Optimizing crunching cost / benefit

Castaris
Castaris
Joined: 18 Sep 20
Posts: 11
Credit: 107,096,839
RAC: 1,402
Topic 231037

Hello!

A long time has passed since my last post. I have kept crunching for a long time, but nowadays I don´t have any dedicated machine running 24 / 7.

When I just joined, I put all my hardware to use and managed to reach the top 50 in weekly credits, more than a million credits a day. Since then, performance has been very variable and I never really got back to earning that many credits, even with the same hardware. It isn´t a hardware issue. 

I noticed that some tasks give more credits than others. I would like to know what is the difference of the many types of tasks and what is more cost effective. Gravitational wave tasks for example, tend to give 10000 credits, but they are few. Am I doing something wrong? What do some people do to reach such enormously high RAC amounts? I would like to get advice so I can join again with all my spare computers, which are many. 

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3,899
Credit: 43,536,822,644
RAC: 63,102,726

Right now the O3AS GW work

Right now the O3AS GW work gives the best credit/day. But you can also get some better efficiency by running two or more tasks at the same time, if you have the VRAM for it. 
 

however, in your case I’m not sure you will be able to do that with only 4GB cards on windows. The GW tasks from my experience use a little more than 2GB each. 
 

A few things have changed with the project. They discontinued the FGRPB1G project, which was the previous best points per day for GPUs, and that’s probably what you were running before. Even with the points boost of 10,000 credits for the GW tasks, it’s still only about half the overall points per day compared to that old FGRPB1G project. 
 

there are other factors like AMD vs Nvidia, where Nvidia generally performs better with the available CUDA apps here, and also Linux performs better usually, as under Linux and with Nvidia you can use additional tuning tools like CUDA MPS to get even more efficiency. 

_________________________________________________________________________

Castaris
Castaris
Joined: 18 Sep 20
Posts: 11
Credit: 107,096,839
RAC: 1,402

Oh... I see. I guess that to

Oh... I see. I guess that to get a more stable performance I would have to use more CPUs in a linux computer. Or are CPUs not worth it compared to the electricity cost? I could get some used workstations for dirt cheap, but I am not sure about it. 

What kind of hardware are you running? 

Castaris
Castaris
Joined: 18 Sep 20
Posts: 11
Credit: 107,096,839
RAC: 1,402

In my case, I own 4 AMD cards

In my case, I own 4 AMD cards from an old Ethereum miner. By putting all of them to work, I managed to get decent RAC levels, but probably they aren´t the best card for doing this and they are all 4GB cards. What are your recommendations? 

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3,899
Credit: 43,536,822,644
RAC: 63,102,726

if you want to stay with AMD

if you want to stay with AMD cards, i would get cards that have at least 8GB of VRAM. running Linux vs Windows probably isnt much different with an AMD card. the AMD apps are pretty much the same between them.

if you don't mind moving to Nvidia cards, I would recommend something relatively modern (Volta/Turing architecture or newer with at least 8GB VRAM). and then run under Linux so you can run the available CUDA application.

_________________________________________________________________________

Comment viewing options

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