Einstein paying better?

[MindCrime]

I just started running Einstein again, I'm currently running it on a 270x and some testing on a 7970, immediately I noticed my cards are producing a lot more credit than they did in the past. They're running the newer app: "Gamma-ray pulsar binary search #1 on GPUs" (FGRP_opencl1k_ati)

Not sure if Nvidia is getting the same love, though.

[denim]

That newer app seems to be paying fairly well. I also recently started working on E@H again. Need that 5m.

[FourOh]

It's paying about 3x better than the Parkes app. Here's an up to date app_config to run 2x concurrent:

<app_config>
<app>
<name>einstein_O1MD1CV</name>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
<app>
<name>hsgamma_FGRPB1G</name>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

[denim]

Thanks for the code.

[MindCrime]

I wish I would have doubled up the concurrency on the 7970 sooner. I went from ~700sec per wu to ~1150 with two running. So twice the credit but only 60% slower.

Edit: watch your vram usage, i'm currently seeing 1600MB with 2 going.

[Bryan]

For a point of reference a GTX 1080 does 2 in 1025s. It does 4 in 1997s so 4 is actually a little faster. I don't care about the credits but I need 1 itty bitty star to get a big star on WuProp :D

BTW 4 takes 3755M of VRAM on Nvidia.

[MindCrime]

I've been running einstein on a 270x and 7970 for about 6 days straight. The 7970 runs them faster and thus goes through more work units than the 270x. The weird thing is the 270x has 7 pages of pendings while the 7970 is only on 3. All WUs on this app are quorum 2.

[Mumps]

Maybe an issue comparing different aps? Could the 7970 GPU be pairing up with like performance cards and the 270x is pairing up with a larger collection of slower cards? Or even you're hitting scole as you wingmate more often on the 270X so you're stuck waiting for his fictional 50 GPU's to get to the tasks? :D

[scole of TSBT]

Are you making fun of my collection of slower cards? :p

[Mumps]

I'm making fun of your collection of "50 GPU's" in a single system for caching purposes. :))

I'm the one with the slower cards. (You would be hard pressed to find *anyone* with as many cards as I have with an average GFLOPs even close to as miserable as mine.)

lol

[Bryan]

I'm making fun of your collection of "50 GPU's" in a single system for caching purposes. :))

I'm the one with the slower cards. (You would be hard pressed to find *anyone* with as many cards as I have with an average GFLOPs even close to as miserable as mine.)

lol

Do you actually have any cards that are measured in Gflops? =))

[Mumps]

Pretty sure I still run a couple measured in MFLOPS. :)

[denim]

I have a few cards that just went FLOP. :-|

[scole of TSBT]

I'm making fun of your collection of "50 GPU's" in a single system...You mean this rig :)

2660

[denim]

That's unreal. Can't imagine what it can do.

[scole of TSBT]

That's unreal. Can't imagine what it can do.That's actually a scrypt mining setup. I don't know the guy. I just ran across the pic one day.

[MindCrime]

Hey that's only 30gpus, unless they're 5970/7990? He probably had to sell that lambo for the electricity.

[Al]

So this was my wingman (http://www.aei.mpg.de/24838/02_Computing_and_ATLAS) on a very long Einstein wu...and this wingman (http://www.aei.mpg.de/24838/02_Computing_and_ATLAS) errored it out at 374k seconds. I guess I'd expect more out of my hardware. ;)

[DrPop]

So this was my wingman (http://www.aei.mpg.de/24838/02_Computing_and_ATLAS) on a very long Einstein wu...and this wingman (http://www.aei.mpg.de/24838/02_Computing_and_ATLAS) errored it out at 374k seconds. I guess I'd expect more out of my hardware. ;)
Bummer...man, what we wouldn't give for that rig in the Pentathlon, huh? :D

[Mumps]

Just ask Dr Bruce nicely to join our team for any Einstein leg. :) He used to be at my local state college campus. :)