Core26 released to full FAH ?

[Nicolas_orleans]

Hello

I just noticed one of my rig downloaded a new core, Core26, for a 17650 work unit that used to be folding under Core 23. Except the very obvious change from OpenMM 8.0.0 to OpenMM 8.2.0 between those 2 cores, anything else we should know ? Notably, can a trajectory / PRCG mix Core23 and Core26 results ?

Code: Select all

20:16:50:I1:WU1087:*********************** Log Started 2025-01-15T20:16:49Z ***********************
20:16:50:I1:WU1087:*************************** Core26 Folding@home Core ***************************
20:16:50:I1:WU1087: Core: Core26
20:16:50:I1:WU1087: Type: 0x26
20:16:50:I1:WU1087: Version: 8.2.0
20:16:50:I1:WU1087: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
20:16:50:I1:WU1087: Copyright: 2022 foldingathome.org
20:16:50:I1:WU1087: Homepage: https://foldingathome.org/
20:16:50:I1:WU1087: Date: Jan 7 2025
20:16:50:I1:WU1087: Time: 00:35:47
20:16:50:I1:WU1087: Revision: 4f149b599caa4725076ef2de3b47c8d7ce725787
20:16:50:I1:WU1087: Branch: HEAD
20:16:50:I1:WU1087: Compiler: GNU 7.5.0
20:16:50:I1:WU1087: Options: -faligned-new -std=c++14 -fsigned-char -ffunction-sections
20:16:50:I1:WU1087: -fdata-sections -O3 -funroll-loops -fno-pie
20:16:50:I1:WU1087: -DOPENMM_VERSION="\"8.2.0\""
20:16:50:I1:WU1087: Platform: linux 6.8.0-1017-azure
20:16:50:I1:WU1087: Bits: 64
20:16:50:I1:WU1087: Mode: Release
20:16:50:I1:WU1087:Maintainers: John Chodera <john.chodera@choderalab.org> and Peter Eastman
20:16:50:I1:WU1087: <peastman@stanford.edu>
20:16:50:I1:WU1087: Args: -dir CeddI94gO-bI8Ageq9jJUp8DuZbjA1qwPSg2bOEixZc -suffix 01
20:16:50:I1:WU1087: -version 8.3.18 -lifeline 1367 -gpu-vendor nvidia -opencl-platform
20:16:50:I1:WU1087: 0 -opencl-device 0 -cuda-platform 0 -cuda-device 0 -gpu 0
20:16:50:I1:WU1087:************************************ libFAH ************************************
20:16:50:I1:WU1087: Date: Jan 7 2025
20:16:50:I1:WU1087: Time: 00:29:24
20:16:50:I1:WU1087: Revision: c7d2824a47eb025fa8cda8968c7a5e971585d90c
20:16:50:I1:WU1087: Branch: HEAD
20:16:50:I1:WU1087: Compiler: GNU 7.5.0
20:16:50:I1:WU1087: Options: -faligned-new -std=c++11 -fsigned-char -ffunction-sections
20:16:50:I1:WU1087: -fdata-sections -O3 -funroll-loops -fno-pie
20:16:50:I1:WU1087: Platform: linux 6.8.0-1017-azure
20:16:50:I1:WU1087: Bits: 64
20:16:50:I1:WU1087: Mode: Release
20:16:50:I1:WU1087:************************************ CBang *************************************
20:16:50:I1:WU1087: Version: 1.7.2
20:16:50:I1:WU1087: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
20:16:50:I1:WU1087: Org: Cauldron Development LLC
20:16:50:I1:WU1087: Copyright: Cauldron Development LLC, 2003-2024
20:16:50:I1:WU1087: Homepage: https://cauldrondevelopment.com/
20:16:50:I1:WU1087: License: LGPL-2.1-or-later
20:16:50:I1:WU1087: Date: Jan 7 2025
20:16:50:I1:WU1087: Time: 00:28:59
20:16:50:I1:WU1087: Revision: f1cd4c791e8c40a35dcfeab3ab85d910949cc0cb
20:16:50:I1:WU1087: Branch: HEAD
20:16:50:I1:WU1087: Compiler: GNU 7.5.0
20:16:50:I1:WU1087: Options: -faligned-new -std=c++11 -fsigned-char -ffunction-sections
20:16:50:I1:WU1087: -fdata-sections -O3 -funroll-loops -fno-pie -fPIC
20:16:50:I1:WU1087: Platform: linux 6.8.0-1017-azure
20:16:50:I1:WU1087: Bits: 64
20:16:50:I1:WU1087: Mode: Release
20:16:50:I1:WU1087:************************************ System ************************************
20:16:50:I1:WU1087: CPU: Intel(R) Core(TM) i5-3550 CPU @ 3.30GHz
20:16:50:I1:WU1087: CPU ID: GenuineIntel Family 6 Model 58 Stepping 9
20:16:50:I1:WU1087: CPUs: 4
20:16:50:I1:WU1087: Memory: 15.57GiB
20:16:50:I1:WU1087:Free Memory: 7.40GiB
20:16:50:I1:WU1087: OS Version: 6.8
20:16:50:I1:WU1087:Has Battery: false
20:16:50:I1:WU1087: On Battery: false
20:16:50:I1:WU1087: Hostname: amandine-MS-7751
20:16:50:I1:WU1087: UTC Offset: 1
20:16:50:I1:WU1087: PID: 189217
20:16:50:I1:WU1087: CWD: /var/lib/fah-client/work
20:16:50:I1:WU1087: Exec: /var/lib/fah-client/cores/openmm-core-26/centos-7.9.2009-64bit/release/fahcore-26-centos-7.9.2009-64bit-release-8.2.0/FahCore_26
20:16:50:I1:WU1087:************************************ OpenMM ************************************
20:16:50:I1:WU1087: Version: 8.2.0
20:16:50:I1:WU1087:********************************************************************************
20:16:50:I1:WU1087:Project: 17650 (Run 51, Clone 1, Gen 409)
20:16:50:I1:WU1087:Reading tar file core.xml
20:16:50:I1:WU1087:Reading tar file integrator.xml.bz2
20:16:50:I1:WU1087:Reading tar file state.xml.bz2
20:16:50:I1:WU1087:Reading tar file system.xml.bz2
20:16:50:I1:WU1087:Digital signatures verified
20:16:50:I1:WU1087:Folding@home GPU Core26 Folding@home Core
20:16:50:I1:WU1087:Version 8.2

I was more expected a Core24 update due to the communication issues between FAH core and client (information from Justin in one thread in this forum), any news on that ? I still see errors only with P18237 with attempts after a free energy error to resume calculations from last valid checkpoint, when no attempt appears to be actually made based on logs ?

Thanks and best regards

Nicolas

[Nicolas_orleans]

Hello

Complement on OpenMM 8.2.0 based on GitHub. I saw no material difference for my Nvidia card compared to Core23, but it could make a difference for recent AMD cards performance / PPD ?

OpenMM 8.2.0

This release is a major update with a focus on improving performance.

For the first time in many years, this release introduces a new platform. It is based on AMD's HIP framework. When running on recent AMD GPUs, it delivers roughly double performance compared to the OpenCL platform. [...]

source: https://github.com/openmm/openmm/releases

Best regards

Nicolas

[prcowley]

In reading this thread about core 26, I went and had a look to see if I have received any.
Yep, I have been processing some core26 projects for a couple of days.

The main thing I noticed is how short the TPF time is (project 17652) - just 8 seconds or there about. Processing a medium sized project of 40,00 atoms took just 17 minutes on a NIVIDIA 4080 super GPU.

What I found interesting though was that before my successful work unit were multiple failures from other folders. Maybe there needed to be some tweaks to the projects to get them to tun successfully.

Cheers
Pete

[BobWilliams757]

Per ToTow on the Discord channel, HIP support is not yet included in the Core 26 release. I sort of assumed that was the case, since none of the AMD crowd was speaking of the performance increases.

I also go the core update, but really paid to attention to it. It might be worth watching to see if there is any performance improvement with the core either way.

[muziqaz]

We are still assessing if there is any performance increase. Core26 in its current incarnation was not targeting performance increase, main target was core errors and possibility to do more variations of "science".

[muziqaz]

Core24 p18251 - Radeon VII (Windows) - crash
Core26 p18252 - Radeon VII (Windows) - 2m8s
Core24 p18251 - 6900XT (Linux) - 3m0s
Core26 p18252 - 6900XT (Linux) - 1m50s