Originally Posted by Falkentyne
I posted this in the blackscreen thread, but hopefully someone with read points and a DMM can find out just what is going on with the memory voltage, and this should be posted over here, too.
I may have some information which may or most likely, may NOT be related to people having black screens at STOCK, but should explain some things about black screens when overclocking the memory, or possibly solving them by either underclocking the memory or raising the core voltage.
I know, from testing, that my card (Hynix) runs stable at 1400 MHz memory with 0.961v in idle (2d/desktop). That's +0mv (default)
I also know that 1450 MHz and +0mv seems game stable but may trigger a completely random black screen at idle, when any GPU usage causes the clocks to go to 1450 mhz.
I also know that 1500 MHz +0mv will **ALWAYS** instantly trigger a black screen on the desktop as soon as any load whatsoever causes the memory clocks to rise--something as simple is putting the mouse over a desktop icon will cause this.
It seems like the black screen is simply a GPU shutdown from an error fault (which can be anything from VRM, too high GPU temps, memory issue, etc), but an unstable GPU (from overclocking) will either VPU recover (crash), or artifact.
It also SEEMS that the black screen from a memory related fault ONLY happens during a switch from 150 MHz to boost speeds. There could be other causes, as I doubt Battlefield 4 would ever cause the memory to downclock, but maybe the current core voltage drops too low in a cpu limited area..
With this, I found out a few things:
1) 1500 MHz and +25mv (0.984v idle desktop) is stable, and I can't seem to trigger the black screen.
2) The GPU speed/overclock speed seems to be irrelevant, but I've only tested this at 1100 mhz. I say it is irrelevant because the GPU never even clocks past 400 MHz when the black screen happens.
With this, I tried something.
I set my monitor (VG248QE) to 144 hz, which forces maximum RAM clocks at all times (seems both NV and AMD cards have this or similar quirk at these refresh rates).
2) I then set the memory directly to 1500 MHz at +0mv (default). So now my memory was 1500 MHz in windows without any downclocking.
What I noticed was bizarre.
The entire screen was flickering in horizontal lines, wildly. I could read the desktop, but the image was jumping around with flickering horizontal lines all over the place. HOWEVER IT DID NOT BLACK SCREEN! (gpu idle voltage=0.963v)
I then set the core voltage to +25mv.
The image INSTANTLY stabilized itself and became crystal clear.(0.984v GPU).
Then I loaded the final fantasy benchmark XIV in borderless mode to avoid any downclocking.
While it was running, I alt tabbed to MSI afterburner (which ran over the top of the running benchmark like an overlay) and set the core voltage back to +0.
There was zero flickering and the benchmark just kept running. I left it running for 15 minutes like this. There was no noticeable corruption or flickering, EXCEPT momentarily when the benchmark "reset' itself after all the scenes were done.
At this point I realized something.
It seems that the CURRENT CORE VOLTAGE (whatever the GPU is currently running at--NOT really what you actually set) somehow relates or influences whether the memory is stable or not. Almost as if the memory voltage, memory bus (which is supposed to be Aux voltage/VDDCI) is somehow either feeding the memory, or feeding it as some sort of skew or added voltage.
This may explain why -many- people have said they can get higher memory overclocks by raising the core voltage.
What someone needs to do is to take a multimeter (DMM), find the readpoints for the memory GDDR5 voltage (which is supposed to be spec'd at 1.5v), then see if raising or lowering the core voltage has any effect on the direct memory voltage itself, or if its somehow related to the memory controller itself (which may be different from the Aux voltage).
For you guys who are having black screens on the desktop when NOT overclocking,
Try 1 of 3 things first:
1) If you are using a 120/144 hz monitor, try setting the highest refresh rate, or anything that can lock the memory speed into boost speeds. Or use a utility (even the old GPU-z 0.7.3) that forces 3D clocks in windows. Does this stop the black screens?
2) Raise your core voltage by +25 or +50mv instead of underclocking the memory. Then check the current gpu voltage in gpu-z. Does this stop the blackscreens?
Before we can say the cause is flaky memory, someone first needs to find the memory GDDR5 read point and see if the core voltage is imaking this go higher or lower than 1.5v.
(note: raising Aux voltage to even 50mv did NOT help: my case seems to be purely a problem with insufficient core voltage in 2D, to stop the memory /controller/RAM/whatever from erroring out past 1400 MHz (I suspect 1425 MHz at 0.963v should be fine too).