Credits gravity vs gamma

Mr P Hucker
Mr P Hucker
Joined: 12 Aug 06
Posts: 838
Credit: 518,671,567
RAC: 38,363
Topic 227469

Why am I getting far more credit for Gamma on GPU than gravity on GPU?

If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4,896
Credit: 18,443,325,997
RAC: 5,856,536

All because that is what the

All because that is what the admins decided for fixed credit awarded that they do here.

 

Mr P Hucker
Mr P Hucker
Joined: 12 Aug 06
Posts: 838
Credit: 518,671,567
RAC: 38,363

I thought Gravity was the one

I thought Gravity was the one they were pushing?  So it should get more credits.  If I don't express a preference, I get inundated with Gravity.  Pity, since I have a machine with 2 GPUs, one can do Gravity, one cannot.  Impossible to put gravity on the good one and Gamma on the old one.

If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.

petri33
petri33
Joined: 4 Mar 20
Posts: 123
Credit: 3,718,145,819
RAC: 7,080,478

How about running 2 boinc

How about running 2 boinc instances with exclude_gpu?

Mr P Hucker
Mr P Hucker
Joined: 12 Aug 06
Posts: 838
Credit: 518,671,567
RAC: 38,363

petri33 wrote: How about

petri33 wrote:

How about running 2 boinc instances with exclude_gpu?

I could do but I have enough hassle running 7 machines and keeping track of everything.  I just wish Boinc was easier to tell what to do.  One simple thing and you have to give it 5 seperate instructions, not knowing which one takes priority.  I'll just run Gamma on both GPUs on that machine.  Why can't they make Gravity work on older cards?  Does it needs newer instructions?

If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.

mikey
mikey
Joined: 22 Jan 05
Posts: 12,552
Credit: 1,838,709,267
RAC: 13,516

Peter Hucker wrote:  Why

Peter Hucker wrote:

 Why can't they make Gravity work on older cards?  Does it needs newer instructions?

Is that the one that requires 4gb of onboard the gpu memory? I know one of the apps does.

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4,896
Credit: 18,443,325,997
RAC: 5,856,536

I don't run O3AS work but I

I don't run O3AS work but I believe they reworked the app or the task parameters so that the tasks use less than 3GB now.

 

Mr P Hucker
Mr P Hucker
Joined: 12 Aug 06
Posts: 838
Credit: 518,671,567
RAC: 38,363

It does, I have 3 gravity

It does, I have 3 gravity tasks running at once on a 4GB Fury.

Unfortunately the card just one version behind that, the Tahiti, doesn't do them, and I have 6 of those, it tries, but fails near the end.  You'd think the server would refuse to hand them out to Tahitis, but to expect common sense from server operators is pointless these days.

If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.

mikey
mikey
Joined: 22 Jan 05
Posts: 12,552
Credit: 1,838,709,267
RAC: 13,516

Peter Hucker wrote: It does,

Peter Hucker wrote:

It does, I have 3 gravity tasks running at once on a 4GB Fury.

Unfortunately the card just one version behind that, the Tahiti, doesn't do them, and I have 6 of those, it tries, but fails near the end.  You'd think the server would refuse to hand them out to Tahitis, but to expect common sense from server operators is pointless these days.

I don't know the specifics but sometimes you just can't specifically identify something like "Tahitis" and exclude them from getting such and such tasks

Mr P Hucker
Mr P Hucker
Joined: 12 Aug 06
Posts: 838
Credit: 518,671,567
RAC: 38,363

mikey wrote:Peter Hucker

mikey wrote:

Peter Hucker wrote:

It does, I have 3 gravity tasks running at once on a 4GB Fury.

Unfortunately the card just one version behind that, the Tahiti, doesn't do them, and I have 6 of those, it tries, but fails near the end.  You'd think the server would refuse to hand them out to Tahitis, but to expect common sense from server operators is pointless these days.

I don't know the specifics but sometimes you just can't specifically identify something like "Tahitis" and exclude them from getting such and such tasks

Easy in app_config.xml locally.  Just list the device numbers you don't want to get them.  Trouble is if I limit gravity to a,b,c and limit gamma to d,e,f cards, Boinc gets all confused and downloads 1 task at a time for 6 cards.  They're looking into the bug: https://github.com/BOINC/boinc/issues/4738

As for on the server, surely there's something in the Boinc server code that can allow forbidding tasks to certain models of card?  Boinc reports what cards I have to the server, I can see them listed on their webpage.  Maybe not "Tahiti", but they can list each specific model (like R9 280x) which should be excluded.

If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3,913
Credit: 43,793,599,309
RAC: 63,450,542

he means the project cant

he means the project cant really differentiate the AMD cards to be able to exclude them. the project scheduler only has limited information, and the scheduler can only base it's response on the displayed (first) GPU in the system. if you have a newer/better GPU paired with an older GPU, only the better one is shown. this is what you see when you look at your own hosts list. you'll see [n]FIRST_GPU_MODEL, where [n] is the total number of GPUs but the scheduler can really only see specs and make decisions based on that FIRST_GPU_MODEL.

this is a BOINC limitation, not a project limitation.

but for the systems that are "tahiti" already as the first GPU, again, the project isn't really aware of what that means. that's just what the driver identifies the device as. and it's not even very specific or correct. Tahiti is a codename for a wide array of models. it's a bit unreasonable to expect the project to keep a running list of all the models that do and don't work.

 

it's a little easier for projects to differentiate Nvidia GPUs because Nvidia groups their generations under a metric called compute capability. this is exposed to BOINC and projects can easily cut off generations using this value. want to limit your app to only Maxwell generation or newer? set your scheduler to only send it to systems with compute capability 5.0 or greater. and so on. unfortunately AMD doesn't appear to have a metric like this exposed through their driver to differentiate generations, while Nvidia has been doing this for 15 years. maybe you should contact AMD and tell them to create a metric like nvidia does and expose it in the driver, then contact BOINC and have them capture that metric so the projects can see it.

_________________________________________________________________________

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.