Milkyway@home

Gilthanis

[H]ard|DCer of the Year - 2014
Joined
Jan 29, 2006
Messages
8,718
I am adding this app_config info in the first post so that any whom come here will have easy access. This allows for 2 tasks to run on your GPU at a time. You can change the numbers to run more, but it is up to you to decide what maximizes your points on your hardware.:


<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.3</cpu_usage>
</gpu_versions>
</app>
<app>
<name>milkyway_separation__modified_fit</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.3</cpu_usage>
</gpu_versions>
</app>
</app_config>

Edit: I created an account called HardOCPtest so that anyone whom wanted to test or contribute anonymously could do so.

to attach a computer to your account without using the BOINC Manager.
To do so, install BOINC, create a file named account_milkyway.cs.rpi.edu_milkyway.xml in the BOINC data directory, and set its contents to:
<account>
<master_url>http://milkyway.cs.rpi.edu/milkyway/</master_url>
<authenticator>1027956_e04658bf8505999426067a2b005c2f74</authenticator>
</account>

teamdaily.php
 
Last edited:
Wait - I'm confused - so if I am running Milkyway on my 7970s, should I be using these? Right now I only used what BOINC downloaded.
 
I believe so EXT64. I have not ran them yet since I just found these...but they should run faster than stock apps.
 
Hmm, well I'll give them a try this afternoon. The MW Separation file is an older version number than what I have (1.30 vs 1) and the other has the exact same file name. I'll give it a go, but I wonder if these optimizations were already incorporated into the general release? A quick googling didn't turn up anything.
 
Well...the only other suggestion I would have is to ask in Milkyway's forums.
 
I tried the one that had a current version number (not the one that appeared outdated). Performance was identical between 'optimized' and normal. 27 seconds per WU (this was the shorter of the two standard sub-projects). So, it appears to me that these optimizations were already rolled into the general release.
 
I tried the one that had a current version number (not the one that appeared outdated). Performance was identical between 'optimized' and normal. 27 seconds per WU (this was the shorter of the two standard sub-projects). So, it appears to me that these optimizations were already rolled into the general release.

Thanks for taking the time to test them EXT64. I'm sure this will help others.
 
Thanks, EXT64. You saved me the trouble of trying these out myself.
 
Just wonder: any security concerns with non-stock apps ? Would MW project have some safety net before release stock apps ?
 
ChristianVirtual, are you meaning a way to tell if the non-stock apps are producing valid results? Because that is usually what the validator and having a quorum is for. Otherwise, I think security with a non-official release may always have security concerns.
 
I got this in an email today.

Dear MilkyWay@home Volunteer,

We are so close!

Last July I sent out a plea to MilkyWay@home volunteers to help save our project after our National Science Foundation grant was not renewed. I am happy to report that we have so far raised $37,991 towards a goal of $40,480!

If you are willing to donate but have not yet done so, then now is your chance. It is only with the help of volunteers, family, and friends like you that we will raise the remaining $2,489 needed to reach our goal.
Donate Now
Don't have a credit card? Click the 'donation options' button on the right to explore other contribution methods.

We have great plans to discover where dark matter is in the Milky Way over the next ten years. That goal requires significant advances in our computer algorithms that run n-body simulations and that requires the efforts of many students to achieve. Any funds received in excess of our goal this year will be used on additional student salaries. Since most of the funds will be expended on student salaries in summer 2015, we will post a report of the funding used and results achieved in September, 2015.

Many thanks to the wonderful MilkyWay@home volunteers, my family, and the parents of my former graduate students for keeping us afloat thus far. Donations have come in by credit card, check, PayPal, donor advised funds, and we received over $3000 from BitCoinUtopia, which is collecting cryptocurrencies to support our research.

The MilkyWay@home team and I thank you so much for keeping our research alive!
In appreciation,
Prof. Heidi Jo Newberg and the MilkyWay@home team
 
http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=3653#62669

Dear MilkyWay@home supporters,

In the past day, we received $4432.50 in donations, which puts us over our goal to be able to run MilkyWay@home through summer 2015!!! This is a tremendous relief to the development team, and gives us time to find ways to operate into the uncertain future. We are extremely thankful to our generous donors. We will be working harder than ever to produce new and cool science results, develop new algorithms for MilkyWay@home, and keep the operations running smoothly.

Thanks again,
Heidi Newberg and the MilkyWay@home team

They are now over the amount they were asking to raise. Kudos to everyone whom donated whether directly or through BitcoinUtopia.
 
I did a 24 hour test running 1 task of MW@h on my 7970 and then 24 hours running 2 tasks.

Running 1 task I got 270200 points. I was running at 975 mhz most of the day. It got hung up for about 20 minutes so I turned it down. I had the fan at 50% to keep it around 80*

Running 2 tasks, I got 302300 points BUT I was clocked down to 925 mhz, which was the lowest I could go. I needed to keep the fan screaming at 60% to keep temps in the low 80s as well. Also, occasionally it would 'kick out' and only run 1 task. I tried to kill one off the tasks halfway to keep it offset. I ordered an arctic accelero 280x gpu cooler, so that should make it practical.

Right now I am trying einstein@home. It was designed where I got my B.S. degree (UWM) so I thought I would run a few points with them.
 
You will see much lower scoring with Einstein work units. But certainly a solid project.
 
Yeah, I support the project - some good dudes out of the physics department there. Weird that it only runs my gpu at 70%, but at least it is not jumping up and down since the work units are long. How does MW@h compare to the other projects. Getting another card this week, so I will probably fold with both projects.
 
Well...some of the projects deliberately set the work up to not push the card all out on a single work unit. They do this so that it is still usable if you let it run while using the computer. Obviously not all cards can do this...but they don't really have the lower end cards in mind.

As far as MW@home comparing to other projects, what exactly are you comparing? MW uses double precision where as only one other project I know of does that. PrimeGrid's GFN sub project uses double precision. Everything else uses single precision.
 
Thanks for the info. I will stick with the double precision project - it's where the 7970 makes its money!
 
Yeah... If you are wanting points, then go with PrimeGrid's GFN world record sub project. Each work unit is worth about 500k a piece. But those work units are monsters, so don't expect finishing one in seconds or minutes.....
 
A 7970/280X can get ~1 million points per day at Collatz, when running the Large WU's. Just throwing it out there as another option...
 
Thanks for the tip guys. I'm really not liking the ultrashort work units of mw@h. Even when running app_config, it still runs eratic, like how it will occasionally pop out of 2 interations. Also, app_config really screws up when I try using it with cc_config. I'm sorry, I realize I am the new guy, but WHY THE FOK is the user still forced to tinker with this kind of stuff in 2015?? This stuff may seem like childs play in these forums, but it is a huge turnoff for the average person. F@H is SO much more streamlined in their interface. There, you can make tweaks w/o having to type up a computer code. I saw that Einstein@home has made it easier, but this sort of thing should be at ALL the projects.
 
That alone doesn't do anything. That is just the app name of one of the apps Milkyway uses.

That looks like the first line of an app_config.xml file to me. Depending on what is below that app name will dictate what that file does exactly. Typically, app_config.xml is used to tell projects the number of WU's you want to run on each GPU and how many CPU cores should be reserved per GPU.
 
Well...first... FAH is a single project with a dedicated client to it. BOINC is not a project. It is a client designed to be used by many. So streamlining a client used for unknown possibilities leads you to a client that can either be highly customized or extremely limiting. BOINC does not NEED cc_config or app_config in general. Those are for people wanting much more control over their systems. FAH is extremely limited in my opinion and though it may seem streamlined in some areas, it was still pretty archaic in others too. BOINC also has come a long way. For example, does FAH support ASICs? How about FPGA's? How about MIPS? ARM? Not yet? Why not? It doesn't even take advantage of Intel GPU's from what I last saw. By the way, does FAH let you choose a specific app to run only without many tweaks or overhead? BOINC projects typically do with little headache.

Now to answer why the user has to tinker with this stuff is because you are talking about ADVANCED users looking to maximize and squeeze out every bit of production they can from more than just the average user's computer. We haven't even begun to get you into an app_info.xml. Yes there is a third type and is becoming less common. And for good reason. IIRC, even the old FAH client needed some know how to maximize it and wasn't a simple install this and you are done setup. I'm not sure what you mean about Einstein being easier, but I'm sure if you describe what you mean, I can explain why or give more examples...

This <name>milkyway_separation__modified_fit</name> basically is just an identifier of the app's name. Milkyway has more than one application.
 
Now if you really want to compare outdated clients, take a look at GIMPS or distributed.net. Prime95 and Majestic12's clients even feel like late 90's style apps these days. To be fair, BOINC is one of the most advanced and project supportive clients out there. I don't think there is any that even comes close to it's versitility.
 
Yeah, its starting to work well for me now. Disabling the iGPU helped, I will be upgrading to an 860k anyway. Also, using .33 gpu instead of .5 gpu works great since it seems to keep the work offset.
 
From MilkyWay@home's Facebook Page:
Bad news Not only has MilkyWay@home lost its funding, but now Newberg's other project, a collaboration with the Chinese LAMOST project, has also lost its funding (though it had the highest possible reviewer ratings). The NASA program that could have funded MilkyWay@home was canceled this year due to lack of funding. The purple part of the graph shows the projections for NSF funding for individual astronomers as a function of time. What it does not show is that NASA has also reduced science funding, research costs have gone up, and that the fraction of proposals funded has dropped fairly steadily from 50% in 1990 to 13% in 2014/15. We will be running another funding campaign this year. Look for more information in late summer.

https://www.facebook.com/Milkywayat...41839.589753444464934/806376566135953/?type=1

Now, this doesn't mean they will shut down. However, they may have to do like last year and ask for more donations to keep them afloat. BitcoinUtopia contributed a good chunk of change, but I'm not asking anyone to put resources there as it tends to be a very touchie subject. (If you would like to run BU anonymously, you can attach using the weak account key for the HardOCPtest account found in the first post of the BU thread http://hardforum.com/showthread.php?p=1041104829#post1041104829 I will try and remember later this weekend or beginning of next week to setup different profiles for CPU, GPU, and ASIC hardware ) However, if you like the project, please consider making a direct donation.

http://milkyway.cs.rpi.edu/milkyway/donate_index.php

If you need help with a profile change, feel free to contact me via PM. That is the best I can probably do at this point.
 
Last edited:
Sad to hear this, had a rig working for milkyway for a long time. Hopefully they will figure something out to save the project.
 
I'm sure I will catch hell from my teammates about this, but I have reattached to Bitcoin Utopia and am contributing solely to the Milkyway project there.

I am only doing this for a very short time; I need 6 million points to reach 50 million in BU (another F-DC Mega Milestone) and then I will detach my machines from BU again. At that point, I will just contribute money directly to Milkyway.
 
I'm sure I will catch hell from my teammates about this, but I have reattached to Bitcoin Utopia and am contributing solely to the Milkyway project there.

I am only doing this for a very short time; I need 6 million points to reach 50 million in BU (another F-DC Mega Milestone) and then I will detach my machines from BU again. At that point, I will just contribute money directly to Milkyway.

That didn't take long... Hit my 50 million goal overnight and removed BU from all of my machines this morning. I probably made about $.01 for Milkyway. :rolleyes:

Running just a few of my GPU's (the AMD's) at BU got me ~6 million points in < 12 hours. No matter which way you slice it, that is fairly ridiculous and why most people do not like BU.

I can't imagine ever running BU again. However, my OCD just couldn't stand that I was sitting so close to 50 million points... It was a moral imperative that I cross that hurdle. :D
 
front page news

Nbody Status Update
Hey All,

Just a quick update on the status of Nbody. I discovered that the hanging that people were experiencing were due to the Newton Raphson root finding algorithm which would get stuck in an infinite loop. After looking for why this occurred for sometime, I decided to switch to a bisection method which is insured not to get stuck.

The slow nature of the initialization of the plummer was found to be an error in the calculation of one of the limits of integration in a crucial integral that was being calculated. What a time it was finding that! With that limit fixed, it seems that the initialization takes a few seconds!

I am currently unit testing the code and hopefully should get a lovely new version in a couple of weeks.

Cheers,
Sidd 25 Jun 2015, 16:28:41 UTC · Comment
 
Front page news from the 2nd.

2015 Fundraiser for Milky Way Research
Hello Everyone,

Our fundraising campaign kicks off today. If you are willing and able to financially support Milky Way research (I know many of you already support us with your computations), we would really appreciate it! Click the link to find a letter, video, budget, and a description of our thank-you gifts. The goal seems daunting, but donations of $5, $10, or $20 really do add up and make a difference. If you can afford $100, then we will send you a very nice astronomer-designed T-shirt as a token of our appreciation.

For more information take a look at our fundraiser page here.
 
After the pent, decided to run my only 7970 (crimson driver 17.7.2) on MW. Card runs with core freq 975MHz at 1.04V, 650MHz memory. No other CPU task running on this windows 7 rig (2695v2). Card temp is hovering around 60°C. Pulling ~230W off the wall.

Running 4 tasks simultaneously on one card across 2 boinc manager instances instead of one to fill up enough cache in case server not responding.

There are two tasks with different credits. One with credit of 227.5x (x = 1,2 or 3) and the other 244.01. Average run time per task shown below:

upload_2019-5-27_13-20-50.png




upload_2019-5-27_13-26-5.png


Other MW benchmarking result can be found here and here.

Use this neat tool to calculate statistics up to 10 pages.
 
true dat! he just demolished everyone! I couldn't even think of a way to bunker against him:banghead:
 
Back
Top