AMD HIP API support is coming near you

[toTOW]

The client will be fixed in the next version:

https://github.com/CauldronDevelopmentL ... c6f21314d7
https://github.com/FoldingAtHome/fah-cl ... 4ef8b7037f

There's also a bug in AS code for v8 clients that don't cast the available memory value received correctly ... this has been holding back the release of huge projects for more than a year.

[FaaR]

It would seem that the newest AMD Adrenaline driver no longer includes the ROCm v6 runtime library, is this going to cause any issues with folding@home?

Thanks!

[muziqaz]

Yes, and some heads rolling within AMD.

[muziqaz]

Ok, AMD knows about this. Hopefully will rectify this with next driver release

[CaptainHalon]

Any update on HIP and Core28?

[muziqaz]

Any update on HIP and Core28?

Still waiting for the core to be rebuilt with latest patches

[maciej_ficek]

"No support for Polaris or older" - You mean no support for OpenCore28 but they will continue to get OpenCore<=27 tasks, or end of support at all, so the oldest AMD GPUs working in FAH would be RDNA1 GPUs from 2019?

[muziqaz]

"No support for Polaris or older" - You mean no support for OpenCore28 but they will continue to get OpenCore<=27 tasks, or end of support at all, so the oldest AMD GPUs working in FAH would be RDNA1 GPUs from 2019?

This is just for core28 hip mode. core28 opencl will work on everything

[CaptainHalon]

Any update on HIP and Core28?

Still waiting for the core to be rebuilt with latest patches

Cool, thanks for the update. Is it still going to require 64GB of RAM to get a WU?

[Joe_H]

That requirement for 64 GB of RAM is just for the first project being worked on to use Core_28. It is very large. My assumption is that there will be smaller projects at some point in the future that will not require that much RAM.

Edit - found the info, very large is 7.9 million atoms in the simulation.

[maciej_ficek]

Can you run AMD and nVidia side by side in the same machine, or is that a bad idea? I may be misremembering but I seem to recall that this was not recommended in the past, I'm curious if this changes anything in that regard.

One of my nVidia 1660 Super cards died this past winter; I haven't gotten around to replacing it yet but I'm wondering if this expands the range of possible candidates. This Linux Mint machine still has one 1660 Super in it & I'll be looking at adding another card in the $200-300 range.

Edit:

And the first project will require 40GB of free RAM on the system and a high end GPU (which includes 7900xtx).

Well, maybe I'm jumping the gun a bit at my price range...

I run simultaneously RTX 2000 Ada and Radeon Pro W7500; both in two separate eGPUs via thunderbolt. Practically, I do not notice any decrease in PPD for any of these two cards when I run them both simultaneously; I also do not notice bigger temperatures. Sometimes the tasks crash after downloads, but this happens for both the cards even when I run only one of them.

[maciej_ficek]

Can we donate some money in any way to speed up the HIP/ROCm implementation? I think many people waited for that for years. It would move humanity forward a lot.

[muziqaz]

Can we donate some money in any way to speed up the HIP/ROCm implementation? I think many people waited for that for years. It would move humanity forward a lot.

I don't think we can afford to buy out another developer full time