overclocking MSI GTX1070 Gaming X

M

Momo

Gawd
Joined
Mar 3, 2005
Messages
869
hey folks,

I'm taking the plunge and finally overclocking my GPU. All 'seems' well so far.
I hit a wall on memory at +825mhz if i recall, without a GPU overclock.
I hit a wall on GPU overclock at +150mhz

so i'm currently at +125 core and +750 RAM.

I'm using Unigen Heaven with settings cranked to find instability walls.

With these slightly lowered settings, how long do i have to run unigen to consider this "stable"?

or is there a better program these days for stability testing a GPU?

0kBe5


edit #3 why the heck is this not working as adding an image url?
http://imgur.com/a/0kBe5
 
Not to shabby. I'm getting the same clocks on my 1070. There's a Power Limit Mod another member posted up that i've been looking at. This shunt mod will lower the TDP so the card will not start to throttle back the clocks and have better power delivery flow with less restriction. Not saying you'll get massive gpu core clocks, but will help some.
 
ah thanks!

turns out 125/750 isn't stable in game (Project Cars) got artifacts after a few mins then a crash.
dropped it down to +100 and +700 if i recall, or +725, and played L4D2 for 45m without issue.

... what is the gold standard for GPU stability testing these days?
FurMark ?
 
Momo said:
ah thanks!

turns out 125/750 isn't stable in game (Project Cars) got artifacts after a few mins then a crash.
dropped it down to +100 and +700 if i recall, or +725, and played L4D2 for 45m without issue.

... what is the gold standard for GPU stability testing these days?
FurMark ?
Click to expand...
My personal test is playing a game for 4 hours.

If you want a higher core clock you probably need to back the memory off. I would start with +0 on the memory and then push the core as far as you can to when you can pass my test. I like to start with +25 at a time until I hit +100, then I'll nudge it in 10 MHz increments. I'll push the memory in +50 increments after getting the core stable. I didn't see as much improvement in pushing the memory on my Titan X compared to my GTX Titan X, just FYI. I settled on +500 for about 11 GT/s, which allowed me to squeeze a bit more on the core.

They're not comparable, but just for the record I run my Titan X at +120 core and +500 memory using EVGA Precision X OC with no extra voltage and power limit set to max.
 
the numbers i posted above as max were independent of one another.
s0 150mhz was the wall GPU only
825 was the wall RAM only

I'm now in experimentation mode with real games.
 
so with my core set to the same as the above picture, +125, in FurMark i get around 1797-1809 boost clock.
it shows 100% load on the core
it shows around 90% load on the memory controller


with the same settings in Unigen i get ~2062mhz core
it shows 99% load on core
it shows ~60% load on memory controller


Why is that?
why does that make such a drastic difference in the core boost?
 
Momo said:
so with my core set to the same as the above picture, +125, in FurMark i get around 1797-1809 boost clock.
it shows 100% load on the core
it shows around 90% load on the memory controller

with the same settings in Unigen i get ~2062mhz core
it shows 99% load on core
it shows ~60% load on memory controller


Why is that?
why does that make such a drastic difference in the core boost?
Click to expand...
That's because furmark loads everything and you get a higher overall temperature - since gpu boost depends on the gpu temperature, higher temps = lower clocks. In unigine heaven you have a lower total load on your gpu, which allows higher gpu boost at the same temperature. And/or more load on the memory controller means more power is needed there. Since the gpu has a total power limit more of it is used by the memory/mem. controller and less power is left for the core.
Don't expect to have the highest boost clock in all games unless you manage to keep the card, if I remember correctly, somewhere below 70°C and even then it may reach the power limit before it reaches it's maximum core clock if something else, other than the core, is using more power in that certain application.
 
ahh ok, it's a give/take thing, with a finite limit!
That's why in FurMark it shows limits: 1 power, 0 heat, 0 everything else i suppose
 
soooooooo, what is a better stability test. furmark with lower clock but more overall card load, or unigen with more core clocks and less overall card load?

i'm guessing there's no straight answer here.
 
Momo said:
soooooooo, what is a better stability test. furmark with lower clock but more overall card load, or unigen with more core clocks and less overall card load?

i'm guessing there's no straight answer here.
Click to expand...
Unigine is the better one because it simulates loads that you will most likely see during a gaming session. Furmark is only good for killing components and 3DMark is only good for comparing e-peens.
 
You must log in or register to reply here.
Back
Top