7970m...how high can it overclock?

ok thanks...learned a new thing again today Mission Accomplished

 

So I've had just for benches vram to 1660 with low core and 1050core with low mem. But I can't get either that high at the same time. 1.075v is what worked for me but 1.1v didn't help me any it moslty hindered.
1.075v for gaming was less stable at 950 1350 for me and stock volt at 950 1350 worked. At 995 1560 I got p6813 I think. At 1025 1350 I got about p6800. At low core 1650 mem with 11k physics it neared p6800 also. But I can't seem to get 1000+ core benched with 1400+ mem
Like slickdude I see 950core 1350-1375 mem stock volt fairly good and in some major games and long term sessions maybe slighlty less on that core like 940-925. These cards could burn out easier at lower temps then last gen since the fabrication process has gotten so small. Gaming 62c max so far with 950 1350 clocks stock volts and when MSI isn't causing my card not to clock down it idles at 27c when temps are aloud to level off.

 

i think as someone said before your lower score at higher freq is because of errors. It is having errors and having to rerun the stuff or whatever.

 

Okay so the mem is the sore spot in those attempts. Interesting the Hynix is rated for 1250.

 

So at stock voltage we can go 950/1300 fairly easily?
About a 12% OC and that is stable w/ no heat issues or game issues?

That would be nice as it puts the card just about at 6990m CF levels. Haha.

 

So long as memory bandwidth is ample OCing the core will always bring much better results. If I were you I would bring the core up to the point where you start to see performance being mem bandwidth bottlenecked. Then simply up the mem frequency till the bottleneck is removed. This way you will get a better balance and only push the mem as much as is needed... yes this means patience and alot of testing...

 

^ Isn't that how it should usually be done? Or does that depend on the GPU? I thought GPUs, at least high end ones, are generally given enough mem bandwidth that you do not see performance going flat with increasing core clocks for quite a while. THEN you go up on memory, rinse and repeat? Or is that not correct?

 

That is the way I would do it yes. Memory OCing takes alot of extra power and creates lots of heat too. Many people though simply raise both equally thinking a memory OC will always be beneficial...

Considering the desktop 7870 has the exact same mem bandwidth but clocked at 1ghz I think the mem could quite easily be left at stock without much noticeable performance loss. While the card will run cooler and more stably.

 

Except improvements are seen just clocking the memory.

 

950/1300 is easy. on my card, 1ghz/1.5ghz is on stock voltage

I drop it down to 950/1400 for games cause I don't think you need more for everyday

The mem clock at 1200 is severely limiting this card. You see noticeable improvements all the way up to about 1480mhz vram on my card. After that, i don't see any significant improvements so I'll say that the vram really should be at about 1450

 

Yes. It would be pointless to add an option to increase the memory clock if there were no improvements.

That being said, improvements are seen in both cases, AFAIK. I do not have the 7970m so I do not know if it is bandwidth limited to begin with. If that is the case then sure, increasing the core clock will be pointless unless you increase the memory. However, for a GPU where this is not the case (including the 7970m if it does not apply to this carD) it is only logical to increase the core clock until you are bandwidth limited, then increase the memory clock and increase the core clock again until you see a flat and keep going.

 

Define flat? Flat would depend on the application.

 

Sweet so in my P150EM I'm going to try 950/1350 maybe 950/1400. Hehe. As long as it stays below 75C I'm happy

 

Flat = no more gain. I do not understand why it would depend on the application. If I OC'ed the core and the memory was limiting the GPU, would different applications give different results?

 

Look at my 7970m overclocks on memory , up till 1650 memory I saw an increase in fps, proof right there

 

Because you are basing it on the erroneous assumption that every game has the same bandwidth usage. They don't. Different games react to memory clocks differently.

When one game would "flatten" another would not, so I find your definition of flat lacking.

 

Im just hoping this 7970m that i get will be as overclockable as my old 6970, ran it on an everyday overclock of 850/1100 with temps under 80c (topping out at 81 after a long load)

 

That must have been one special card!

 

Unless you get a stinker card, you should get there easy. I'm getting 1015/1470 stock... seems most people are topping out around there.

 

If increasing memory clocks improves fps values, then you are bandwidth limited to BEGIN with, no? That means you first increase mem clock until you see no more of an fps gain, then go up on the core clocks, then go up on memory. I'm obviously too dumb to understand your one liners so if you could explain better, that would be helpful.

 

I am not asking for proof. I am not trying to challenge. I do not know if that is how my language is coming across. I apologize if it is. I am trying to UNDERSTAND.

 

Ive never had a gpu that loved bandwidth so much , I myself don't even understand. This architecture just benefits from faster memory, which is weird cause my 7970 desktop card doesn't even see this much of an improvement

 

Some applications/games are more memory bandwidth intensive than others. For example, if you play BF3 you might stop seeing improvements above 1400mhz VRAM speed (made up numbers, just for an example) but the value could be 1250mhz if you're playing Metro 2033 or something.

In other words, the amount that the VRAM is bottlenecking the GPU differs for every game, so you can't just say there's a single mhz value at which all VRAM bottlenecks are gone (aside from the obvious, "if you OC the memory to 10ghz there will be no VRAM bottlenecks").

 

this is truth

 

This is interesting feedback. Come on where is GDDR6? LOL we obviously need it

 

7970 desktop card has a 384bit mem bus.

As for me I am stuck at 1.05V as 1.075V makes my system shut off halfway through the benchmark.

 

Similar to me I left it at stock voltage 965/1500 but otherwise I went up to 1030/1500 @ 1.1v for benchs and that was OK (shutdown for BF3 tho, even at 1.075v)

Our 16F2 lacks a bit of power.

 

Did you have the 560M version? The 570M version might do better.

 

Yep 460/560M version too - not sure the 570M's will do better, someone needs to try I guess.

 

Well the power circuitry is different. They certainly should.

 

Was 965/1500 stable for gaming such as BF3/Skyrim/Crysis and not overheating? [above 75-76C? I like to stay down below 80 by a few].

I was hoping on my P150EM when it comes that I could hit up 935/1440 ish stable.

Thats 10% on core and 20% on memory clock.

Which one makes a bigger difference in gaming?

This would make this thing within 15% of a GTX 580 desktop card. Pretty ridiculous....

 

Yeah that's 100% stable @ BF3 ! Max GPU temp is 76°C after a long session of BF3.

I can bench @ 985/1500 otherwise - everything @ stock voltage, I don't overvolt.

 

So far stock volts 940core 1250 is real fine for me. 950 core crashes after an hour and 1300mem up does too. 925 core 1300 crashed on me but 940 1250 does fine. 950 1250 I'll try again soon but seems not for my system as with vync off it will crash in just Diablo 3 while playing hard.

 

What are your temps?

Sent from my SGH-T989D using Tapatalk 2

 

I don't break 62c in gaming. And benches it's even lower. When I do have probs temps are at 62c I had to drop to 925 1250 in diablo3 no vsync full graphics after playing awhile 940-930 was an issue for me.

Bench wise I can max core at 1050 stock ram and 1660+ vram on stock core all at 1.075 but for gaming I only see what I mentioned stable. I think it's drivers. Originally tested on beta driver and now on the one MR.FOX recomended. I'd like to game on 1.075v but when I do I seem to crash at same to lower clocks than I get on stock volts. Just been using. One card for the whole time.

To mention room temps are lately near 70F-80F often toward the 80F lately. I'd rather be stable in heat than try to stabalize in a cold room and use it one day just to find I have to test and lower more.

 

hey guys. kinda new to the built laptop world so i would be happy for any help.

I just bought a M18X R2 3920xm @3.69ghz, 32 Gb ddr3 1600, dual 7970m crossfire, and I ran it through 3dmark 11 and got 10272 all stock with no updates. then got 10072 when I installed 12.6 beta drivers. the gcard score went way up but the physics score dropped by 1000, aren't the physics dealt with by the CPU mainly?

also is that what the laptop SHOULD be scoring? seems kinda low to me.

 

Physics is a combined GPU/CPU benchmark.

Make sure you are in high performance mode as the CPU will directly affect the score.

 

guys are scoring 11K+ on the m18x forums running the 12.5 beta modded driver. you can look in my 7970 guide (in sig) for the link. Performance will increase as 7970m xfire support matures. So you are a tad low

if you want tips, head on over to the m18x forums

 

the laptop itself? its always in Max mode. its always plugged in. I went from the old drivers to 12.6 and my score in physics went down 1000 but my graphics score went up about the same amount. so the final score was exactly 200 lower. also those drivers seem to run crysis 2 dx11 high Res pack all ultra at Max Res fine but it keeps shifting sometimes AS I'm playing it from normal and high Res textures and I can see it happen.

 

the slider of mem clock in my afterburner only max out at 1560MHz and i cannot get higher even i type the number. how to solve this?

 

what? 30% OC not enough for you lol.

Just close afterburner and reopen...you'll have space again on the slider

 

are you talking about the 7970m? because on msi afterburner i can top out at 1105/1560. Im on 12.7 driver.

 

i have tried still no luck. i can run furmark for 10min with gpu clock 1080MHz very stable but now just bottlenecked by the mem.

 

ok this is what you have to do.

open afterburner and set your clocks. CLick "Apply"

Now close afterburner (click X top corner)

Open afterburner again and you will see that there is lots of room to overclock more now.

Another thing to consider is that I'm using Afterburner 2.2.1...I don't know if there is a difference in 2.2.2

 

I already tried many many times but still the same. maybe because my one is sager not alienware. or because i am running stock driver. thanks anyway.

 

hmmmm...that's weird.

you're getting 1080 core? that's amazing! And i agree, your memory should go higher. Have you tried with Afterburner version 2.2.1?

 

I cant understand why ur downclocking can make ur system unstable since im experiencing the same thing, absurd as it looks, i underclock my card(vbios set on 900/1300/1.075) to 885/1300 using afterburner made my screen flicked, does anyone have an explaination for it?

 

yes i am using 2.2.1. i have tried other benchmarks and 1080 core is unstable. now down to 1045 and so far so good. still impressive for stock voltage.

or the old atitool still work on 7970m? should i try?

 

forget the bencmark scores, do you guys actually get any performance gains in games line bf3 when you overclock?

 

4.2FPS gain in Metro and...18 in BF3 (i still cant figure it out why but it seems nothing wrong with my settings) my clock was 900/1300 then and i downclocked memory to make it safer.