UNSTABLE_MACHINE CoreStatus = 7A (122) out of nowhere

[7im]

Just to be clear, when you say "other client" is running okay, you mean the GPU2 client, right? And only GPU3 is failing, and you only have ONE instance of the GPU3 client installed, correct? Let's try to start eliminating some of the possible variables that could be causing the problem.

While the GPU3 client not running "symptom" makes it appear that the GPU3 client is "cocked up" we really haven't seen a similar issue reported by anyone else. That lack of cocked up reports "symptom" suggests is may not be the client, so then we need to look at all possible causes. And you said yourself that the client did work for a while before starting to throw up errors. Another symptom suggesting the client works, and something else has failed.

[ElectricVehicle]

In theory, your power supply has enough headroom for a single GPU (even a really big one like the 480). In practice, splitting the 900W between 6 rails is bad news for a really big video card because that means it can only deliver a maximum of 250W per rail. You really are much better off with a quality single-rail power supply.

Single rail power supplies can smoke very quickly when there's a short, thought that doesn't happen very often. You can have troubles with connector melting if you deliver too much power, like the entire powr supply rated power through a single connector.

With split rails, you need to make sure to identify and allocate the power limited rails properly.

PCI Express motherboard slots provide a maximum of 75 watts
The 6 pin PCI Express power cable provides up to 75 Watts (6.25A).
The 8 pin PCI Express power cable provides up to 150 Watts (12.5A).

Since I only have one GPU in my system at the moment, and my power supply has 4 rails, I dedicated 2 rails to the GPU - 1 for the 6 pin PCI Express and the other for the 8 pin PCI Express power cable. Each rail is rated to 25A (300 Wattts) on mine.

Quality of the power supply is more important than split vs. single rail. Efficiency is also important in many applications since it reduces power cost, heat generation and leaves more of the power availablein a 15A or 20A 120V branch circuit that feeds the outlet the power supply is plugged into for other purposes.

Update: Rats! I forgot to add the link to the site where I found the best explanation of PCI Express power information, and many other power supply tidbits:

All about the various PC power supply cables and connectors

[7im]

EA makes a good point. And while he only described load balancing of the split rails, I think it needs to be said straight out. Load balance your split rails. Share the load, literally.

[fiCaks]

Just to be clear, when you say "other client" is running okay, you mean the GPU2 client, right? And only GPU3 is failing, and you only have ONE instance of the GPU3 client installed, correct? Let's try to start eliminating some of the possible variables that could be causing the problem.

Both are A3 clients - 1st client set up and then copy of it made just changing in configuration the machine ID to be able to run them both simultaneously.
One of them folded for some time till started to throw errors. So, fired up the other one. Surprisingly it folds w/o any complaints for last 13hours.
This strange difference of 2 similar clients made me create the thread, not h/w issue. Before even creating it I stressed the system to maximum(Furmark, OCCT, Memtest, PSU stress test, physical test of PSU) to make sure there's no soil for accusations of me claiming s/w fault while my h/w is "broken" or pushed over limits.
Please focus on fact that "copy of client" is happily folding while the "original" won't return anything but errors.

[PantherX]

I don't understand why you would want to run 2 GPU3 Clients when you only have 1 GTX 480. Is there a reason for doing that?

[fiCaks]

I created them to check whether there would be PPD gain whilst running 2-on-1 like it was with Core_14 units. After analyzing results of 1st 2 WU submitted, came to fact that the trick does not work with Core_15. So, I'm running just 1 instance of A3 at the time since then.

[fiCaks]

Sorry for stearing the water!
Just deleted client.cfg and went through -setuponly for fresh for "original"(failing) client. After that it started folding again. Hurray!
However, two lines in client.cfg, created by client, are differing now:
local=1156
local=1
In "reanimated" previous client these are missing. Before I deleted client.cfg, they were stating local=1768 local=1
Would it be that client over the time is creating those lines, first of which reaching certain limit is blocking client from running?

[Leonardo]

-

gpu of both clients is -gpu 0

Each GPU client startup shortcut, whether pointing to multiple GPUs on the same card or multiple GPUs on separate cards, must have a different -gpu flag number. Enter the flag "gpu 0" for one client and "gpu 1" for the second.

[fiCaks]

Each GPU client startup shortcut, whether pointing to multiple GPUs on the same card or multiple GPUs on separate cards, must have a different -gpu flag number. Enter the flag "gpu 0" for one client and "gpu 1" for the second.

Thanks for outstanding load of help! Is anyone else thinking that i've reached 8m points mark in less than 1.5 year time, operating just a single client w/o a clue how to configure it or run/configure multiple clients?

[7im]

Just deleted client.cfg and went through -setuponly for fresh for "original"(failing) client. After that it started folding again. Hurray!

Great news. I knew it has to be something local, otherwise we'd get 200 more of the same complaints in the forum.

Would it be that client over the time is creating those lines, first of which reaching certain limit is blocking client from running?

No, local just counts the number of WUs uploaded since the client was installed. There is no limit. Sounds like the problem was a corrupted configuration, and deleting it and recreating it fixed it.

[fiCaks]

Great!
7im, Can this thread be deleted to stop h/w trolls coming after me

[7im]

I'm sure they are well intentioned. Maybe having a Mod lock the thread is sufficient. It helps to have threads like this around the next time the error pops up, even if it went off topic a little.

This is your thread, feel free to do with it as you please.

[fiCaks]

WOW! Thanks for usefull hint - never new it. Although, if you say this might become usefull to someone, I shall not use my rightfull powers

[Leonardo]

I'm glad you got the problem fixed.

[ElectricVehicle]

Just deleted client.cfg and went through -setuponly for fresh for "original"(failing) client. After that it started folding again. Hurray!

Great news. I knew it has to be something local, otherwise we'd get 200 more of the same complaints in the forum.

Would it be that client over the time is creating those lines, first of which reaching certain limit is blocking client from running?

No, local just counts the number of WUs uploaded since the client was installed. There is no limit. Sounds like the problem was a corrupted configuration, and deleting it and recreating it fixed it.

It's unknown at present, but possible this thread I started may be related to improper formatting of the client configuration, which the client seems to do all by itself using the Configure option from systray: GPU3 systray client can't enter passkey properly this hasn't been reported / re-created by a second party yet, but there is something definitely different about GPU3, Passkey and client.cfg that's odd at the least wihich we haven't gotten to the bottom of yet.

Specifically the local line in GPU3 client.cfg looks ok at the moment, showing the single occurrence of the local= line:
local=48