Could someone please explain what it's going on with this 1070M......

Just wondered if someone could help me with overclocking & undervolting my x7 v7 GPU please.

So if i run Superposition benchmark i get a score 10417. This is with my CPU @ 4.0Ghz with a -155mV My average GPU clock is 1720mhz @ avg 0.946 V

I use MSI afterburner and I can run 1848mhz @ 0.875 V & 200mhz mem overclock stable. Temps are much better but i only get a score if 10100? What gives?

Any idea?

Sent from my SM-G955F using Tapatalk

 

One of Pascal's quirks is that if the voltage is too low for the given clock speed, instead of crashing it reduces performance. That's what you're seeing.

 

Interesting. So the key to undervolting it is to keep the max volts at my settings and then lower it until I see a drop in performance, NOT a black screen.
Cheers buddy.

Sent from my SM-G955F using Tapatalk

 

Here are my results. Every time i underclock it lowers the score :/

Default GPU
4.0Ghz @ -155mV
Firestrike Score 14685

+100 core +100 mem
4.0Ghz @ -155mV
Firestrike Score 15121

1797Mhz Clock 0 mem @ 1.000v
4.0Ghz @ -155mV
Firestrike Score 13604

1797Mhz Clock 0 mem @ 0.963v
4.0Ghz @ -155mV
Firestrike Score 14161

1797Mhz Clock 0 mem @ 0.925v
4.0Ghz @ -155mV
Firestrike Score 14131

1797Mhz Clock 0 mem @ 0.900v
4.0Ghz @ -155mV
Firestrike Score 13980

 

How are you altering clocks/voltage?

 

delete

 

Through msi afterburner

Sent from my SM-G955F using Tapatalk

 

The thread or your comment mate?

Sent from my SM-G955F using Tapatalk

 

I posted a video someone else posted so i wanted the mods to delete my message.

 

What does your voltage/frequency graph look like?

 

Basically got a flat line that tuns at the desired 'clock'
I shoes that I'm holding it at thee set volts but just doesn't seem make any speedy difference. If fact, as you cam see, it makes it perform worse!!

N. B. Not my graph but looks like this

Sent from my SM-G955F using Tapatalk

 

OK. if you're not power modding it will be dropping back down the boost graph under power limit, and not using your overclocked 'top of the cliff' max voltage point.

Run FS in a window, with that graph open, along with gpuz/hwinfo/AB graphs showing power use. Watch where your power use sits (~115W on default msi 1070 for example), and also watch where the crosshair sits on the graph.

If you haven't touched the left end of the curve, and power limit keeps it away from your clifftop point, it will be running pretty much exactly the same as stock.

When I generate a graph, first I set a +offset clock, THEN flatten from the desired max voltage. That way the low end of the curve is also overclocked should clocks fall back down under power limit.

 

Cheers I'll try that tonight. I think mine is 115w as well.

When I've had the graph open when running heaven benchmark. It goes straight to my desired clock (1797mhz) and I'm not hitting 115w. Think it's about 106w. I saw the video about false clock readings hence i tried with 1.062v at that speed and still no joy.

Sent from my SM-G955F using Tapatalk

 

With 100% core utilisation? And no power throttle flag in Hwinfo/GPUz/etc?

 

I'll try tonight an post up a screenshot

Sent from my SM-G955F using Tapatalk

 

First pic default, then +100 mem and core, and then 1797mhz @ 0.981mV
Hope this sheds a little light on the situation.

 

Why are you testing at such a low resolution? You're CPU-limited at times, so differences in GPU performance, especially minor ones, are gonna be masked.

 

Just so i could see what was happening. Running it in full screen still has the same performance hit on the gpu.

Sent from my SM-G955F using Tapatalk

 

Run it in fullscreen 1080p.

The middle image has the highest score because on average it's boosting higher than the top and bottom, and also has a memory OC.

 

The results i posted up are 1080p firestrike. Yeah I always get higher scores when i overclock. It's just when i set a clock that's higher than the average over an overclocked, or the same. The score is much lower.

Sent from my SM-G955F using Tapatalk

 

From the 3rd image, it doesn't look like you followed @bennyg's advice to add an offset and then flatten the curve. Your graph should look something like this:



What you're doing right now, that is, flattening the curve without adding an offset first, is you're limiting the card's max boost, or underclocking it. Obviously, that's gonna reduce performance.

Since your 115W 1070N is mostly power limited anyway, I'd just add an offset that's stable and be done with it. If you flatten the curve after the fact, you're clamping the max voltage/frequency your card can achieve in less power-limited scenarios.

 

Hi,
Unfortunately i cant view the image but i think i now understand what he meant. The reason i mainly wanted to clamp them down is to help with temps. Im gonna try this.

added +100 core and mem with it flattened @ 1.000V and see how it affects the score

 

So heres an update

I went +100 on core and mem and then flattened off 1898mhz. This didnt work and failed to increase clock when benchmarking (stayed at 1443mhz) so i lowered it to 1850mhz @ 0.975v it seemed to of worked!!!!

my question is can i now straightened the line to the left to lower the volts?

 

So I've just done a FS run with 1860Mhz Core & +300Mhz Mem and it seems to be ok. I tried lowering the voltage to 0.963v but it adversly affected the scores.
Do you think my score is pretty solid?

 

Score looks good.