This is gonna be a HARD one - Gtx 970

How many sticks of ram do you have?

 

I check it and it says its OFF.

 

I have 4 sticks and in the right side if i remember right is 2 sticks that are 4 GB each and the other side 2 are from my old PC and they are 2Gb each so, 2 ram of 2GB, and 2 ram of 4 GB. in total 12 GB

 

Post a cpuz like you were asked earlier, pics of all tabs.

 

http://www.nvidia.com/object/IO_89570.html

Error Correction has been in place since Fermi cards... read Page 17. If the card detects an error in memory or cache. It will request to resend it over the PCI-e bus. Resending the data would cause hitching or stuttering.

 

Here, sorry it took me a while.

I hope u can see it, i put all in one photo.

http://postimg.org/image/fdcl3swlv/

 

Yes.... it would wouldn't it.... hmm.

This is extremely interesting to me. Thank you for posting this sir.

Would be very nice to have an application that could check current VRAM bandwidth AND check for errors at the same time. I assume even looping the current VRAM bandwidth checker (think 970 VRAM debacle) batch file style would put some stress on the VRAM.

Wondering if there is a way for us to see WHEN an error happens and ECC is used. That would be extremely useful. I am positive no program currently exists that can do this however I am thinking it can be done with CUDA or OpenCL. Without a doubt a stress test can be made but we'd need an ECC Error log.

I wonder if CUDA dev tools maybe have this already somewhere.

**Edit**

"All NVIDIA GPUs include support for the PCI Express standard
for CRC check with retry at the data link layer. Fermi also supports the similar GDDR5 standard
for CRC check with retry (aka “EDC”) during transmission of data across the memory bus."

<-- Monkeys Uncle.

No. Sorry. You need a K series CPU for that. You might be able to get a few MHz out of it via a BCLK overclock however this usually only ends you up with an unstable system.

 

I see, thank you. I will then not try it.

 

No problem man. We will figure it out soon enough. Make sure you keep a positive attitude.

 

That sounds good, and thanks i will try and btw HAPPY NEW YEAR TO ALL OF YOU AND THANK YOU FOR TRYING YOUR BEST I REALLY APPRECIATED :hatty:

 

I'm personally not knowledgable enough to know if this affects gaming and if it has any relation to overclock instanbility, but interesting info, thx.

 

It without a doubt does Ganja. Read the white paper and take it as literally as you can. A CRC retry takes time and that would result in the same command being sent twice.

This is why since Fermi a lot of the time you won't even see artifacts with unstable memory, it will just work slower and error correct or just plain crash with a black screen. Older cards GT200 & lower did not act in this manner. From what I remember you could always visually see the VRAM being unstable from artifacts.

 

Yep. seen it first hand myself, error detection on gddr5 when pushing vram too far. Mentioned that to the op, though don`t know if he understood tbh.

 

Actually I did some reading, and it seems ECC doesn't exist on most gaming GPU's, certainly not the 970/980/980ti's.

From what I understand, only quadro tesla and titan have ECC?

So I don't think ECC would be the cause of this.

 

Is there a way to me to check if there is a error detection on gddr5, ECC, CRC?

 

Try running Heaven 4.0 custom preset everything maxed @1080p, your GPU at those clocks should give a score in the 1500 range. CPU wont really be a factor in that test.

 

Why dont you put your 970 back to stock clocks, it is not a good idea just to whack in numbers and hope for the best. Most def lower that 8ghz vram clock speed.

 

he still did not try stock clocks?:bang:
you told him that 3 days ago

 

Not that I read? vram clocked too high is the likely culprit with gddr5 error detection, maybe even too high gpu core as well.

 

GDDR5 memory controllers can perform basic error detection on both reads and writes by implementing a CRC-8 hash function
run stock gpu and cpu clocks and if that don't work take the 2 2g gig sticks put of your mb
then tell us whats up