Page 4 of 5
Re: Core26 released to full FAH ?
Posted: Fri May 09, 2025 9:08 pm
by Frogging101
Thanks. Although ROCm is a pain in the butt. rusticl is the first thing I managed to get working on this GPU.
Re: Core26 released to full FAH ?
Posted: Fri May 09, 2025 9:39 pm
by muziqaz
rocm for fah is not the pain at all. All of my AMD GPUs are under rocm. All work fine
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 1:21 am
by arisu
boboviz wrote: ↑Fri May 09, 2025 12:38 pm
2 days ago OpenMM released the beta of OpenMM 8.3
There are brand new beta projects that are still using OpenMM 8.0 and 8.1 so there doesn't seem to be a big rush to upgrade. Kind of a shame because 8.2 added some significant performance improvements.
Does 8.3 come with any major performance improvements? I can't find much about it. But there are really just three things I want to see OpenMM support:
- Dynamically adjusting number of kernel threads based on memory-to-compute ratio (more threads hides memory latency)
- Dynamic parallelism (launching kernels from other kernels instead of the CPU)
- Support for multiple simulations in a single OpenMM context to avoid the (already very low) overhead of MPS, but more importantly to allow parallel simulations on Windows, not just Linux
Dynamic parallelism would be amazing. We could fold on a 5090 over one PCIe 3.0 lane without slowdown, we'd would no longer have to reserve a CPU thread for each GPU, and the Windows and Linux performance would be identical!
Unless 8.3 is closer to acheiving any of those, I'm not gonna be too excited.

Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 1:26 am
by calxalot
I think a future core is being built with openmm 8.2.1
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 1:38 am
by arisu
calxalot wrote: ↑Sat May 10, 2025 1:26 am
I think a future core is being built with openmm 8.2.1
8.2.1 doesn't exist. The latest commit changes 8.2.0 to 8.3.0 directly:
Code: Select all
commit 371f3a3bbf66e622281c825cb9b073402db74d65 (HEAD -> master, origin/master, origin/HEAD)
Author: Peter Eastman <peastman@stanford.edu>
Date: Mon May 5 12:23:56 2025 -0700
Update version number to 8.3 (#4925)
diff --git a/CMakeLists.txt b/CMakeLists.txt
index f7f6ffce4..661050259 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -156,7 +156,7 @@ ENDIF (NOT CMAKE_CXX_FLAGS_RELEASE)
SET(OPENMM_LIBRARY_NAME OpenMM)
SET(OPENMM_MAJOR_VERSION 8)
-SET(OPENMM_MINOR_VERSION 2)
+SET(OPENMM_MINOR_VERSION 3)
SET(OPENMM_BUILD_VERSION 0)
ADD_DEFINITIONS(-DOPENMM_LIBRARY_NAME=${OPENMM_LIBRARY_NAME}
Maybe what you remember is some internal FAH version? I wonder if that is why core27 calls itself version 8.2.1 when it's built with OpenMM 8.2.0.
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 2:43 am
by calxalot
The major and minor are from openmm. The patch version belongs to the core.
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 5:16 am
by arisu
I wish they would at least release a changelog even if they don't release the actual source code. So the new core will have the same version that core27 has?
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 6:10 am
by calxalot
I don’t expect the version to be the same. Core 27 comes from the same repo as 26.
Maybe muziqaz knows.
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 7:09 am
by muziqaz
Core26 and core27 are both 8.2
8.3 will not going to be used for quite a while.
FAH upgrades to new OpenMM versions only if it fixes something very important, or if it introduces some functionalities which are needed by researchers, or if they bring support for some other APIs.
I believe OpenMM posts their release notes.
FAHcore change notes are not that important for end user
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 8:27 am
by arisu
It's too bad the change notes aren't at least released to beta testers. They used to be released, back when FAH was a Stanford project.
8.3 doesn't seem to come with any very major improvements it looks like. Just getting HIP into the cores would be a better use of dev time than rebasing on 8.3.
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 5:55 pm
by boboviz
arisu wrote: ↑Sat May 10, 2025 1:21 amUnless 8.3 is closer to acheiving any of those, I'm not gonna be too excited.
I'm not excited, i only report the news.
And i hope that new versions will help to debug/fix the problems...
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 5:57 pm
by boboviz
muziqaz wrote: ↑Sat May 10, 2025 7:09 am
Core26 and core27 are both 8.2
8.3 will not going to be used for quite a while.
That's for sure
Re: Core26 released to full FAH ?
Posted: Sat May 10, 2025 5:59 pm
by boboviz
arisu wrote: ↑Sat May 10, 2025 8:27 am8.3 doesn't seem to come with any very major improvements it looks like. Just getting HIP into the cores would be a better use of dev time than rebasing on 8.3.
Meantime i'm crunching Core 23

Re: Core26 released to full FAH ?
Posted: Mon May 19, 2025 10:11 pm
by teaough
Every time I get a core-26 or core-27 work unit my system runs for 1 second and then dumps the WU.
AMD Ryzen 5/NVidia 4060
Submitting to forum as instructed... but I'm not certain that this is the correct thread for this.
Thanks.
21:55:13:I3:Running FahCore: /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 -dir AXgevWMTa4yTRRP1WX1D519FIyNV-Fq6u6HvqomGJXs -suffix 01 -version 8.4.9 -lifeline 1789 -gpu-uuid 4a6400bf-af2c-e395-2010-1f7e749e1a75 -gpu-platform cuda -gpu-vendor nvidia -cuda-platform 0 -cuda-device 0
21:55:13:I3:WU1631:Started FahCore on PID 81474
21:55:14:E :WU1631:Core returned WU_STALLED (127)
21:55:14:E :WU1631:The folding core did not produce any log output. This indicates that the core is not functional on your system. Check for missing libraries or GPU drivers. Make a post about your issue on
https://foldingforum.org/ to get more help.
21:55:14:E :WU1631:Run did not produce any results. Dumping WU
Re: Core26 released to full FAH ?
Posted: Mon May 19, 2025 10:13 pm
by muziqaz
Install icd loader