Milestones VI

I am pleased to note that I have a system that has passed the 9.5M RAC mark. And has achieved the highest ranking I have ever achieved on this project.

Tom M

I'm not sure if this is a good place for this or not, but I saw this in the BOINC website which has the Top 100 Multi-Project BOINC Participants.  I was quite surprised to find myself and two of my teammates on this list.

If you're curious if you are on the list as well, you can find it at:  https://boinc.berkeley.edu/chart_list.php

Tigers_Dave wrote:

Lurking on the forums, it appears that Mac GPUs and eGPUs are much less productive than equivalent PC GPUs.

you'll have to be more specific about what a "Mac GPU" and "eGPU" actually are. and what subproject specifically you're running. since your computers are hidden, we cannot make any real suggestions.

 

the only Apple GPUs are those present in the M1 processors, but they are Metal only and do not support OpenCL, so they cannot be used for any BOINC project. If "Mac GPU" means "an AIC GPU in a Mac", then you'll need to elaborate on which GPU you have specifically, Nvidia or AMD, and which model(s). 

same with the eGPU(s). on it's face, the reduced bandwidth of a GPU in an external enclosure shouldnt matter much for Einstein. but you'll need to expand on which GPUs are in those enclosures and their specific setups.

 

Einstein has two different subprojects, Gamma Ray (FGRPB1G) and Gravitational Wave (O3AS). the gamma ray tasks run faster and award a lot more credit/unit-time.

The gravitational wave app is slower, and more CPU bound, fast GPUs are easily bottlenecked by a slow CPU.

 

With more recent advancements and improvements in the Gamma Ray Nvidia application, Nvidia cards generally outperform AMD cards for Einstein. but to get the faster application you need to make sure you are running compatible GPUs with recent drivers (Maxwell and newer, and Driver 470 and newer). If you are running older GPUs and/or older drivers, you will still get the older/slower application. upgrading the drivers to latest will allow the newer app to show up (v1.28) and 40-100% boost in GPU production (YMMV, depends on model). this applies to gamma ray only, the gravity wave app hasn't been further optimized.

You might be out of luck for Apple/Nvidia due to the huge driver issues/headaches and Apple’s general lack of support for Nvidia, at least recently. It seems they’ve abandoned each other. Your 680 card is too old to use the new faster app, but your 980s potentially could. But again the problem is drivers on Mac. the new application uses (and hence requires) features only available in OpenCL 2.0 and newer. Historically, Nvidia drivers only included legacy OpenCL 1.2 support. But starting with their 465 driver branch and newer, Nvidia bumped that to OpenCL 3.0. The project also will not send you the new application unless they detect that you have compatible drivers installed that report at least OpenCL 2.0. 
 

I’m not super familiar with mac/Nvidia drivers since I don’t run macs, but a quick Google search shows the latest drivers as some CUDA 10.1 drivers released in 2019. This is well before change to OpenCL 3.0 in Spring 2021. Unless there are newer drivers available to get the latest features, there’s probably nothing you can do outside of changing to a Linux or Windows OS which has better Nvidia driver support. 
 

nothing you need/can do about the AMD cards, just keep chugging along. 
 

One thing you could try on all your systems (if you’re not already) is to try running multiple tasks to see if you get a little better production. Try 2-3 at a time on each GPU. The each task will run slower, but shouldn’t be slower than the multiple, leading to a slight boost in overall output.