Formula-BOINC 2017

Skillz said:
One of my 1080s
Click to expand...
I recalculated your PPD. It's like 2.65 mins per task, not 5 mins. So PPD is about 124K. I think we are in good shape.

My 1080 with two tasks per GPU comes out to about 1.92 mins per task. 55% power limit set i.e. underclocking to run the card cool.

upload_2017-8-23_21-22-10.png
 
Milkyway on a 1080TI with no power limit reduction
1:53 when running 1 task at a time
1:28 average when running 2 tasks at a time. The computer screen becomes laggy though.

Think around 218,000 ppd.

Edit: I did an NSLookup on milkway.cs.rpi.edu and got 128.113.126.23 I'm planning to download a bunch of WU and use windows firewall to block 128.113.126.0 to 123.113.126.255. Then remove the rule tomorrow night. With Gridcoin crunching so many wu, I'm guessing it won't take long to get ours validated.

Edit 2: Looks like I'm not getting anymore than about 75 tasks. No point in bunkering with that limit. Their servers are only giving me the same number of tasks as I report back to them.
 
Last edited:
I had to drop the <cpu_usage> line in app_config.xml down to 0.3 from the default 0.8, since BOINC Manager started only running 1 GPU task. I guessed on the 0.3, maybe something higher would have worked also.

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.3</cpu_usage>
</gpu_versions>
</app>
</app_config>
 
Is there any website that shows the progress of our Team during the Sprint ?

What settings should I use for my rig from sig with GPU. The Xeons run CPU-only with default settings, any advices ?

:)

edit: found it myself, sorry bothering you guys.

There are so many stat pages involved one gets dizzy sorting it all out..LoL
 
Sorry Skillz, I was juggling between screens and had the wrong team on that one. Thanks for pointing that out. I will look again when not at work. (Work blocks the site because for some reason it comes up as a gaming site)
 
Yeah... with BOINC there are so many factors when bouncing between projects because they are all different. The huge advantage is only having to learn one client unlike the old days. There used to be a lot more independent projects with their own clients.
 
MN Scout said:
Edit 2: Looks like I'm not getting anymore than about 75 tasks. No point in bunkering with that limit. Their servers are only giving me the same number of tasks as I report back to them.
Click to expand...

Well it's probably too late now, but the thing about bunkering projects that require a wingman is you'll have a higher chance of the wingman returning his end of the task during the event the sooner you get the task. So the first 75 tasks you grabbed have a much, much higher chance of being validated in the 3 day sprint. Therefore, those 75 if you were to have held them until the event started have a chance of 100% validation rate while task you grab now could be in lingo (waiting for wingman) after the sprint is over.

More to bunkering than just getting a lot of tasks before the event starts.
 
Skillz said:
Well it's probably too late now, but the thing about bunkering projects that require a wingman is you'll have a higher chance of the wingman returning his end of the task during the event the sooner you get the task. So the first 75 tasks you grabbed have a much, much higher chance of being validated in the 3 day sprint. Therefore, those 75 if you were to have held them until the event started have a chance of 100% validation rate while task you grab now could be in lingo (waiting for wingman) after the sprint is over.

More to bunkering than just getting a lot of tasks before the event starts.
Click to expand...

Wise, very wise. Would have been easy enough since I looked up the IP address. Doh. Sorry all.

Of the 105 tasks I did early this morning, 15% are waiting for a wingman.
 
Most all my tasks are done and I'm just waiting to unload them. Those GPU tasks go by fast. What's the IP? Are you sure it works to block incoming/outgoing connections? I might run a rule in my router so those rigs can do something else for the time being.
 
Skillz said:
Most all my tasks are done and I'm just waiting to unload them. Those GPU tasks go by fast. What's the IP? Are you sure it works to block incoming/outgoing connections? I might run a rule in my router so those rigs can do something else for the time being.
Click to expand...

I didn't try the IP in the post 683 I striked out through it since I didn't actually try blocking that range. I saw how fast MilkWay tasks were, and how fast the tasks were validated, that I decided to go back to my PrimeGrid performance testing.
 
MN Scout said:
Milkyway on a 1080TI with no power limit reduction
1:53 when running 1 task at a time
1:28 average when running 2 tasks at a time. The computer screen becomes laggy though.

Think around 218,000 ppd.

Edit: I did an NSLookup on milkway.cs.rpi.edu and got 128.113.126.23 I'm planning to download a bunch of WU and use windows firewall to block 128.113.126.0 to 123.113.126.255. Then remove the rule tomorrow night. With Gridcoin crunching so many wu, I'm guessing it won't take long to get ours validated.

Edit 2: Looks like I'm not getting anymore than about 75 tasks. No point in bunkering with that limit. Their servers are only giving me the same number of tasks as I report back to them.
Click to expand...

I can get the 218k ppd higher by upping the tasks just a little bit more.

Average task length continued:
3 tasks:
1:20 average - 245,000 ppd
4 tasks:
1:20 average - same time. I tried dropping my CPU allocation down to 80%, so the GPU instances would have a bit more play and not fight with the WCG tasks for cycles, BUT it had no benefit. Still at 1:20 average.

I'm going to stick with 3 tasks. I had to lower the cpu_usage value down from 0.5 to 0.2 to get my Boinc setup to actually run the 3 tasks. I don't understand why, but it appears to be working.

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.33</gpu_usage>
<cpu_usage>0.2</cpu_usage>
</gpu_versions>
</app>
</app_config>


And it looks like the team has their A game going. Way higher production than the next team in our league.

I'm going to keep playing around on my end with things.

Edit: Looking at the top computers on Milky Way at home, certainly shows those old AMD cards are a BEAST at this program. 60 seconds a task.
 
Last edited:
can you explain this ^ ?

I have no idea what you say but guess I should understand to contribute as much as possible ;)

Thank you !
 
BitMaster, you can create a text file in Notepad and name it app_config.xml. By using some of the parameters above or slightly modified based on cards ability, you can actually run more than one work unit on each GPU in your box. The reason for this is that the application may not utilize your cards to 100%. By running more than one at a time, you can push the card to that 100% or near limit. By doing this, you are putting out more work units over a given time and thus increasing your PPD (points per day). So, MN Scout's app_config that he posted is for running 3 work units per card. The one I posted was for 2. Essentially all you have to do is change gpu_usage and cpu_usage lines for adjusting how many tasks to run. I would start with the default and see what your GPU utilization is first. If it is near 100%, don't change anything. If it is less, than you can make an app_config like above. I would move to 2, then 3, etc.. I have heard people with Titans running up to 10 work units on each card....(I had misread previously and thought it was 20 per card but it was 10 per with 2 cards)

In Windows that app_config.xml goes into the c:\Windows\Program Data\BOINC\projects\ <insert project here> folder
 
Last edited:
Gilthanis said:
BitMaster, you can create a text file in Notepad and name it app_config.xml. By using some of the parameters above or slightly modified based on cards ability, you can actually run more than one work unit on each GPU in your box. The reason for this is that the application may not utilize your cards to 100%. By running more than one at a time, you can push the card to that 100% or near limit. By doing this, you are putting out more work units over a given time and thus increasing your PPD (points per day). So, MN Scout's app_config that he posted is for running 3 work units per card. The one I posted was for 2. Essentially all you have to do is change gpu_usage and cpu_usage lines for adjusting how many tasks to run. I would start with the default and see what your GPU utilization is first. If it is near 100%, don't change anything. If it is less, than you can make an app_config like above. I would move to 2, then 3, etc.. I have heard people with Titans running up to 10 work units on each card....(I had misread previously and thought it was 20 per card but it was 10 per with 2 cards)

In Windows that app_config.xml goes into the c:\Windows\Program Data\BOINC\projects\ <insert project here> folder
Click to expand...

Thank you for the nice detailed explanation.

The 980 is at 100% with 1 task only, guess I don't need to change a thing for now. Still, good to know !
 
I've never had a video card manufacturer give me grief about it. However, this has been a concern for DC'ers for years. The commenter in that thread saying the "stealthily" added verbiage isn't completely accurate. Some have had that statement for a while only adding the mining portion more recently. nVidia wants people to buy the Quadro line for what we do. They just know that they sell more cards by allowing us to do this. Hell, look at EVGA. They encourage DC'ing.
 
BitMaster said:
can you explain this ^ ?

I have no idea what you say but guess I should understand to contribute as much as possible ;)

Thank you !
Click to expand...

Maybe a picture of the GPU usage will help show why I loaded up more than one task on my card. I'm trying to remove the dips and be at or near 100% load, since that is wasted GPU time. The caveat of running more than one gpu task is that user tasks like typing may become not smooth anymore, but the tradeoff may be a lot more points.

I think the rest of the team did a good job describing how (create app_config.xml file in the project folder, then open the BOINC Manager and under the Option menu choose "Read Config Files", and it will instantly load the new setup).

Here is the grand GPU Usage picture:
upload_2017-8-25_12-20-11.png
 
Just keep in mind that closer to the end of the challenge, you may need to remove the app_config to squeeze the last few work units in before deadline. Since we are talking minutes per work unit, probably won't be worth the hassle, but some care about that.
 
What are people seeing for validation rates? Here: http://milkyway.cs.rpi.edu/milkyway/home.php then click on "Tasks - VIEW". 465 Valid and 125 Validation Inconclusive AKA waiting on the wingman to return the same result. So roughly 4 out of 5 have been counted into the Sprint scores. Pretty good ratio.
 
My 980 ( downclocked to 1138MHz for computing )

...hmm..wont UL my jpg..bahh...

OK, no jpg upload for whatever reason, what I wanna actually say: my zigzag only has minimal drops every so many seconds...will check if its worth loading more tasks onto the GPU.
 
about 1/4 waiting, ca. 1200 total, 113 are marked without result (?, german translation ). This morning I had 0 waiting, its up&down I guess
 
What is that grey cable? Is it necessary? That is blocking a PCIe slot is it not? You could use that slot possibly. I certainly wouldn't use that PSU to power another 1080Ti though. Probably already pushing to its limits.
 
Grey cable is USB. The card would be sitting on the PSU fan. I think I once ran it that way a long time ago. Likely not the best setup for 100% load with the intention of keeping the GPU cool, not to mention the PSU.

Not sure why the sleeving is melted. The computer is at 100% full load right now and it is not warm.
 
That doesn't mean the cables aren't getting hot. And yes that would make it hard for the GPU to breathe. Not saying that I haven't done it multiple times... but for a card that expensive... I would invest in a bigger PSU and possibly a bigger case.
 
The US Army killed us in Rosetta and looks like the Christians are not praying for us but going to kick our butts ...
 
I enabled Rosetta on my computer.

I'm going to try and decide on some resource share numbers for all the CPU applications I have running. Kind of set it and forget it for a few weeks and see what happens. I have one computer, but maybe I can help in a small way to a handful of projects.
 
IMO, just let the BOINC client sort it out. The only thing I use the resource share feature to manually manipulate what has priority is setting a share of 0 for projects I don't necessarily want running, this way if the projects I want running run out of work those 0 resource share projects will fill in the gap.

Otherwise, the BOINC client does a decent job at load balancing multiple projects.
 
I have all the projects at resource share point of 1, but the GPU projects at 10. I was thinking of upping the CPU projects that we have competition in and are at risk of losing Formula Boinc points in a resource share point or two.
 
My MilkyWay errors on my GPU now, had 66 computational errors..hmmm ??? WTF ???

I updated the driver but I think it did not fail right after the newest NV driver 385.41 but cannot exclude the driver to blame.

Any ideas ?

Went back to Seti for now

*edit: found the culprit: NVpanel Optimize_for_Computing on ON = error; Off ( default ) works.

My bad :(
 
Last edited:
Don't be surprised to see lower point payout at Rosetta. It is a known low scoring project.
 
OK, subscribed to Rosetta just now as well...for the future when it may be on a sprint etc.. . Also added Asteroids. Whatever works best and makes some sense.
 
You must log in or register to reply here.
Back
Top