I think that there is no need to suspend the project while these problems persist. Configure your MilkyWay@home preferences to only accept CPU work and not GPU work.

I think that there is no need to suspend the project while these problems persist. Configure your MilkyWay@home preferences to only accept CPU work and not GPU work.

Sorry I try to collect gold badges from primgrid with cpu tasks.

Is anybody having there "bundle" tasks validated? Mine are all failing validation.

This is for a UBUNTU 16.04 system using the 1.40 MW app.

wb8ili asked:

Is anybody having there "bundle" tasks validated? Mine are all failing validation.

This is for a UBUNTU 16.04 system using the 1.40 MW app

Yes, I have some validated for Ubuntu 16.10 with NVIDIA 970's.

Hey everyone,

i'm a first day newbie in milkyway (POEM finished the project, so i was lookin for another gpu-project with AMD support), but i'm also a distributed computing user since 1999...

My first thoughts after attaching to project were "nice, it downloaded cpu and gpu tasks fromt the same batch, let's see how my one week old AMD RX 480 performs, especially regarding to the DP mode and how the gpu- and cpu perform against each other..." example of wu:

Milkyway@Home 0,751C + 1 AMD/ATI GPU 1.42 milkyway (opencl_ati_101) de_modfit_fast_19_3s_136_bundle5_ModfitConstraints3_3_1478900148_101814_0

next thought was "wtf???" to keep it short:
the gpu-tasks take as long (~) as the cpu-tasks, the gpu is completely unemployed (2% avg. gpu and memory controller load), the wu well finshes, but avg cpu load per wu is 99,8% - on a gpu-task!
OK, the FX-8320E isn't a hell of a machine and i read then in the forum that this client runs better with more than one gpu-task assigned to a gpu, but 10 times 2% is still just 20% gpu-load, but with the need of simultaneously 20 cores??? c'mon my FX8320E isn't that bad ;-)

I was close to desperation, wondering myself if this would be my first project wich i immediately deteach from or if i should crunch just the cpu-tasks, but at last i found this thread and now calmed down because of "oh bad luck at the beginning, seems to be a faulty new client version 1.42" .

So thx4information and until it's solved i guess i have to wake up my gpu with einstein ;)

So, ehrm, jepp, can confirm exactly the same behaviour like in post 65714-65716 and so on.

Have suspended all GPU tasks for now.

Still having same issue with tasks restarting at 60% on both my Intel and AMD machines, Even after resarting pc and client.

Anyone else having this issue?

Brian

Hello everyone,

There are some anomalies with the new version that was released and Jake has been putting in some long hours. Please give him some time to address them. We are just going through some growing pains with the restarting of % done and the GPU work units being executed on the CPU instead of the GPU. I'm sure not much longer and he will have things ironed out.

Thanks Jake for all you effort and get some rest. We always perform better rested.

Rich

captianjack --


I looked at your computers. I see two LINUX machines. I only see one task between the two machines that is a "bundle" and it did not pass validation.

What am I missing?

Is anybody having there "bundle" tasks validated? Mine are all failing validation.

This is for a UBUNTU 16.04 system using the 1.40 MW app.

Most of mine are being validated in Mint. Before the update I had a several % error rate. I had checked some of the top users awhile back and it seems to just be linux related as Windows users didn't have any errors. Now it seems the error rate went down but they are invalid now. Still most are valid. 202k points so far today, #3 overall on Free-DC with a stock 280x.

wb8ili,

When a task is sent to multiple computers and too many of the computers get an error, the status of the task is set to "Completed, can't validate. If you look at the work unit in question, it was sent to 3 other computers and they all got errors. The error text says "Too many errors (may have bug)".

I am running some more right now and you should be able to find some that did validate.

Hope that helps.

Is anybody having there "bundle" tasks validated? Mine are all failing validation.

This is for a UBUNTU 16.04 system using the 1.40 MW app.

Are they failing, or are they ending up in "can't validate" status? (See below.)

I've had a lot of bundle tasks validated (Ubuntu 16.04 with a laptop NVIDIA GPU and Ubuntu 14.04 with a low-powered AMD GPU)

Tasks seem to be of two types - one requiring a wingman (or two), the other not requiring confirmation by comparison.

I haven't had any of the former fail to validate (yet - I expect problems eventually with the AMD GPU 'cos it occasionally seems to mis-compute the integral and single-task jobs would become invalid because their results didn't match the "wingman" results...)

However, I've had several bundle tasks end up "Completed - can't validate" and all of those have failed out because of errors on Windows platform tasks - I've not yet seen a failed Linux wingman...

I agree with others who say we should give Jake the time to sort this out when he's not exhausted! The fact that the only [serious?] issue I can see on my Linux platforms is the way the jobs report the percentage of the individual task completed rather than the percentage of the batch(!) suggests that this is a build issue rather than a [severe] logic issue.

Cheers - Al.

Just to clarify - I am referring to "Can't Validate" bundled GPU tasks under UBUNTU.

Here is an example - (can't figure out to make it a link)


http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=1886765954

Downloaded milkyway_1.40_windows_intelx86__opencl_nvidia_101.exe for by XP machine -- all WUs immediately ended in error (that was the latest compatable app that I cound find).

Downloaded milkyway_1.42_windows_intelx86_64__opencl_nvidia_101.exe for my Win10 machine and WUs appear to be running OK -- so far.

However -- I just noticed that there appears to be pretty heavy CPU usage for them. So maybe not so OK. . .

---

My GPU units are not executing on the GPUs, they're running on the CPU and sucking up CPU time. My machine is running win7, and i typically run 2 units per card.

Oh well, looks like I'm waiting for the next fix. Off to SETI!

I think that there is no need to suspend the project while these problems persist. Configure your MilkyWay@home preferences to only accept CPU work and not GPU work.

Ehm, the whole discussion over here is all about GPU tasks which, because of their short duration and the limited amount which a single machine can retrieve, are hammering the server down in a DDoS attack-like fashion.

Michael.

P.S.: Although there is the "no load issue" regarding GPU core and RAM clocks (going down to 300/150 MHz as soon as an MW task is initiated), both types of WUs do validate - I checked one of each of the two types on my AMD cards.

---

President of Rechenkraft.net e.V. - This planet's first and largest distributed computing organization.

Same issue here, the GPU isn't utilized by the new version. (1.42 opencl_ati_101)
CPU load is high.

wb8ili asked:

Just to clarify - I am referring to "Can't Validate" bundled GPU tasks under UBUNTU.

Here is an example - (can't figure out to make it a link)


http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=1886765954

wb8ili, it looks like all your "bundle" tasks are failing. On that particular task, it was sent to two other computers and they both failed. After failures on three computers, the Milkyway host decided that it might be a bad work unit and gave it a status of "Can't Validate".

At this point, the real question is why all of your "bundle" tasks failing. The only major difference I can see between your setup and mine is that you are using an older (February 2014) version on BOINC on your machine. I am using the current one that is distributed by the BOINC software installer. You might try to upgrade BOINC and see if that helps.

I report that I'm successfully completing GPU WUs on a Radeon R9 290 under Windows 10. They take 43.5 minutes to complete, and I've been given 37 of them at once. Excellent!

I report that I'm successfully completing GPU WUs on a Radeon R9 290 under Windows 10. They take 43.5 minutes to complete, and I've been given 37 of them at once. Excellent!

...and if you look really careful, you'll recognize, the GPU is absolutely not used by them, only the CPU. X)

---

Aloha, Uli

Damn, I thought my graphics card fan was quiet today. What's going on with the "fake" WUs?