Putting all Your CPU threads to Use

[Gorgon]

I have a 2700x in a Linux System that has two NVidia GPUs installed. The GPUs each need a thread to keep them fed leaving me with 14 threads available for CPU folding.

 

However, if I try to assign all 14 threads for a single CPU "Slot" when a Work Unit (WU) starts I get first a message that the number of threads has been decreased to 13 since 14 is a multiple of 7 a "Large Prime" and then a message saying the threads have been further reduced to 12 as 13 is another "Large Prime".

 

There is a bug in the A7 CPU Core where CPU Slots where the number of threads are a Large Prime or a multiple of a Large Prime will error out on Core A7 WUs due to issues in the Gromacs libraries as detailed here at Folding at Home and on the Folding Support Form.

 

A Large Prime is defined as "above 5" for Folding so being reasonable and only looking at numbers below 128 threads we see CPU slots with 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, 73, 79, 83, 89, 97, 101, 103, 107, 109, 113, and 127 or Multiples thereof will all have this issue.

 

Not wanting to leave any performance on the table instead of letting the FAHClient define the CPU slot as 14 threads as it will using the default thread setting of "-1" I instead created one CPU Slot with 8 threads and a Second with 6 threads thus using all the threads on the Processor. Creating 10 and 4-thread CPU Slots would also work as would 12 and 2-thread CPU Slots and might result in better performance due to the more threaded Slot having a much higher Quick Return Bonus.

 

Note: this system is not used interactively but is managed via the command line and using all your threads on a system that is used interactively using a Graphical User Interface such as X-Windows on Linux or under Windows will create problems so the general rule of thumb is to reserve at least one or two threads for use by the Operating System for interactive systems.