Core return BAD_FRAME_CHECKSUM 112

[PhilTheNet]

Hello,

I have encountered an error (BAD_FRAME_CHECKSUM, 112) and the WU goes into DUMP mode when restarting the FAH client after an unexpected shutdown.

Is there a way to recover the WU to restart its calculation without DUMP?

Thks

[muziqaz]

Not from these errors.
Also, check your GPU, it probably is unstable. Or the system RAM

[PhilTheNet]

Ok,
Is the checksum calculated by the core or the client?

[muziqaz]

Ok,
Is the checksum calculated by the core or the client?

Core, I believe

[PhilTheNet]

Too bad, I can modify the client to fix the problem, but I don't have access to the source code for the core, which is a shame.

[muziqaz]

Too bad, I can modify the client to fix the problem, but I don't have access to the source code for the core, which is a shame.

You cannot fix your hardware instability with core tweak

[PhilTheNet]

This is not due to hardware instability but to the fact that use shuts down the machine. I would have been interested in adding this recovery feature, but if we don't have access to the core source code, it's not possible.
How to find out how the core calculates this famous chesum...

[muziqaz]

This is not due to hardware instability but to the fact that use shuts down the machine. I would have been interested in adding this recovery feature, but if we don't have access to the core source code, it's not possible.
How to find out how the core calculates this famous chesum...

I feel that you are trying to come up with the solution to overstep your system instability.
Look at why the checksum is failing, not how to recover from it.
There are reasons why WU is dumped after this

[PhilTheNet]

You don't understand my approach, I'll try to figure it out on my own.
Thank you for responding.

[FaaR]

You don't understand my approach, I'll try to figure it out on my own.

Your only recourse is to not shut down your computer while the client is running. You must shut down folding first, since the exact GPU state isn't saved when you shut down the OS, so if you have the client still running when you do that, data will be lost. There's no way to get around that, due to the way GPUs and the driver model running them currently function. There's many many megabytes of data flowing through a big GPU at any one time (AMD Vega64 has over 40MB of SRAM total on the chip in caches, registers and various buffers, and that's an old design today); it's not like CPUs where you have maybe a few dozen registers per core and a flush of the write buffers to keep track of for a state change.

So, don't shut down the computer while the client is running. You could say, "just resume from a previous checkpoint", but you have to realize that to the folding client, abruptly shutting everything down mid-folding looks just like a fault with either the hardware or the work unit itself, and since accuracy and reliability is the most important thing with folding (because actual science relies on these results; F@H is not a competition to gather the most points ), they will take the safer choice and just throw out the work unit completely it would seem.

So don't shut down your computer while folding is running. That is the solution you are looking for!

[prcowley]

What version of FAHClient are you using?

Most of the clients have a way to either pause or wait until the current work Unit has completed before pausing. (at Least versions 8.xx have it)
Once the client is paused or completed then you can safely shut down the computer.

Cheers

[muziqaz]

V8 saves the work progress fine on the stable systems upon restart.
If the system is not stable nothing you change with the client gonna fix the issue. It will be just a lottery and extra work for the user.