Anyone have a spare GPU to throw at Milkyway?

[NeoGen]

Hey guys

I've been on MilkyWay for a few weeks already pushing on with my Radeon HD 7970, making nearly 400K a day, and there' a few more people contributing to it but I suspect they are using CPU's instead of GPUs because the daily points are very low.
Well, the thing is... there's one team behind us (Ellison Crunchers) only 5M points away from us and slowly closing in each day. They have 1 guy who's pushing just a little bit more than me and the others all combined on our Team.

If anyone got a spare GPU they can throw at this project to keep Ellison Crunchers at bay please join in, I'm trying my best not to lose our current spot on the stats. :)

http://www.quickmeme.com/img/8c/8ca00d1556f2c61785cb35ec5cf1dd9953ed6672c2b490445a 97101defc3bf07.jpg

[Dirk Broer]

My GTX260 died, Windows 10 broke OpenCl 1.1 for my HD 4770, the only 'card' I currently have that can crunch MilkyWay (DP required) is the IGP on my A10-5700 APU...

[vaughan]

I'll see if I can put a GPU on it later today.

[Brucifer]

I just put a GTX-970 on it. The first unit came back as validation inconclusive............ ????? Anyway I selected all the choices for work units for GPU so we will see how that goes. The other issue is heat....... :-( so again we will see how that goes. :-)

edit: just crunching the milkyway units now, no "n"body. getting fewer inconclusives now.

[AMDave]

I'll throw some cycles this way.
I need the ranks :P

[Dirk Broer]

Perhaps I can divert the HD 7790 from the 'fast reaction force' to the A10-5700 system, and place its A8-3850 (plus 8GB RAM) in my (reserve force) ASUS F1A75-M, instead of that mobos present A6-3500. I can then test my HD 7750 in that board to see whether that Graka has died too....The A6-3500 so will end up where it once started: on my ASUS F1A75.

[Brucifer]

I removed the gtx-970 and replaced it with a gtx-750. gonna have to joggle a few system gpu's around to be able to crunch some stuff I also wanna crunch.

[NeoGen]

I don't know if it's really better or just me, but I'm using an app_config.xml to crunch about 10 WUs at a time. It's slower than 1 at a time but I think it's squeezing a bit more of performance out of it.

[NeoGen]

Heeelp! Team Ellison is now just 2M away from us and closing in each day. :icon_eek:

[Dirk Broer]

To make it worse: team Ellison (aka Ellison Crunchers) consists out of thirty-seven people and are active on only five BOINC projects.
One of their members, just one (Gary Roberts), is active on MilkyWay@Home....

[Nflight]

I decided to help out a little here, I put the GTX-1080 on the project and let's see what I can get out of it. Right now each work unit takes just 2:11 to complete! The points values aren't really that good for a GPU card to be frustrated by all the network activity that it has to manage plus crunch.

[Nflight]

Would you be willing to share your app_config file?

[Nflight]

I don't know if it's really better or just me, but I'm using an app_config.xml to crunch about 10 WUs at a time. It's slower than 1 at a time but I think it's squeezing a bit more of performance out of it.

Would you be willing to share your App_config file?

[NeoGen]

I think I fixed the thread. :)

In my latest iteration of the app_config I lowered the number of instances to 6 because I'm running an R9 380, instead of my good old 7970. As much as I liked the 7970 and it gave me great amounts of credits on a double precision hungry project like Milkyway, the difference in the monthly power bill is noticeable now in the summer. Not just because the 7970 sucks up twice the watts or more, but also because it churns out a lot more heat, putting pressure on the AC.

This is a pretty simplified app_config, it probably could use better tweaking, especially as I never actually measured the CPU usage, I just threw in 0.1 of a core per WU and figured it should be enough.


<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.16</gpu_usage>
<cpu_usage>0.1</cpu_usage>
</gpu_versions>
</app>

<app>
<name>milkyway_separation__modified_fit</name>
<gpu_versions>
<gpu_usage>0.02</gpu_usage>
<cpu_usage>0.1</cpu_usage>
</gpu_versions>
</app>

</app_config>

[KarmannGaz]

I just remembered why I don't let MilkyWay try to use my GPU - it needs to be a double-precision GPU, and my HD6670 is only single-precision, so all the tasks i got just failed! Switching preferences back to CPU only and will let MilkyWay have a look in again once the Pentathlon is over. :icon_rolleyes:

[Dirk Broer]

They're a pretty stern lot over there at MilkyWay: Not only do they require DP capability, they want you to run a pretty recent GPU as well.
My HD 3850 (AGP) became useless that way, as well as my HD 3870, my HD 4770 and my son's HD 4850.
All perfectly capable cards, DP-wise, but lacking in the OpenCL capabilities (absent with the HD 3800 series and limited to OpenCL 1.1 with the HD 4700 and HD 4800 series).
Last thing I read however is even more shocking: the IGP of a AMD A10-7700K was recognised as ATI GPU R600 (R38xx) does not support OpenCL, but that person runs Ubuntu 16.04 with its rotten AMD Radeon support.
Personally I've crunched MilkyWay on the IGP of my A10-5700..

[KarmannGaz]

They're a pretty stern lot over there at MilkyWay...

lol, no worries, just means Einstein gets exclusive use of my GPU for the time being :icon_lol:

[SuicideCabbage]

Are the legacy apps no longer available? I remember two years ago or so taking the 3850 off the shelf and setting it up as anonymous platform with the old pre-OpenCL app. Worked fine then, by the way the two 3850s are still shelved, and I will probably be phasing out 2-4 5830s soon if anyone is interested. The 5830s are still good cards, but I am annoyed with trying to get them to work on newer Linux kernels, the 3850s may be hard to get rid of though. To bad they are not AGP, I could bring some boards back from the dead and do something on them. If memory serves the 3850 did a Milkyway in 17-19m (non-OpenCL) and the 5830s clear two units every 2m or so.

[Dirk Broer]

It's all OpenCL nowadays at MilkyWay (http://milkyway.cs.rpi.edu/milkyway/apps.php), GPU-wise I'm afraid.
This rules out the HD 3000 series as modern OS-es do not support OpenCL for them.
The HD 5000 series should work under Windows, using this driver (http://support.amd.com/en-us/download/desktop/legacy?product=legacy3&os=Windows%2010%20-%2064), but MilkyWay seems to have a very rotten GPU detection, spitting out 'Message from server: ATI GPU R600 (R38xx) does not support OpenCL' more often than not, and seemingly regardless of the actual GPU your system is toting.

The OpenCL support for AMD cards under Ubuntu and related OS-es (e.g. Lubuntu, Xubuntu, Kubuntu, Linux Mint, etc) became a bit dramatic after release 16.04.
Even the R3 that is in my Socket AM1 Athlon 5350 can't crunch OpenCL applications, as the OpenSource driver is at OpenCL 1.1 and FGLRX does not work from this version on.
That same Athlon 5350/R3 combo works flawlessly in Windows 10 and Xubuntu 15.10, btw.
On the AMD driver page you can read wich cards are supported with the new AMDGPU-Pro Driver Version 17.10 for Ubuntu ​16.04.2 (http://support.amd.com/en-us/kb-articles/Pages/AMDGPU-PRO-Driver-for-Linux-Release-Notes.aspx)

[Nflight]

I have returned to run some Milkyway. I tried NeoGen's optimized 6 wu's at a time and saw BSOD twice in 30 minutes. So I downgraded the operation to just 4 work units at a time. With 4 running at a time I am finishing them at just under 10 Minutes. Roar I hate this project..... BUt I need to raise my score so I will keep trying...

[NeoGen]

I guess it might be because of the double precision calculations but this project raised the temps of my GPU a handful of degrees Celsius higher than any other project I noticed, even with my AIO liquid cooling rigged on it.

Nflight, do you keep an eye on the temperatures with GPU-z or other software?

[Nflight]

After 4 hours of work I can run 36 GPU work units at once, but in order to make it work, I need 4 CPU cores to manage the whole brunt. This raises the heat from the GPU's to nearly 80°C so I am back to running 4 at one time and the Temperature levels off at 57°C which is much nicer.

Since I run CPU cores for another project I need as many CPU cores as GPU cores working as efficiently and effectively as I can muster.

As for observing the Motherboard heat production i Have an Asus brand M/B which already has a built in sensor array, this also tells me which Fans are running and at what speed, plus voltage of around the parameters of the board. All this happened last night so now I am allowing the system to run for 24 hours before I attempt anyother re-configuration to enhance my system. Hopefully I won't blow it up!

[Nflight]

I would like to take this moment and say congrats to Paul_AMD for reaching the Top 20 in our Team. That is a feat especially in this Project, Your a rare case NeoGen. If I had just a wee bit more patience I would try and dethrone you, maybe in the future!

[NeoGen]

It was a painful trek uphill, I don't know if I'll do another one of those any time soon again. I pretty much ran it all on one single 7970 GPU. But it was worth it :)

And congrats Paul, you are going up the ranks fast. :icon_thumright:

[Paul AMD]

Thank you. That bad news is I'll be out of town for a few days so production will fall off until my return. Keep up the good fight!

[Nflight]

I just reached out to our newest member who is showing up on our screens, Robert (http://milkyway.cs.rpi.edu/milkyway/hosts_user.php?userid=1261833) is his nickname on the Team Effort. I am hoping he joins us as I would love for him to share with us details about his Ryzen 7 16 core machine he has running... :blob3: