Milkyway@home TeAm competition - 10/15 to 11/15

[Fardringle]

@fardingle, 4x should be fine for MW. I don't think it comes close to saturating a 8x or 16x pipe line. You will definitely pull that 1060 off MW when you see just how badly it does. lol

Normally I wouldn't put it on MW at all. I was only considering it for the team challenge/race.

In the short time they have both been running together, the GTX 1060 has completed 8 tasks in an average of 294 seconds per task. The 280X has completed 53 tasks in an average of 46 seconds per task.

Only running one task at a time on each GPU right now. Not sure I want to mess with multiple simultaneous tasks just yet...

Edit: After running overnight, it looks like the 280X is aiming for about 340K per day and the 1060 is getting about 61K per day. That 61K is quite terrible! My super ancient Quadro 4000 in my old Xeon machine has been getting close to 80K per day on MW... The GTX 1060 is definitely going back to SRBase!

 

[Assimilator1]

Icecold
Lol, Milkyway guy, a variation of the Milkybar kid? , mmm I loved Milky bars!

Err, anyway, I thought it's about time I ran WUs in parallel, can someone remind me how?

 

[Icecold]

To run multiple WU's you need to create an app_config.xml (paste the info below into notepad and save as app_config.xml) file in the Milkyway project folder. If running Windows 10 with the default path it will be in C:\ProgramData\BOINC\projects\milkyway.cs.rpi.edu_milkyway This is the contents of my app config running 4 WU's at a time with 0.25 CPU usage, but I think I have seen as low as 0.01 CPU usage entered into these. I do 4 WU's at a time on a Firepro S9150 and 2 at a time on a 7950. @Fardringle if I had to guess you'll want to do 2 WU's at a time, so use the following but just change it to 0.50 GPU usage and see how it does. It should definitely be more PPD than doing 1 at a time if the card is fast enough.

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.25</cpu_usage>
</gpu_versions>
</app>
</app_config>

 

[Icecold]

Icecold
Lol, Milkyway guy, a variation of the Milkybar kid? , mmm I loved Milky bars!

 

[Icecold]

Day 1 Stats:

crashtech​	5,265,507.00​
phoenicis​	3,129,598.00​
Icecold​	2,325,371.00​
biodoc​	1,625,443.00​
emoga​	473,681.00​
waffleironhead​	245,853.00​
Stig​	143,625.00​
Walrus of Apathy​	134,588.00​
flywaldo​	108,251.00​
Fardringle​	79,414.00​
Brent​	61,213.00​
Lane42​	46,620.00​
IcefireQuickSilver​	38,375.00​
elfenix​	6,899.00​
Endgame124​	1,820.00​
Silverthorne​	455.00​
Christopher​	297.00​
Shu Gao​	37.00​

In pie form:




If your numbers seem low, don't get discouraged it may just be tasks waiting to validate. Every bit helps regardless. Great first day TeAm!

 

[Fardringle]

To run multiple WU's you need to create an app_config.xml (paste the info below into notepad and save as app_config.xml) file in the Milkyway project folder. If running Windows 10 with the default path it will be in C:\ProgramData\BOINC\projects\milkyway.cs.rpi.edu_milkyway This is the contents of my app config running 4 WU's at a time with 0.25 CPU usage, but I think I have seen as low as 0.01 CPU usage entered into these. I do 4 WU's at a time on a Firepro S9150 and 2 at a time on a 7950. @Fardringle if I had to guess you'll want to do 2 WU's at a time, so use the following but just change it to 0.50 GPU usage and see how it does. It should definitely be more PPD than doing 1 at a time if the card is fast enough.

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.25</cpu_usage>
</gpu_versions>
</app>
</app_config>

Thanks for the info. I'll probably try that at some point, but not right away. I want to let it run for a day or two with one task at a time so I can have a benchmark for comparison when it is doing two at a time.

In the meantime, you wouldn't happen to have a suggestion to get BOINC to let me move the GTX 1060 in the same machine back to SRBase? I configured my user preferences on MilkyWay to only use AMD/ATI GPUs, and on SRBase to only use Nvidia GPUs, but the client just says "Not requesting tasks: don't need (NVIDIA GPU: not highest priority project)."

There aren't any other GPU projects enabled for the Nvidia card right now, and BOINC doesn't have any MilkyWay tasks left for the Nvidia card...

 

[Icecold]

Thanks for the info. I'll probably try that at some point, but not right away. I want to let it run for a day or two with one task at a time so I can have a benchmark for comparison when it is doing two at a time.

In the meantime, you wouldn't happen to have a suggestion to get BOINC to let me move the GTX 1060 in the same machine back to SRBase? I configured my user preferences on MilkyWay to only use AMD/ATI GPUs, and on SRBase to only use Nvidia GPUs, but the client just says "Not requesting tasks: don't need (NVIDIA GPU: not highest priority project)."

There aren't any other GPU projects enabled for the Nvidia card right now, and BOINC doesn't have any MilkyWay tasks left for the Nvidia card...

I know there are some issues with SRBase and multi GPU. I'm sure there's a way to make this work with an app_config or changing a cc_config or something, but I'm really not sure Hopefully somebody else will chime in.

In the meantime could you try switching the Nvidia one to something like GPUGrid or Primegrid and see if it picks up work? I'm just curious for troubleshooting if it's SRBase specific.

 

[Fardringle]

Maybe the BOINC Manager just had to update a timer or something. Without me doing anything else, it started downloading some SRBase tasks for the Nvidia GPU some time in the last hour or so...

 

[Assimilator1]

To run multiple WU's you need to create an app_config.xml (paste the info below into notepad and save as app_config.xml) file in the Milkyway project folder. If running Windows 10 with the default path it will be in C:\ProgramData\BOINC\projects\milkyway.cs.rpi.edu_milkyway This is the contents of my app config running 4 WU's at a time with 0.25 CPU usage, but I think I have seen as low as 0.01 CPU usage entered into these. I do 4 WU's at a time on a Firepro S9150 and 2 at a time on a 7950. @Fardringle if I had to guess you'll want to do 2 WU's at a time, so use the following but just change it to 0.50 GPU usage and see how it does. It should definitely be more PPD than doing 1 at a time if the card is fast enough.

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.25</cpu_usage>
</gpu_versions>
</app>
</app_config>

Ah, I'm running Rosetta alongside so I don't want take away from that what I don't have to, what could I lower CPU usage to which won't impact MW significantly? (running 2 WUs).

Btw re stats, I wouldn't have been surprised by low stats on my 1st day, but I'm not even listed! I know I've produced some.

 

[Fardringle]

You have two computers on your account that have reported results recently, so I'm not sure why you aren't on Icecold's stats either...

The one with the i7-4930K started getting results validated a couple of hours ago (after the update from Icecold), but the computer with the Ryzen 5 3600X and RX580 had at least some tasks validated last night so those points should show up on the report. Maybe you'll just have extra on tomorrow's update.

For that matter, it looks like the 79K points I do have on the first stats update are pretty much just what was already being done by my old Quadro 4000s, so maybe my new stuff isn't there yet either..?

 

[Icecold]

Ah, I'm running Rosetta alongside so I don't want take away from that what I don't have to, what could I lower CPU usage to which won't impact MW significantly? (running 2 WUs).

Btw re stats, I wouldn't have been surprised by low stats on my 1st day, but I'm not even listed! I know I've produced some.

I don't think Milkyway hits the CPU much regardless of what that's set to. My app_config is like that because I'm running 3 pretty strong graphics cards on a low end older i3, so I don't run any CPU projects on it. I think I would try setting it to 0.05 CPU usage in the app_config and then run GPUz and make sure it shows 100% load and if so you should be good.

I'm grabbing the Milkyway stats from Free-DC, and it only seems to update once a day from what I can tell. I'm not sure if that's a Free-DC thing or a Milkyway thing, but it seems to update the daily total around 4 or 5 PM EST. It's not 100% ideal, but my thought process was I'll just post the total each morning that Free-DC shows around the same time, and that should work out to be pretty consistent. The first stats update is probably more like 5PM the day before EST to 5PM yesterday if I had to guess, but I'm just going with what the stats site shows when I grab the data. If you're not on tomorrow's stats, though, that would be an issue so I'll check that when I post them If you look at Free-dc it shows 0 points for everybody today because the stats haven't updated yet - https://stats.free-dc.org/team/mil/77

If there's a better way for me to do the stats that isn't super labor intensive I'm open to ideas I appreciate the participation so far!

 

[Fardringle]

edit: Ignore me. I misread what I was looking at, and accidentally wiped out this whole question anyway.

 

[emoga]

@Fardringle Try this for GPUGrid. If you still arn't getting close to 100%, switch the gpu_usage to 0.5

<app_config>

    <app>
        <name>acemd3</name>
            <gpu_versions>
                <gpu_usage>1</gpu_usage>
                <cpu_usage>1</cpu_usage>
            </gpu_versions>
    </app>

</app_config>

An easy way to find what an unknown app_configs 'name' is for a different project is to use a wrong 'name' in the app_config.xml and then check the message log.
It will then log what the potential name is.

 

[Assimilator1]

Icecold
Not a clue, unless a different stats site would help? We'll see what happens tomorrow.
Btw, is EST 7hrs behind GMT?

Re app config, ok I'll try that, thanks

 

[Icecold]

Icecold
Not a clue, unless a different stats site would help? We'll see what happens tomorrow.
Btw, is EST 7hrs behind GMT?

Re app config, ok I'll try that, thanks

EST should be 4 hours behind GMT. Free-DC will likely update in the next couple hours, but my plan was to post stats around the same time(morning) each day for consistency sake.

 

[Fardringle]

I think simplicity is important, and Free-DC makes things easy since the scripts already exist. We just need to recognize that the numbers won't always match the real-time stats on the project site. The points will get credited eventually.

 

[TennesseeTony]

Regarding the app_config/CPU usage for MW, on AMD cards, MW uses almost NOTHING from the CPU, you can set that to .01 if you like. I could easily run 5 cards/15 tasks on a dual core CPU with room to spare. Also I found .33 GPU usage to be a good performer (3 tasks at once, versus 4), on my 280X cards. For the short time I owned a Radeon VII, I think I had about 36 tasks running at once, over 12 clients, and STILL ran out of tasks frequently, lol, that card is a DP MONSTER!!! 1X PCIE is plenty of bandwidth as well, with any card.

 

[Icecold]

Regarding the app_config/CPU usage for MW, on AMD cards, MW uses almost NOTHING from the CPU, you can set that to .01 if you like. I could easily run 5 cards/15 tasks on a dual core CPU with room to spare. Also I found .33 GPU usage to be a good performer (3 tasks at once, versus 4), on my 280X cards. For the short time I owned a Radeon VII, I think I had about 36 tasks running at once, over 12 clients, and STILL ran out of tasks frequently, lol, that card is a DP MONSTER!!!

I didn't realize you no longer had the Radeon VII. I kept thinking "man I could have sworn @TennesseeTony posted at one point about having a Radeon VII I wonder if he'll jump in on this competition"

 

[Fardringle]

Regarding the app_config/CPU usage for MW, on AMD cards, MW uses almost NOTHING from the CPU, you can set that to .01 if you like. I could easily run 5 cards/15 tasks on a dual core CPU with room to spare. Also I found .33 GPU usage to be a good performer (3 tasks at once, versus 4), on my 280X cards. For the short time I owned a Radeon VII, I think I had about 36 tasks running at once, over 12 clients, and STILL ran out of tasks frequently, lol, that card is a DP MONSTER!!! 1X PCIE is plenty of bandwidth as well, with any card.

I'll give 3 tasks a try to see how it goes. Thanks!

 

[TennesseeTony]

I didn't realize you no longer had the Radeon VII. I kept thinking "man I could have sworn @TennesseeTony posted at one point about having a Radeon VII I wonder if he'll jump in on this competition"

I bought it for the Pentathlon. I would have kept it if it didn't crash on both Winblows and Linux daily. I even under-clocked it. MW didn't get picked for the Pent, and the driver issues prompted me to return it to Amazon. Sorta wish I had kept it now that the prices are ridiculously high. Only paid $450 and tax.

 

[Endgame124]

Regarding the app_config/CPU usage for MW, on AMD cards, MW uses almost NOTHING from the CPU, you can set that to .01 if you like. I could easily run 5 cards/15 tasks on a dual core CPU with room to spare. Also I found .33 GPU usage to be a good performer (3 tasks at once, versus 4), on my 280X cards. For the short time I owned a Radeon VII, I think I had about 36 tasks running at once, over 12 clients, and STILL ran out of tasks frequently, lol, that card is a DP MONSTER!!! 1X PCIE is plenty of bandwidth as well, with any card.

Thanks for the tip - I’ll need to adjust that so I can run an extra WCG task. I’m tempted to try 2 tasks on my r250X, but I’m going to imagine the bottom barrel card of that gen is only good for 1 task at a time.

 

[VietOZ]

I can confirm what Tony said. AMD cards used literally 2 or 3 seconds of the CPU, that's it. However, it's a different story on Nvidia. I haven't run Milky for a while, but if my memory served me right, it could use almost a whole thread/task. Even when Boinc Manager said it only used a fraction of a thread, but actually if you check Task Manager and you'll see it used way more than that. I couldn't remember exactly if it was Einstein or Milky, but I do know they're kinda similar on Nvidia cards. I'm not sure about the Titans as I don't own one.
Milky also had this annoying thing. Maximum WUs you can get was 900 WUs, doesn't matter if you spoofed 10 GPUs or 64 GPUs. All you can eat is 900 WUs. The things is ... you'll have to finish all 900 WUs before you can get new work. Then sometimes it caused your machine just sitting idle for hours before it got new works. I don't know if they fixed this or not, but that was pretty annoying when I ran it a while back.
While the VII can run 30 plus tasks at once, but the most points it can produced is still 3 tasks. The VII is beast, but it's also very tricky to work with. 1X PCIE will work, but won't get the max performance out of it. Even the CPU load can affect its performance, specially when undervolt. A Radeon VII should be good for 1.75M to 1.8M per day on Milky. If anything less than that, by a significant amount, then you're doing something gimping its performance.
That's AMD favor projects like Einstein and Milky. On Nvidia favor projects like PrimeGrid/Collatz/Amicable/etc ...etc ... the VII can hold its ground too. As y'all can see, most of my damage on Sprints/Marathon, on GPUs came from my VIIs. I don't own anything higher than 2060 on Nvidia's side. But I still managed to stay competitive when Nvidia favor project announce. Part of it is speculative bunker, other part was the VII.
The only thing I hate it was the heat. I literally haven't turn on my heater for the last 2 years during winter. Just pop Einstein or Milky on and my whole house will be warm enough

Anyways, Milky is not in the way of current FB competition. So i thought I'd share some of my experiences on it for the fellow crunchers.

 

[Fardringle]

It's hard to tell for sure since I haven't been running 3X tasks for very long on my 'new' 280X, and those tasks are mixed up with the 2X and 1X tasks that I ran previously since a lot of those are still being validated. But picking out the ones that look like 3X runs based on their run time, it's looking like I should get around 460K per day if I were to let it run 24/7, compared to about 340K per day that it was looking to average while only running one task. That's a pretty good improvement and crazy results for a 7-year-old card!

It won't be those actual numbers since I have to pause BOINC while playing games on the computer, which I tend to do far too much of on weekends, but during the week it should be pretty close to 24/7.

 

[VietOZ]

I assume that you're talking about your RX580. For the 580, best output is 2x. You can even undervolt it a bit and overclock the memory for efficiency. This is strictly for Milkyway

 

[biodoc]

Milky also had this annoying thing. Maximum WUs you can get was 900 WUs, doesn't matter if you spoofed 10 GPUs or 64 GPUs. All you can eat is 900 WUs. The things is ... you'll have to finish all 900 WUs before you can get new work. Then sometimes it caused your machine just sitting idle for hours before it got new works. I don't know if they fixed this or not, but that was pretty annoying when I ran it a while back.

The "time between reload" of new WUs is several minutes the last time I watched. My only workaround to minimize this issue is to run 4 boinc instances simultaneously on the Radeon VII. Ideally, while one instance is waiting for a reload, I have 3 others still feeding the card.

A Radeon VII should be good for 1.75M to 1.8M per day on Milky. If anything less than that, by a significant amount, then you're doing something gimping its performance.

I'm seeing 1.6 M now but I have the power capped at 150 watts. I'm on linux so I use Rick's amdgpu utilities to control the settings. I could push the power up a bit now that winter is coming.