M17x R2 7970m CrossfireX Ultimate Installation and Tweaking Guide

PM sent with details.

Good luck bro

 

Thanks a ton Buddy!

 

Flingin, not sure i will beat that bro, but i know i will try.

Not sure i can do it today but will try later.

Did you have to OV the cards to get to above 1K?

 

Isnt it important to go for same kind of card that you have before if you go X-fire?

Has anyone mixed original dells and clevos?

 

What OS are you running and have you performed a fresh install after installing the card? If not, try a fresh OS install or if that's too extreme try the ATIman uninstaller.

The cards are the same only diffrentiated by enduro and non enduro vBIOS's. The only issue that I am aware of is that the 1.1 hardware revision doesn't play nice with the vBIOS that works with out RGB screens. So maybe request the seller send you a photo of the back of the card so you're sure you're not getting the 1.1 revision.

Ask around for anybody who received a key with their desktop cards. Most users dont use it and it comes bundled for free with some cards.

I think it should work but its best to go for the same brand cos if you encounter any issues you'll always be wondering whether the different brand of cards is causing it.

 

I used overvolted Vbios 1.075V

I know i can go more on clocks, but every 3Dmark i do causes my hard to increase rate to about 250/Per Minute....i just fear my R2 will switch off during the test and GG....too many times in the past i have witnessed that, and everytime i am pushing it i fear the worse :/

 

Well Flingin.
Ive given it a try going for overvoltage on GPU and the CPU.

I get a flat out shutdown during this.
My CPU will take about 110W when running as heavy as this.
Adding overvolted GPU's with extreeme overclocking is probably adding some watts to the 100W the cards have each original.
My guess is its my PSU that doesnt want to play very much over 330W and adding only 10W on each GPU during this OC/OV is a minimum i guess.
Add normal usage on the system and above 350 is not out of the question.

I will try more, but the apx 8390 i got this morning before going to work might be my limit.

 

That is weird.

To even start overclocking my CPU to 28X i have to bump voltage to +150mV ( if i add some 2-3% BLCK i get restart as well, but not PSU shutdown, i think this is my 920XM limit )

When not overclocked our systems draw about 280Watt Max, overvolt on cards may bump it to 340 Straight away, and then +150mV on CPU just kills everything.

2x120+Watts cards
Cpu about 100 Watts +

=340 Watts already...

I think it is time for Dual 330W PSU MOD

 

Mhm, mainboard, memory, screen, disks, lights, fans, everything adds to the 340.(if the GPU's have 120 and my CPU has 110 im already at 350)
Fans and disks alone will add 10-20W.

When trying 29X my CPU shuts down, but thats a restart. When pushing everything i get a full blackout on system, needing to repower.

 

Nah, it's not as big as you guys think. I'm pretty sure Nospheratu's 285W reading at the wall was overclocked.

 

Back when I was running a CFX 6990m and a 6970m, my 6990m was a Clevo card and my 6970m was a dell card. Both had Clevo vbios's on them. It worked, but wasn't exceptionally stable. Fans were stuck at 100% and I would get strange gpu throttling for no reason at all. This was fixed when I swapped out the Dell 6970m with a Clevo/Eurocom 6970m and then when I swapped that card again for a Clevo 6990m I still didn't have issues.

A Dell + Clevo CFX combination worked, but was barely usable imo due to the throttling issue. The only way I could get CFX to engage was running both cards on Clevo vbios's. Using Dell vbios's killed stability even more and resulted in to CFX capability.

I'm curious to see if anyone has mixed Dell and Clevo 7970m's and what their experiences were because these cards seem to be more similar to each other between the brands than the 69XX series cards were. I suspect better luck would be had combining a 7970m mixed brand pair.

I'm running Windows 7 64 bit and I haven't done a fresh install. I'll try ATIman uninstaller and see what that does.

Doing a fresh install is kind of out of the question right now, but I might try setting up a Windows 8 partition to see if that does anything.

 

I did extensive testing on the 330W PSU and the Dual/Triple 240 Watt PSU combinations. What I found is that for some reason the 330 Watt PSU (I tried 3 different ones - each with the ID chip mod done to them) can only draw about 240-280 Watts from the wall before its internal overcurrent protection circuitry trips.

The 240 Watt PSU is more varied. I'd tested 4 different ones. Some can handle up to 280-290 Watts of draw from the wall before tripping while others can't even handle 250 Watts of draw. Some forum members on here have reported the 240 Watt PSU tripping at even below the 240 Watt rating.

Since my setup with dual 6990ms overvolted and overclocked (I did a hardware level voltage mod to trick the voltage regulators into thinking the card was under load and thus seeing a voltage drop, so they would ramp up the voltage to above stock to compensate) and my overvolted 920xm with 110 watt TDP drew around 400 watts, I needed dual PSU's of some kind at least.

I found when you combine two PSU's you don't get a linear increase in potential current delivery before the overcurrent protection kicks in. Adding a 2nd PSU seems to give only at best 50% more headroom and a third only at best maybe 25% more headroom above that. It might have something to do with the voltage loss through the isolating diodes thus resulting in a lot wasted power eventhough I'm using schottky diodes with only a 0.7V forward drop.

Two 240 Watt PSU's in parallel would trip at 350 Watts. Not enough power for me. I added a third and I'm able to draw over 400 watts from the wall and not have a problem. Idk where the limit is for 3, but it's probably not much higher than 425 watts or so from the wall.

Since the 240 Watt PSU's actually have the potential to deliver more sustained current than the 330 watt PSU's I would highly suggest doubling up on those instead of bothering with a dual 330 Watt PSU configuration. You'll probably get more headroom from the 240's.

 

Yep. All i can say is that my cards worked out of the box, both from Eurocom.

After 300W PSU Mod my machine is perfectly stable, and, to be honest, i expected directly the opposite.

Mixing up cards can end up with mixed feelings, that is what i knew from the beginning, so i ordered 2 same cards.

I am almost sure that if i would encounter any problems in getting this running i was prepared to sell one card and just use one card with nice OC.

So far i am verry happy, but when i see 780M performance, especially in SLI, it leaves this moment of not being 100% accomplished after all i have been trough.

NVidia SLI certification Pain in the @$$

 

My cards are also stable now after going for Clevo drivers on Dell cards.

On techinferno they say the cards are slightly different from eachother, but i have to say i dont know how or what it will do if combined.

My normal PSU (240) probably gives me 260-270 making me able to run with UV on GPU stock clocks, and very little OC on the CPU.

Imsolidstate has made an upgrade on the PSU update after discovering the 240 W problem you describe getting nearly 380 out of the 330W. You sure that update has been done on the one you tested? The one i use now has the 330W upgrade.

I would however guess that like the 240 there are differences in the 330's output.

Need to check my actual power consumption. Have to get a meter for that. Quite sure my shutdowns was caused by PSU, but will try same GPU OC with stock CPU to verify.

 

Wonder how much higher a 780 SLI would go on the M18. (or 7970M for that matter)

They will also have to be close to power related issues if thats what we see if OC'ing GPU and CPU.

 

Yes I have the meter. Undervolted gpu setup draws 275watts. It is our CPU that kills everything. 45nm technology draws 110 watts from 920xm cpu where 4th gen hasswell can perform the same while drawing half of it....

Sent from my SM-N9005 using Tapatalk

 

Have you checked the full go like you have done now? Is that where you have seen 340?

 

No bro, when i was doing the tests yesterday i have not measured the power draw....forgot about it...

 

Record for 780m's is something like 20K GPU score extremely overclocked (I think it was Mr. Fox but I may be wrong).

Average for 680m's/780m's/7970m's stock or overclocked is no different from ours.
Noshperatu had some bombing scores.

If your cards OC well, you get good scores. Since this machine is not throttling much either due to power or temps, it OC's very well if your cards are ok.

Don't imagine that every m18x or AW18 out there OC's well and can put up monster scores.
In fact, if that machine isn't using 4700MQ or stronger, with at least 7970m's, then it has no chance to beat you.

 

Highest power draw I got was 325W at the wall with GPU's running at 1000/1400 with 1025V and XM at 25x across all cores with 90/82 TDP/TDC in 3DMark11.

 

So imagine this.

For this test

AMD Radeon HD 7970M video card benchmark result - Intel Core i7-950 Processor,Alienware 014M8C

i had CPU on +150mV

TS 105/93 28/28/28/28

GPUs 1.075V overvolt 1025/1350 OC

 

GPU driver version? Also that OC is clearly overkill in the combined test. The score is so low it's hard to figure out what killed it, but I suspect the crossfire is starting to hiccup.

Do you mind running those same specs with same OC only single-card (crossfire off)?
- disregard this


ALSO, THAT CPU SCORE IS INSANE.

EDIT: I'm an idiot. Forgot it's not 3dmark11 again.
EDIT2: This is the top highest valid score for 7970m's on 3dmark.com. As you can see, you're pretty damn close.
http://www.3dmark.com/fs/389803

His CPU was overclocked, 2 gens newer and second strongest after extreme series.

 

Yeah that's.... crazy!

That 330W would have definitely gotten warm after that run. I'm stuck on +-5700 for Firestrike. It's up to date so I don't know why I'm getting that score. It's as if the second card isn't scaling well.

But to be honest I don't use 3DMark 2013 much and I haven't really tried to resolve the issue. I'm running the free version and sitting through those demo's and first two benches is torture, for me and my R2 running at those clocks.

 

I'd love to know how my machine deals with Fire strike, but it still freezes whenever the Cloud Heaven physics test starts, even underclocked. Some kind of weird, i suppose.

I hate the fact that they force you to take the other tests. Yea, give me the demo, ok, but forcing me to go through those 10000-fps tests is pissing me off.

 

AMD Radeon HD 7970M video card benchmark result - Intel Core i7-3840QM,CLEVO P370EM

Challange accepted Bi@^!%

 

This i had before i swapped to clevo vbios sangemaru.

Had to lower to 800 or below and 1100 to be able to run all tests.

 

I don't mean to be a wet blanket... but to be fair try to match driver versions for a more accurate comparison.

edit: @sangemaru, I remember you mentioning you tried a bunch of pastes including Noctua NH-T1? How did it perform on your XM? I've only got a little Shin Etsu X23D left but almost a full tube of NH-T1 from my HTPC cooler so I was considering using that for repasting the XM. I want reliable performance over a a long period as repasting the XM is more work than the GPU's.

 

Not sure i can go any higher. Still behind

AMD Radeon HD 7970M video card benchmark result - Intel Core i7-940 Processor,Alienware

this is as high as my GPU goes, going higher makes it shut down, I think. (gonna try some more)

 

I see Bro Throme71 is getting greedy

 

LOL-
Nah, i see you can beat the crap out of my GPU.
But i want to push you above 8500
I can and will probably go for 29X and 150mV but if im to do that i have to do the test in a colder enviroment than the 23 deg livingroom.
Finally found a test that my "mark11 crappy CPU" really really exels at. Want to see how high it can go. Think like have been mentioned in here before that it could be the 3Dmark11 program not running well on my pc.

Have you had any success going for the highscore?

 

Did not tried it yet, but as far as i know, this is the best score for 7970M with normal Tesselation Level
AMD Radeon HD 7970M video card benchmark result - Intel Core i7-3740QM Processor,CLEVO P370EM

I do not think i can push my R2 that far, if even yes, i am scared that something will blow up because of Overvolting.

I really want to keep my R2...intact.

By the way, why should i push my 7970M cards, i have already preordered a AMD QS R9 290XM which is almost 20% faster than 780M GTX, only i have to wait for shipping till the end of this month

And it looks almost the same as 7970M card just 4GB of ram and same 28nm.

Good thing the chip is the same as desktop R9 290X just the shader count is a bit smaller and 384Bit instead of desktop 512Bit

Fingers XXXXX

 

You have done that? LOL you mad bro

 

Problem is, I can't really tell if it didn't work right because of bad heatsink mounting, or because the paste was a bad match for our machines, but it was nothing to write home about (I managed to get 100TDP @ 95C, but quickly degraded from there). Still, since it seems like I'm pretty much the only one with this kind of issue, you should try it, reviews have it as "AWESOME".

That being said, if you want to lengthen the intervals between necessary repaste/maintenance, you HAVE to lower your temps (TDP). The only way I can see for us to bypass the need for maintenance is to get some HeatSpring TIM foil or similar malleable metal with a coherent shape.

I hesitate to propose this, but do any of you guys feel like experimenting with HeatSpring TIM? I have great expectations regarding performance after my experience with using thermal pads for cooling the CPU.

For 300$ we'd be able to get 50 square inch pieces (as far as i can tell there isn't a single mobile chip with a length greater than an inch - 940xm is a 7/8 inch long).

I'm thinking if 10 of us were interested in chipping in 30$ each, we'd get 5 pieces each to play with (i'd say 1 piece is enough for both GPU's). It would be best for the buyer to be US-based (unfortunately, I'm not), so the

This thing doesn't need maintenance, replacement, it bonds to the copper heatsink over time increasing thermal conductivity...

Unfortunately, I'd have bought the 300$ pack by myself if I could afford it, but I can't justify spending that much myself just for thermal interface.

 

It would be interesting to try this.
I can be one of 10 but im not in the US.

 

I would be interested too but I would have to see results before I commit.

How did that sample testing go? Were there results posted?

Also that sheet is uniform thickness and quite thin so if theres a larger gap between the heatsink and die no compression of the TIM will take place...

 

There was a discussion in this thread http://forum.notebookreview.com/har...ltrabook-owners-heat-spring-tims-testing.html

Perhaps some of the answers are in there?

 

I'm interested, but from what I can tell, it's not reusable. Also, is there a possibility of in essence "soldering" your heatsink to your die so that you essentially can't remove the heatsink when cold?

That thermal efficiency is insane though - but I'm not sure how much of a difference it will make with the cooling setups in our R2's. It seems to me that our heatsinks get saturated pretty quickly already and that's the main limiting factor. A better option might be to look at a way of improving the heatsinks and fans - maybe M18X heatsinks can be modded for use. Or figuring out a way of directly interfacing the heatsinks to the aluminum chassy.

There's a company out there who offers a fully enclosed water cooling solution for the M18X without external components for ~$300. I know they used to make one for the M17x R2 as well. I've never seen or heard of someone using that setup in real life though . . .

 

Hopelesslyfaithful has some encouraging information in his thread, though it never did pick up the kind of pace to allow for a wide amount of testing across various platforms.

Yea, the sheet is very thin, and requires pressure of at least 30psi to be effective (about 70psi is the sweetspot). IC Diamond have done lots of pressure tests across multiple laptops and found that typically laptop heatsinks offer pressure between 30-80 psi on average, so that wouldn't necessarily be the problem.

Unfortunately, obtaining actual performance numbers is very hard to get. I'll be looking through the thread to see if I can find anything, though.

It IS reusable, you just have to make sure not to damage it. Also, if over a long enough period of time, the indium does bond to the copper, then removing it may become impossible (not that that would be a problem). Indium bonds to copper over time, but does not bond to Silica, so the processors would be safe.

I know for a fact that my heatsink system can dissipate MUCH, MUCH more than it does currently. There's no real point for me to look into modding heatpipes/fins/fans so long as my bottleneck is at die-level. You can identify the bottleneck pretty easily, by seeing where improvements lie. If you get great temps right after repaste but they degrade over time, and your laptop doesn't care about cold air being blown over the fins, then it's either the heatpipes or the contact to die, and you can check the heatpipes for defects using water (and i should have done that before purchasing my second heatsink).

Regarding the water-cooling solution, I'd probably be totally for it, if they still sold it.

 

Ah, then if it is reusable, then I'm definitely in for the possibility of a group buy.

 

Ok, that makes 4 tentatively interested persons. Let's see if there'll be enough more people showing up to make it feasible.
We need a U.S. citizen for this too, preferably (mostly because I imagine it's faster for the kit to be sent to him/her and then cut and sent, than for it to be flying halfway across the globe first

 

I'm number 5. And I'm an American living abroad in Germany, so I know how the shipping works. Still would like someone else to handle the order and cutting, but if no one else is on board, I'll step up.

 

I think somebody is getting too excited here, this Heat Spring is essentially this Coollaboratory Liquid MetalPad - the innovation of cooling for High-End sytems

Which i tried in my R2 and i could barely get the thing melted on my DIE, more, i almost burned out my CPU, just because the burn in process takes too long, how many of you would watch your CPU go 100'C for 1 minute ?

Anyway i'm in, maybe this is really something new, but i doubt it....

 

Similar concept but I think the similarities end at the fact they both are metal. This doesn't need to be heated up to melt to work. Just place it on the die and the install the heatsink. Boot up and carry on like normal as if you just installed your favourite thermal paste.

It only bonds to the copper of the heatsink over time so we should be able to use it for a week and then slide it off without hassle. I think we may need to do the c-clip mod to optimise its performance though.

 

Yeah, the only thing I'm worried about is the pressure. 80+ PSI is a lot. I'm not sure if even the c-clip mod will produce that much pressure.

 

I would guess the pressure is dependant on the size it is supposed to operate on. Bigger surface= higher pressure, and our GPU and CPU are not identical.
Its not that much.

Not done this in many years but this is how i remember it:

Generally 80PSI = apx 5 bar.

1 bar @ 1 square cm = 1 kg.
If die size is 1 sq cm we need 5 kg pressure.
Less than 1.5 kg per screw.
CPU is probably a little bit bigger needing more pressure.

Screws are very small but has extremely low pitch.
Its sort of a gear ratio.
circumference to pitch ratio is a sort of a gear (minus 50% friction.)

Sorry for doing this metrically. The US system SUCKS since there is hardly logic between different measurement systems. (weight/volume/pressure/etc)

 

Here is some comparisons with Overclocking



Seems like 1.1V can be Stable and...Safe

 

Uuuu, nice. That makes 6. UK would be a really good shipping nexus too, if you're up to doing the spread.

And don't worry, it's very different from the Coollaboratory product, it retains a solid coherent shape, and the bonding is chemical in nature (not soldering requiring melt), and there's no worry about it shorting out your area, although it is capacitive, but extremely unlikely to ever move.

And don't worry about pressure. I should worry, and I'll probably want to stick 2 or 3 on top of each other, but all of you have great performing cooling systems as far as I can tell, and 30psi is extremely easy to get (you'd have to have a really crappy mount for that little pressure averaged out over the area).

That.. makes me really want that heatspring tim..

 

Yajks, i am really bored nowdays, thinking about improoving my R2 all the time. Lets hope that Heatspring will not waste our time...

I have 2 things in my mind now.

1st is

Modyfying the R2 Heatsinks, you will probably see a Dedicated thread for this in a very near future. Only think i can say that it is awsome.

2nd thing is

I am planning in swapping my Crappy R2 speakers with GX660R speaker system from Dynaudio or if it does not play i will use R4 Klipsch Speakers.

I have more things going trough my mind now but these are in early Alpha stage....


OMG what a Batman Video i just found LOL

http://www.youtube.com/watch?v=H8GgGt9kPdk

 

Lol, I'm also thinking about cooling mods for the XM So I'm looking forward to your heatsink mods!

Do you have a link to the source of that image? What machine was it from? Did they say what pressure was used?

I was also thinking of a speaker mod, more specifically adding the subwoofer from the newer laptops.

 

BTW i bought a 920XM OEM to be ready to replace my odd operating 940XM.
I know it scores good in 3DMark, but I also know it doesnt operate as it should, and puts the breaks on for 3DMark11.

Should not be able to OC that heavy at 75mV (5% + 28X) i guess. Can also run 28X with stock BIOS.

Will keep the 940 in for now, but since i found a 920 XM OEM under "new other" i could not pass on it knowing my CPU is strange.

Also want to check my performance in BF4 with the 920 to see if it improves.