Dell Precision M6700 Owners Thread

I've bought items from China many times... I've never had an issue with it showing up, but sometimes it takes 3-5 weeks. I would buy domestic if possible just for the shipping speed but sometimes China is just the most economical option. (Returning the item can be a pain though. I've only done that once.)

 

Can't say much about returning (since the card is working flawlessly), but it took 11 days to arrive and it's working properly.

It won't come with a X-bracket, so be aware of that.

 

I was speaking to nforce4max

The first proper way of testing vram is occt gpu memtest.
Filling the vram usage isn't enough, sometime GDDRx can handle some error (it's ECC memory after all), furthermore vram usage is far different from the actual vram used.

 

I believe that I found the culprit and it looks to be of all things a fan controller chip under the mxm but beyond that absolutely everything else survived. Swapped the board and all is good thus far.

 

Mobile Quadro does not have ECC memory like desktop Quadro does.

 

[Edit note: I have just now learned that not all of the mobile Quadro cards after the 5000M have ECC RAM, and I have changed the following wording accordingly.]

Actually, starting with the Quadro 5000M, at least the highest-end mobile Quadro cards have ECC RAM. By going to nVIDIA Control Panel, then to Workstation, and then to Change ECC state, one can toggle the ECC state off and on. With the 5000M, for example, when ECC is off, the card's RAM shows as 2 GB, and when ECC is on, the RAM shows as 1.75 GB, of course because 0.25 GB is being used for error correction.

 

Hmm, I was previously misinformed then. I’ll check it out… thanks.

 

You are welcome.

I also was in error in my understanding, though. I have just now learned that not all of the mobile Quadro cards after the 5000M have ECC RAM, and I have changed the wording in my earlier post in this thread accordingly.

 

Yeah, I have systems here with a Quadro M5000M and RTX A2000, and neither of them is showing the option to manage the ECC state in NVIDIA control panel. Both the "Manage ECC state" and "Manage GPU utilization" options are missing. Under "Workstation" there is just "View system topology" and "Set up Mosiac".

 

Front what I recall GDDR5 memory technology does have some sort of "ECC" (more like error detection) build itself, that's a feature of GDDR5 and newer (might be "GDDR1" and newer but I don't remember). That way, whenever the memory controller detect an error, it ask the memory chip another transmission of the data or it correct the error if it's not that big. That's explain why in certain "too high" overclocking the performance can decrease due to memory bandwidth starvation because of too much retransmission. Also it explain why sometime you are able to use the card for a bit then get a black screen or BSOD out of nowhere.