P17106 finished in 10m flat (personal record)

[PaulTV]

Hola,

Today I got a job for P17106 which took about 10m to solve. It's a nice personal record, but I guess this project may incorrectly be assigned to high-end cards.

https://apps.foldingathome.org/wu#proje ... =180&gen=0
Project: 17106
Run: 13
Clone: 180
Gen: 0
Slot: gpu:6:0 GA104 [GeForce RTX 3070]
Work server: 54.157.202.86
Collection server: 140.163.4.210
Assigned: 2021-08-07 19:08:46
Logged: 2021-08-07 19:20:28
Credited: 2021-08-07 19:24:24
Credits: 13.587

Cheers,
Paul

[Neil-B]

No it isn't a mistake ... it is a test/benchmarking wu for the new core being issued to all gpus running windows ... deliberately short/quick so as not to waste resource ... the project summary (follow link at to of forum) explains

[PaulTV]

Ah, okay. It still ran on core 0x22 for me, but I suppose that's to compare the performance of the new core.

[Neil-B]

Check the log .. 0x22 but now version 0.0.15 ... or at least should be

[Neil-B]

You should also see the core download and installation

[PaulTV]

That's an affirmative, on both version and core download. Cool!

[Neil-B]

This version resolves an issue that afflicted core 0.0.14 that was released with the moonshot 13456/7 .. further work is ongoing I believe but this needed to be pushed out

[JohnChodera]

Thanks to everyone who helped us test the new core22 0.0.15! Things appear to be working well so far, with three issues noted that we're working on addressing in subsequent updates:

* A number of failures occurred because the client attempted to assign an AMD CPU with AMD-APP OpenCL drivers installed as the GPU. We're working on a workaround to avoid this.
* In this test project, we permitted all GPUs (including all Intel GPUs) to run the test. A number failed (such as the Intel Xe) because they don't support any double-precision operations. We're working on excluding these GPUs.
* We are still building with CUDA 10, so the latest RTX 30x0 cards will fall back to OpenCL. We hope to rush an update out next week that will use CUDA 11.2 to enable RTX 30x0 cards to use CUDA, getting a nice performance boost.

Overall, things look good for core22 0.0.15, and we'll deploy it shortly!

Thanks again!

~ John Chodera // MSKCC

[vivatobst]

Wait - Do I have to manually download new core versions? Just checked and still running on 0.0.13

EDIT: Nevermind. Just read it's still a BETA thing

[PaulTV]

Wait - Do I have to manually download new core versions? Just checked and still running on 0.0.13

From what I understand from my logs, nope you don't. After that p17106 job, I got one for another project, that ran on 0.0.13. So I gather that the required core sub version is specified in the job info, and if that version isn't installed already, the client will get it automatically. I certainly didn't do anything manually to run that p17106 on 0.0.15.

Specifying the core sub version in the job info would give the possibility to test newer core sub versions on specific projects without affecting all other projects using that core type, and do a phased roll-out of the new sub version.

[JohnChodera]

> Wait - Do I have to manually download new core versions? Just checked and still running on 0.0.13

Cores are automatically downloaded when needed by the project, but recent infrastructure updates now let us ensure that projects can set which core version they use for reproducibility and reliability reasons!

~ John Chodera // MSKCC