CUDA Update to FAHCore_22

[psaam0001]

Is there a guide somewhere that would explain some of these extra slot options (at least the ones that are not exclusively for use by the developers/beta testers)?

Paul

[bruce]

Is there a guide somewhere that would explain some of these extra slot options (...?

I'm not sure, but if there is one, it hasn't been updated to include the options discussed above which have not been online for a full week yet. In that regard, we are the best source of such information.

[psaam0001]

I understand Bruce.... There is just so much that is changing with F@H (hardware/software update wise) right now, that the guide would be needing as much TLC as the servers (and client systems) do.

Paul

[PantherX]

Is there a guide somewhere that would explain some of these extra slot options (at least the ones that are not exclusively for use by the developers/beta testers)?...

The second last post has the list of all Slot options and the description: viewtopic.php?f=38&t=26036

When it comes to FahCore_XX flags, there's isn't a single source of truth that's documented. Instead, this Forum can be used for that time period only since the idea is that FahCore_XX should run within any baby-sitting or tweaks. Thus, the flags we provide is only for workarounds until a new version is available which can fix the issue completely, thus, removing the need to document those workarounds.

[Foliant]

Yes. [...] go to the Slots tab, select the slot, Edit, add an "Extra slot option" called "extra-core-args" and set it to "--disable-cuda" and save the settings. This setting will apply to only that slot.

Thanks! Ill try that for the next run. (in 20 hours)


There is another point I noticed after the Core update: Power consumption.
My first thought was maybe very stressful workloads but numbers are staying the same since then.
Compared to openCL my folding rig consumes a little bit more power since using Cuda.
The system is underclocked as before.
On oCL it was folding around 680-705W.
Now, using Cuda, its more like 695-720W overall.

[bruce]

...On oCL it was folding around 680-705W.
Now, using Cuda, its more like 695-720W overall.

I think that 15W is well-spent.

[chabgood]

It would be more intuitive if the options were available as a drop down list.

[psaam0001]

The second last post has the list of all Slot options and the description: viewtopic.php?f=38&t=26036

When it comes to FahCore_XX flags, there's isn't a single source of truth that's documented. Instead, this Forum can be used for that time period only since the idea is that FahCore_XX should run within any baby-sitting or tweaks. Thus, the flags we provide is only for workarounds until a new version is available which can fix the issue completely, thus, removing the need to document those workarounds.

TYVM for that insight.... I was just curious as to what a few of the other "work around" flags were at the moment, and what they did. However, I'm just going to use foldy's help to get me through the next day or two on the Windows 7 system, till the new HD arrives, and I can give it a new OS.

Paul (off topic on this thread for the last time)

[road-runner]

I have a windows 10 rig with 2-1080 FE I leave run year round till winter then crank up the 1080TI. I noticed PPD went from around 800k each card to 1.2k each I have not done anything...

[PantherX]

800K for 1080 Ti is low. On OpenCL, 1080 Ti is between 1.3 to 1.6 Million. On CUDA, it hovers between 2.3 to 2.6 Million

[HaloJones]

800K for 1080 Ti is low. On OpenCL, 1080 Ti is between 1.3 to 1.6 Million. On CUDA, it hovers between 2.3 to 2.6 Million

I think he's referring to the performance on the 1080FE not the TI.

For reference, with the 149xx units my 1070's were getting over 1.2m, have just dropped back with these new 1730x units.

[road-runner]

Yes these are the very first founders edition when 1080 came out. I have 6 - 1080TI thats is my winter heat have not turned them on yet..

[psaam0001]

Happily got 2/2 NVIDIA carded Fedora 32 systems running on the CUDA cores. Using driver version 450.80.02.

Paul

[road-runner]

Ran a 1080 TI last night just to see what it would do..

Min. Time / Frame : 00:01:01 - 2,529,052.0 PPD
Avg. Time / Frame : 00:01:02 - 2,468,110.3 PPD

[gunnarre]

Good, that's what you can expect for a non-Moonshot work unit on CUDA. I've seen Moonshot WUs get 3M PPD on the 1080 Ti.