FWIW:

I'm seeing quite a few failures on the VINA cores as well. In fact the only invalid results I have at all are all VINA cores. Some from my Galaxy S4 which I'm not fused about, it's not mature technology.

My bigger concern is on my 48 Thread machine, its had 100% success on all other units but is generating INVALID VINA results all over the shop.

I'm not going to spend time changing hardware and swapping components when everything else runs fine except VINA. However as already stated my two Intel 3930K Systems have no INVALID results for any VINA unit. I'm not sure this is coincidence as 48 threads vs 24 threads is quite a bit of work being done and over a week is quite a decent sample size.

I've just disabled FAAH on my 48 thread machine and called it a day.

We have been continuing to investigate and have found some infrequent situations where two results diverge slightly that gets them marked invalid. This is causing us to go through recent results that have been marked invalid and identify examples that we can share with the researchers and see if we can modify the validation logic to accept a wider range of answers as the "same". This process will take a some time, but when we have an answer we will update this thread.

KerSamson you are one of the solid contributors for Switzerland. I hope you stay even if times are difficult. I also had in the past some frustrations with projects thast could have a very high erroring rate 30% or more. They did error in the first seconds but still it was causing disruption because I would hit the limit for the daily number of WUs. But very strangely even if my machines are more or less identical some would generate a lot of errors and other not a single one. I had this with varioius projects.
After a lot of efforts I was unable to find out why some machine error highly on some projects and others not. And this was with Windows based systems. So I gave up to try to find the cause.
My solution was to have all machines run a new project and then just select the suitable ones for a given project. Now with very little projects and few machines I can understand it may become an issue.
For a solid contributor like you it would be fully understandable that you put to rest your machines for a few months and come back when conditions seem more suitable as I am doing at the moment. And it will also impact favourably your wallet

Hi guys,
September and October were fine with only few troubles.
Since end of October, some of my Linux hosts earned again a lot (>300) of invalid WUs, in particular with FA@H.
The problem I mentioned during the Summer is still present from time to time.
The high numbers of invalid WU (i.e. successfully computed but considered as invalid by the validator) is boring, frustrating, and finally not acceptable.
I was considering to add or replace some hosts but finally I do not know what I should do, since the mentioned problem is impacting systematically Linux-based AMD systems, and I would like to remain on this particular platform.
On 2013-08-08, knreed promised to investigate the problem.
I did not listen anything since this date.

I also noticed during the last 3 weeks, that Linux-based hosts - working on FA@H - did stop to compute for unexpected reasons. This problem occurred several time during the Summer and finally "disappear". Since beginning of November, this trouble occurred again several times.
I collected the error files.

I would very appreciate to receive some founded answers. I can be contacted directly by the WCG-Tech-Team, for providing more detailed information (e.g. error and statistic files).

In my business work, I have to investigate until root causes are identified. Because of the impact of the above troubles, I think that it is time to look for the particular root causes.

Cheers,
Yves
---
@knreed: JM Boullier has my direct contact data.

maybe try a different linux distro.