tasks. I can still use the computer without hiccups with the PPS Sieve tasks, not so with the Genefer GFN-22 tasks.
 
It's been a while since posting so probably should get an update.

CSG still waaaaay over credits. However, Travis (the admin) is talking about making some adjustments that will lower them in the near future. Anyone wanting those easy points now should probably get them while they can.
Rakesearch has an optimized application you can download and use based on what extensions your rigs have. Thus people are seeing much higher payout of points while running them. I recommend people monopolize the opportunity while they can before the project adopts them as their standard app and realize the points are too high and need adjusted.
GoofyxGrid-NCI project is still encouraging the use of multi clients. You can crank out like 250 clients per host and ring up a lot of points for very little processor usage. VM's are needed to maximize most hosts capabilities but the VM's are less efficient that bare metal operations.
 
GoofyxGrid-CPU project is currently crediting me between .09 and .115 Points per second. The work units seem mostly short too. Would be a good project to set up as a backup project when running a primary project that runs out of work quick. The reason it would be a good backup is because 1. high points and 2. short work units letting the BOINC client check in with your primary project often.
 
Just joined GoofyxGrid-CPU. I supposed the project mission is the same as the NCI version? Can't seem to find more details about the project mission.
 
No...they have a test application for subset sums right now. I'm guessing it is going to be a cpu intensive umbrella project.
 
GoofyxGrid-CPU seems to give around 450K PPD on my 2683. Not bad considering this is a CPU project. I'll reach my first 1M over the next day or so.
 
Yeah, in the forums they are talking about wanting to keep work units worth 10 points per minute. They are reporting that work units are slowly taking longer and longer to process, so they may keep adjusting the points per work unit over time They are still watching it as work unit run time has recently gone up by ~12%.
 
Passed 1M mark. :D

Pity no badges yet but will come soon according to project admin. Will see.
 
Also, for those interested... goofy's NCI project has had major problems the last few days. He plans on getting things back to normal tomorrow. However, he plans on bumping the credit per work unit from 10 points to 25 as a reward for people not being able to turn in work, get work, etc... http://nci.goofyxgridathome.net/forum_thread.php?id=284#2496
If you were running multiple clients before per my how-to, you will see a drastic increase in production when this batch comes back online. That is a 150% bonus to an already exceptionally high crediting work unit anyways.
 
I believe that bonus was for one day only. However, I earned ~1 million points yesterday just from NCI work.
 
I haven't added it to the all inclusive DC list yet but the Team Captains have been testing a new project. I can tell you now that it currently grossly over pays in credit. It is a CPU project.

http://dhep.ga/boinc/
 
Some food for thought. These work units don't have an official deadline. They work different than every other project I've been part of. They are using the trickle credit system. The longer the work units are up and running the better for the project allegedly. The work units require Java.
 
You must log in or register to reply here.
Back
Top