MW Work

[joker]

Don't know about anyone else but I am getting work from MW that is actually working and not giving me computational errors. Hope these work and the double credits are valid now.

[Mike029]

Yub yub, The Milk is flowing again.

[joker]

The Milk is flowing again.

I will try not to turn that into something perverted. ;)

[Maxwell]

How long are they taking each?

[Mike029]

140.90 79.12 213.76
157.94 79.89 213.76
174.44 81.43 213.76
234.71 95.19 267.20

That's the times I have at 900/900

[joker]

How long are they taking each? Based on my last 2 WU's (one with a 5830 OC'ed and one with a 5850 Oc'ed) 2:42 with the 5850 and 2:59 with the 5830.

[Maxwell]

It sounds similar to what it was before - which probably means and app_info file to run two at a time and change CPU priority would be good.

Give me a few and I'll see what I can do. Anyone willing to test it (I'm all-in with Collatz now...)?

[Maxwell]

Here is what I have for you MW app_info.xml enthusiasts:


<app_info>
<app>
<name>milkyway</name>
</app>
<file_info>
<name>milkyway_0.57_windows_intelx86__ati14.exe</name>
<executable/>
</file_info>

<app_version>
<app_name>milkyway</app_name>
<version_num>57</version_num>
<plan_class>ati14</plan_class>
<flops>1.0e11</flops>
<avg_ncpus>0.05</avg_ncpus>
<max_ncpus>1</max_ncpus>
<coproc>
<type>ATI</type>
<count>.5</count>
</coproc>
<file_ref>
<file_name>milkyway_0.57_windows_intelx86__ati14.exe</file_name>
<main_program/>
</file_ref>
</app_version>

</app_info>

This will not work with command line options - I tried, but I think they are going to have to update the code to do so. Regardless, this version works and will run 2 WUs at a time; if you unsync them, you should have a pretty constant high load on the GPU and not have to wait for the next WU to load.

This was standard practice with the older ATI opti app version because the WUs were so short.

[Bryan]

The new app really sucks! It is much slower than the old .23 Gispel version (75s versus 108s on a 950 MHz 5870). It also uses 1/2 a CPU core per wu. On my Q6600 quad running AQUA and driving a 5970 I was only getting 50% GPU loading. Turning off all CPU crunching the load went up to 96% on both GPUs.

On the I7 with a single 5870 I can run AQUA and still get 96% GPU loading.

MW is NOT paying double credits yet. Once they get the current app stable they will make an announcement telling which days you can do double dipping ;)

EDIT: The app_info works great Max ..... thanks!

[Maxwell]

The new app really sucks! It is much slower than the old .23 Gispel version (75s versus 108s on a 950 MHz 5870). It also uses 1/2 a CPU core per wu. On my Q6600 quad running AQUA and driving a 5970 I was only getting 50% GPU loading. Turning off all CPU crunching the load went up to 96% on both GPUs.

On the I7 with a single 5870 I can run AQUA and still get 96% GPU loading.

MW is NOT paying double credits yet. Once they get the current app stable they will make an announcement telling which days you can do double dipping ;)
I ran just a few trying to get the app_info file working. I saw quite a bit of variance in WU times.

And yea, regenerating code is ugly... but they did seem to up the credits.

[dan]

Max I just tried the app_info.xml change. It worked and got two WUs running at the same time, but my system became unresponsive.:(

[Mike029]

Max I just tried the app_info.xml change. It worked and got two WUs running at the same time, but my system became unresponsive.:(

Don't run the opti app. It will validate fine and run on your box but MW will invalidate it on their end. Travis is still trying to stabilize the project. No opti apps for now. >:P

[Maxwell]

Don't run the opti app. It will validate fine and run on your box but MW will invalidate it on their end. Travis is still trying to stabilize the project. No opti apps for now. >:P
Nothing is optimized - just configured to run two at a time. But you're probably right - don't try messing with things until they're stable again. The app_info file should still work then...

[Crazybob]

On one of my boxes, I wasn't able to detach in time. It Dl'd more of the .23 work? Are these valid tasks?

[Crazybob]

Answered my own question. Completed 4 tasks and all say validator error. Detaching and reattaching now.

[Crazybob]

Ok, got another one. My 64bit quad reattached, but all work keeps erroring out. Detached and reattached again with same results? WTF?

[Shadow]

The latest application version is 0.57, so if you have any lingering 0.23's, they'll error out. Opti apps will cause errors too.

[Mike029]

It seems that there is a heavy cpu load running the new app. in MW. Running MW alone I complete taskis in 80 to 100 sec. with 99-98% load on my 5870 900/600.

Running cpu wu's I'm finishing MW wu's in 220 to 280 secs. with 50-70% load. On my duel core I'm suspended cpu wu's as the extra seconds I cut off a wu will more than make up for the loss of cpu credits.

Play with the settings you may be able to run some cpu on 4 and 6 core processors. Post your results.
Cheers.
Mike.

[Crazybob]

I have none of the old WU's and searched my machine for any remnants of the old program without any luck. As soon as I reattach, all new WU's error out.

[Shadow]

Some others are having similar problems in the MW forums and say that updating to the latest Catalyst version fixed it. Looks like a crapshoot for now though. Anyhow, here's the link:
http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=2339

One more link:
http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=2345

[Crazybob]

Thanks for the info. If I remember right, I'm already running the latest. Will have to go over to MW and check around to see what's what.

[Crazybob]

Will have to double check what CCC I'm using. Did read over there, that the new ap has a problem using 100% of 1 core. There is a fix, but not yet released.

[Crazybob]

That was the culprit. Updated to latest 11.3 CCC and looks like they are going through. Thanks for the advice.:)

[Crazybob]

Here is what I have for you MW app_info.xml enthusiasts:


<app_info>
<app>
<name>milkyway</name>
</app>
<file_info>
<name>milkyway_0.59_windows_intelx86__ati14.exe</name>
<executable/>
</file_info>

<app_version>
<app_name>milkyway</app_name>
<version_num>59</version_num>
<plan_class>ati14</plan_class>
<flops>1.0e11</flops>
<avg_ncpus>0.05</avg_ncpus>
<max_ncpus>1</max_ncpus>
<coproc>
<type>ATI</type>
<count>.5</count>
</coproc>
<file_ref>
<file_name>milkyway_0.59_windows_intelx86__ati14.exe</file_name>
<main_program/>
</file_ref>
</app_version>

</app_info>

This will not work with command line options - I tried, but I think they are going to have to update the code to do so. Regardless, this version works and will run 2 WUs at a time; if you unsync them, you should have a pretty constant high load on the GPU and not have to wait for the next WU to load.

This was standard practice with the older ATI opti app version because the WUs were so short.

New version 0.59 This solves for CPU usage issues. I have edited the app_info file above to use with this version.

[Mike029]

Credits are flowing nicely now. Anyone try running 2 per gpu? If so, is it worth it?

[joker]

Now the server is down for maintenance! Guess I will be getting my 5 million in Collatz after all. :p

[Crazybob]

With the App_info file above, you can run 2. On my 4870's, can run 2 in about 6 min. Bad thing is they shut it down again.:mad: Just can't win.

[Fire$torm]

Well the server is back up.

I have had Zero problems with MW wu's. Had issues with MW before until I rolled back CCC to 10.10 for my 4850s. Been smooth sailing since. BTW my OS is Win7 x64.

[joker]

I too am getting WU's now.

[Maxwell]

Fun fact: there is a way to increase the Priority of the GPU to up the load:

<cmdline>-r0</cmdline>

Add that in the app_info file, and you should see differences. Warning: probably not good for non-dedicated crunchers.

Thread where they are talking about this: http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=2343&nowrap=true#47532

[joker]

to up the load:.....

I will just say nothing.... :-B