M15x 7970M throttle FIX !

After seeing what you had said about the fans, I decided to test it for myself using Hitman 3. With the exact CPU and GPU clocks which you have previously used, the game indeed throttled after playing 15 minutes into the game. Though the throttle was only momentary, it happened from time to time as the fans were not at maximum speed before the GPU was loaded with 99% usage. Thus, as you have stated, there might be a strange relationship between the fans and throttling, which is something I cannot comprehend yet.. hmm.

 

This is a great discovery though. Using Hwinfo we can get the fans up to speed before we game thus avoiding the throttle. I am still unsure however as to whether cutting past natural bios control in the spooling up of the fans is just as effective in stopping the throttle as allowing the fans to naturally spool up while sitting on the menu of the games we want to play for a couple of mins.

The M15x is still a sensitive beast with voltage though and will still not allow us to do any beastly OCs and almost certainly no overvolting.

Still however lets find out what the M15x can do onego with these workarounds in mind. Maybe we can unlock certain potential and then fully document any findings and give advice in a new thread to advise other M15x owners how to get the most out of their high specced setups.

 

IMPORTANT UPDATE

Ok I did some testing as follows using the clocks in sig and cpu at 73/60 tdp and 27/25/20/20.

Firstly I put both cpu and gpu fans to max speed and then ran FC3. It started throttling right from the beginning.

I then reset and tried just gpu fan to max and cpu fan system controlled. The gpu throttled even more severely eventually sitting at 450mhz continuously.

Finally I rebooted once more and ran gpuz and its built in render test. Once temps had reached around 60C I quit the test and loaded straight into the game with no time lost. Then I left the machine running at the menu screen until the fans spooled up to max (about 30 secs to a minute) then I loaded into my saved game and NO THROTTLE AT ALL!! I gamed for an hour or so and enjoyed it very much.

So the throttle has everything to do with fan behaviour HOWEVER setting the fans to max with hwinfo does not help! One needs to let the system slowly acclimatise to increasing load and temp and ramp itself up and fans naturally. It seems this way oddly the system is ready for the full stress of running a game at 100% gpu load without throttle issues.

I will slowly test this out using increasingly high clockspeeds and voltages. Considering this seems to be a new finding perhaps this approach may allow us to finally push our hardware as it should be!!

Perhaps this process allows the sensitive gpu voltage circuitry to get used to increasing power comsumption. Perhaps suddenly putting a high load without any prior warning or warmup time is what causes the problems. If this is the case then perhaps the cpu isn't limiting the gpu as much as we thought if at all!

 

Using the above approach I tried 875/1400 at 0.975V with cpu still at 73/60 tdp and 27/25/20/20

No throttle whatsoever and the game is running very well indeed.

 

I have tried 900/1400 with sucess. I had to dial in 1.025V for stability. 1V didn't quite cut it. Even with near stock voltage levels and cpu OCed at a getting on decent GPU OC not a hiccup.

I think we can consider the gpu throttle problem fixed.

 

Wow. I've got to say I am impressed king. I tried out your hypothesis following the exact same clocks that you have right now, from loading the GPU with MSI Kombustor and jumping straight into the game. No throttle or whatsoever. Thank you, my saviour. +Rep

 

Thank you mate. I took quite alot of testing. It seems my card needs increasingly high voltage once I get over 875mhz though. Understandable I guess considering it is going above stock clocks. I reckon at least 1.05V is required for 950mhz!

Next step is to see if the CPU is overclockable further! I will start testing this weekend unless you beat me to it. Perhaps we can do it together as I am so very short on time!

 

Normally I run a fan profile in hwinfo. You should try making one. At 60C my fans are about maxed. (I don't like my stuff running any higher than 70C.)

 

thanks for this thread guys, it has been a great help, I have just installed a dell oem 7970m gpu and its running a stock vbios @1.0v with the newest 13.2 drivers. I have the 720qm cpu so I don't think getting power to the card is an issue for me, I have been using msi's on screen display to monitor temps, load and frame rates in both battlefield and far cry 3, I have noticed that once the gpu hits above 75 degrees it throttles down for 3-5 seconds. My temps on idle are pretty good, 37-40 so I believe the heatsink and thermal paste are working as best they can, I have seen you guys have had great success with 2 different vbios's.

Clocks at 750/1300 @ 0.925v
and
Clocks at 825/1325 @ 0.95v

Which do you guys recommend for a pretty much care free experience? Obviously I don't want to flash the card too often with a new vbios so if i could get it right first time that would be great. I have already downloaded the provided 0.925v vbios (thanks for this btw) but if you think the 0.95v bios would be more suitable can someone post this vbios for me.

In relation to the fan issue, if its any help to ye, i have tried both using HWinfo and setting both fans to max and also letting the system spool them up their selves and still get throttling once it hits over 75.

 

Have you tried undervolting it to 0.975v? You wont really notice any throttle at this voltage unless you are overclocking your CPU.

 

I tired this voltage with a clevo card over xmas and it was still throttling for me, that was once it hit 80 though, I don't know if the limit was higher because of the different make card or an earlier version of the drivers. I tired dropping the stock clocks to 775 earlier in msi but it made no difference.

 

It isn't really the voltage that matters in my findings.

It is down to the way you load the GPU. If you load it 0-100% stress and I mean in game 100% stress then the card throttles almost regardless.

It is sensitive and the process does differ from game to game.

With FC3 it works best to run gpuz render then quit and run game after 30 secs and let it sit at menu for another 30 secs to warm up not using hwinfo to control fans (native fan control) then load into game.

For Crysis 3 the game doesn't run the card at 3d clock speeds until the game literally starts so it is best to spool up the fans with hwinfo and load into the game. Once you have the game loaded and are ready to hit enter to start the gameplay you actually alt tab out and run the render tool in gpuz to get the clocks and voltage up. 10-15 secs is enough then quit render alt tab back in and hit enter fast not giving card enough time to drop clocks. The game runs without a hiccup even with cpu at a massive 23/23/25/27 OC and 87W tdp and the gpu at 900/1400. 950/1450 also works but not necessary as 900/1400 maxes the game smoothly.

The trick here is to allow the gpu voltage circuits time to acclimatise to the new 3d voltage before adding the stress of 100% in game load.

For a long time everyone thought the throttling was 100% to do with the cpu but it really isn't. The circuits are simply more sensitive due to the heavy load on it with powerful cards. Stepping up the actual load on the card in a gradual manner and therefore the power needed gives the circuits a much needed chance to keep up and provide the card with the power it needs.

 

If others can put up advice on how to kill throttle with other games that I haven't tested please do!

I will sort out BF3 this weekend.

 

Ok new update.

I haven't played BF3 yet but decided to see what my 7970M could do and perhaps ultimately what the M15x can take now that we have an effective method to stop throttling.
I found my sweet spot to be 960mhz/1400mhz at 1.05V with cpu at 23/23/25/27 at 82W/65W tdp/tdc values. With this I ran for quite some time in Crysis 3 without a problem.

I then pushed the core to 975mhz and found I needed 1.075V for stability. Then I pushed to 1000mhz and the game crashed. Clearly my ES 7970M likes alot of voltage! OEM chips probably perform better. Oh well it was cheap

So I pushed the voltage to 1.1V and set the card to 1000mhz/1500mhz and ran once more. Amazingly the game ran fine and perfectly smoothly HOWEVER then the system just shut off. Evidently 1.1V with a decent CPU OC on the 920XM is too much for the M15x to handle.

My conclusion thus far is best to keep to stock 1.05V and see what your stable limit is. Perhaps somewhere between 970 (me) to 1000mhz (lucky!) or more if you are really lucky!! Also at these core speeds perhaps the memory could be set to 1450mhz for better bandwidth. I doubt any higher would help unless going for 1050mhz + which the M15x cannot handle unless you are running 720-840qm.

Perhaps someone with one of these cpu's could try to see how far they can push! It is good news however. Stock voltage still gives a decent overclock and with the next gen of cards which are rumored to be more power efficient we should see even higher OC headroom

 

What are you using to change the voltage so often? I have a 740qm so I could give it a go. How are you managing to keep the card cool with such a high voltage and overclock? Also Is it possible for someone to upload the dell .95v 825 / 1300 vbois talked about a few pages back?

 

Sent you a PM. I can also share a few vbios files at various different clockspeeds and voltages if you wish. However I do not take any responsibility for what happens if you misflash or damage your hardware.

 

It turns out actually that since this is a new card and the stock bios is 015-21 that you cant use that program to undervolt cards with the newest vbios already on them. I flashed a .975 bios earlier, version 015-17 and it caused a few issues. Back on stock now. My thoughts on it are that the hardware has changed slightly hence the difference in part no.s too from 117 to 118. Looks like I'll just have to wait until slv7 updates his program to make a custom vbios for myself.

 

If I send you a vbios file and you flash it. Then you can use the program

The hardware won't have changed at all. You just have the newer stock vbios which can't be modified by slv7's program.
Once your card is flashed with a vbios that is older ie like mine you will have no problem flashing.

I doubt SLV7 will update his program until he upgrades to the next gen as there is no point as the older vbios can simply be used.

Let me know if you want it and at what clocks and voltage.

 

thanks for the offer but it seems like if your stock bios came with anything over version 17 then if you flash it back to version 17 you are going to have issues and then change in voltage is not going to happen. here is a good thread that i have been reading that might explain it better than i just have. AMD 7970m - modified VBIOS

 

I thought the throttling was only an issue with the Clevo edition of the card. Is it a problem with both? I was going to order one tonight to replace my 5850 as I thought it was plug and play (ignoring the heat sink mods).

 

There are no heat sink mods? Any card can throttle in the M15x even the lowly 260M. It is well documented. I have a dell card.

Anyways it isn't a problem any more as this thread addresses and fixes it.

 

I have both a clevo and a dell card. Getting your hands on a dell one is a way better option. The fans run nativaly rather than you have to set up hwinfo to control them every time you start your system. You will also get hdmi out with sound working without issues. Some claim they got it working with their clevo card but I couldnt no matter what drivers and bios combo I tired. Best card I would recommend at the moment for the m15x would be a dell card with a 015-17 stock bios. I paid the same for both cards so hang out for one if you can.

 

Agreed. Even if it is a dell QS or ES like mine it will work fine.

 

Great thread. I have been reading bits of it for weeks just decided to join. I love that you are using voltage so much lower than stock, but I have a few questions. I use the 720qm so I don't think power draw is nearly as much of an issue as you 820+ users. It's been so long since I messed with my GPU I forgot what voltage it was, so I put the vbios from gpuz in RBE and it didn't work said it was invalid, but HWinfo says the max voltage is 1.05 and the lowest is .825. I don't even remember if I changed the voltage since I got the card lol. Anyway, I am not sure if I have quite the same throttling issue as you guys. Is it like for 1 or 2 seconds the fps will drop really low or is it more prolonged? I have noticed however that the gpu usage is pretty low in some games, could be a cpu bottleneck. I also have sometimes random shutdowns while gaming intensely. This problem seems to go away for long periods of time then reemerge for new games. Crysis 3 I am having some problems sometimes shutting down. I do not know if lowering the voltage will help with this, but I am willing to give it a shot.

So to sum it up, with a 720qm, and random shutdowns imminent, what kind of vbios should I try? I do have svl7's vbios package, so I can use any of those or the one on pg 3 of this thread.

 

If you have any CPU other than the 920XM/940XM there shouldn't really be any throttling issues. The XM cpus use 10W more power. Best way to check throttling is to enable monitoring of the core clock in msi afternurner. It should be 850mhz on a 7970m but when it throttles it drops down to 450mhz.

For your random shutdowns, it sounds like a temperature problem. At 86C these cards shut down it seems. So you need to make sure the temps are getting to high. HWINFO can be used to control the fans manually. You should check out how hot your temps are getting, MSI Afterburner can be used to check gpu temperature while in game.

BTW how is the performance of Crysis 3 on the 7970m?

 

Have you tried grinding down the x-bracket or at least try putting copper shims?

 

Um, yeah I should def check to see if the clocks are going down when I think it is throttling, but no its not a temp problem the highest it ever gets is 75 in crysis 3, 70 in bf3 or other games. And the performance with the 7970m is well, it looks unbeleivable obviously but i only run it on like low/med settings. Keep in mind that is low-med crysis 3 settings lol... there is not any actual low.

onego, is there a thread that talks about those modifications? My temps look fine on the sensor but if it was a problem on the back of the card I don't know if that would detect heat. I think the x bracket for this one did not fit the heatsink and i stuck on the old one from my 6990m, so maybe that caused the problem.

update: So I played crysis 3 multiplayer (i think single crashes more often) and was getting like 60-100 fps, temps on gpu about 75, I did not see any throttling with afterburner. However, after about an hour of gaming the computer shut down

 

Crysis 3 can be run at 1080p with all graphics settings on "very high" and 16x aniso and 1 xSMAA enabled (any higher antialiasing option and the vram is maxed I once hit 1975mb usage at the above settings and typically it settles around 1700-1800mb!)

This with CPU at 23/23/25/27 at 82W TDP, 65W tdc. All I need to run 23x at 100% load.
7970M at 950/1450 at 1.05V. Temps max around 75C-80C in ambient 20C

The game runs very smooth indeed with no throttle so long as you follow the anti throttle instructions I posted earlier. CPU after prolonged gaming sessions at the settings above gets to similiar temps as the GPU, in fact pretty much identical!

@darthvoldemold. Read this thread thoroughly and you will learn how to stop throttle and then you will be able to max crysis 3 and every other game out there. Maybe not BF3 as your 720qm will hold you back a bit but still much higher than low/med.

It is highly advised you upgrade to the 920xm if you enjoy gaming alot.

 

Two more things temp wise I can thing of:

1) Did you properly install the thermal pads on the areas they are supposed to be installed on.

2) Can you check how hot your cpu gets? It's possible the gpu temps are fine but cpu temps are too high.

 

Modifications regarding the X-bracket could be found here in the post below:

[The backplate problem part 2]
[Guide] M15x with AMD 6970m / 6990m - everything you need to know

Your shutdowns may be caused by the vram modules being too hot as well. Check and see if your vrams make contact with the heatsink and make sure the thickness of your thermal pads are just right.

Lastly, try undervolting your card to 0.975 V to see if it resolves the problem. No harm trying.

 

Just throw away the M15x ^^ it's obsolete now, take a clevo/sager P150SM , but It's a shame alienware discontinued the 15" serie it was a stupid choice, the 14" is weak and quite overpriced compared to the concurrence.

 

@king of interns: after I read your post I did go back and read the rest of the thread and it turns out almost all the information I wanted was there, a good lesson learned haha.. I still won't read the 150 pagers though. I have wanted to get a 920 for a long time and I thought the prices would go down over the years but it seems not... I'll probably get one before crysis 4.

@ssj: When I bought the card, they told me to put heatpads on some other boxes on the card (sorry for layman's terms) other than the vram, but I think I tried that and it did not fit for some reason, so I have settled with the same configuration as past cards, heatpads only on the vram. If there is a thread that discusses exactly how to put the pads on the accessory boxes, I would love to see it. And I don't think the cpu is getting very hot.

@onego: That is a cool thread, I wish I had seen it when I got the card because it came with the tall things on the bracket, and I could have just filed them down but instead I replaced it with my old one. I will definitely check again to see if the vram is in contact, I'm pretty sure im using 1mm pads. And after all that, I will try undervolting.

I also found svl7's program by reading this thread, which I had heard about it turns out I just have to post more on techinferno then I can download it...Can't wait to see how that works.

 

Try using 0.5mm thermal pads with the grinded down x-bracket. This will most likely solve your problem.

 

I'm pretty sure my M15x can take on a P150. It's not obsolete yet.

If you put the thermal pads the way they were on the previous card you should be fine. Have you tried using hwinfo to control the fans and try running with the fans @ 100%?

 

Onego, I will see if the pads the seller gave me are .5, otherwise i only have 1mm, but I didn't really like that bracket because it seemed like in addition to being too tall, the screw holes were a little bit small for the screws in the heatsink. In fact, I might have read a thread where somebody installed new screws to use that kind of bracket, but maybe im just imagining that.

I started using hwinfo to control the 7970m fans because I beleive its a clevo (bought from upgradeyourlaptop on ebay, thats eurocom right?) so it goes to 100% on boot, and in the fan control browser it said "active" under the cpu, so I thought it had put a custom plan on the cpu, im still not sure if it was doing that but now aday I use hwinfo for both. I have never tried moving fans to 100%, but it would not make sense because when it crashes, i have been gaming for an hour usually at max fan speed.

edit: So I decided to try playing crysis 3 with fans pushed to max on both cpu and gpu, and I have to say it might have helped, my reasoning being that I just played about an hour+ of crysis 3, and just for fun I kept ramping up the OC to see if it would crash the computer, I maxed it at 850/1400 and it was stable and beutiful.

 

it isn't so much fan speed , although that helps stability. it is the way you ramp up load. you will be good for at least 950/1450 like me if you read the last couple of pages. i will update the first post

 

Hm. You get 950/1450 at 1.05v? I just tried 875/1400 and it was beginning to throttle a little bit. Does throttlestop stop gpu throttling? I usually dont run it but I tried it and it didnt seem to make any difference. I'm about to open up my computer and make sure about all the contact and stuff people talked about earlier, and I might do the filed down x bracket mod. Heres a few pics of what I got with my 7970m from upgradeyourlaptop. I'm not sure if those screws are supposed to match the bracket, thats what it looks like, but there are already screws in the heatsink, which is why I didn't use these. And the little black ones... if I remember correctly those hold the gpu down onto the mobo but I already had some.


Update: So I spent a pretty good while looking at the gpu, the two backplates, the vram and its pads, and this is my conclusion. I wanted to file down the tall backplate and use it, but I confirmed my premonition that the screws on the heatsink did not fit it. However, the screws in the baggy in the picture do fit the tall backplate. I was going to remove the screws from the heatsink, but I thought "hey what if I can't get them back on" so I decided to come here and ask.

Has anybody on NBR removed the screws from the M15x heatsink? I would love to see a thread.

Besides that, the vram pad that is close to the copper pipe on the heatsink (but not the other one) had little spots of some kind of condensation on top in the middle of where each module is. I wiped it off with my finger, it almost seemed like a thin film of either water or some other liquid. This gave me the suspicion that maybe there is a coil pushing the heatsink away from the gpu, causing low contact with those vrams. However, I tried to look sideways to see if it was making contact, and I could not really tell. So I just repasted it up with IC diamond, blew all the cat hair out of the vent, and shut it back up.

 

I had my 7970m flashed for 0.975v but I'm seeing that since I'm running a 740q CPU it's not going to really benefit it. So I won't to get my power back to stock, I tried flashing the regular Dell7970m.ROM twice and my power is still stuck at 0.975. Anyone know what's going on?

 

I actually just tried flashing the 1000.0 uv bios and it's still stuck at 0.975 according to hwinfo64.

 

Do you run gpuz render tool before you launch into gameplay? Which games are you playing. I have only played Crysis 3 and Far cry 3 so I know the method for those games only. I am confident the approach can be adapted to any game though allowing for max stable frequencies at 1.05V in any M15x with any cpu including the XM when highly OCed.

 

I have been playing basically just crysis 3. I was actually trying the gpu z render thing 2 days ago and was having really good performance and OCability, but I wasn't sure it made any difference so then I stopped. I'll try that again and report back. But how does rendering before playing the game influence whether or not it throttles throughout the gameplay? It only makes sense to me that it would help with throttling occuring as soon as the game starts.

 

Give it a try, it seemed to help my C3 run better.
How did you get your msi afterburner to OC your 7970m? I tweaked my config file for it and now I can adjust the sliders but they're still maxing out at 850/1200.

 

I also did a tweak in the config file. Look under the section called [ATIADLHAL] and find the line that says UnofficialOverclockingMode = 1 and make sure its a 1.

There is one annoying thing though, when I overclock the card it seems to cause screen tearing or artifacts of some kind for a few seconds, but only outside of a game. After I go in a game and then come back it looks normal. It's pretty weird and annoying. Must have something to do with 3d clocks running 2d programs.

 

That's exactly what I did. UnofficialOverclockingMode = 1
And now my clocks are adjustable (the sliders) but they won't push past stock 850/1200. It just reads that as the max. Did you flash a vbios with higher clocks?

 

Crysis doesn;t run the card at 3d clock speeds until you literally start gameplay. What the render tool does is to give the gpu a lesser load than the game. Which in turn means a reduced load on the gpu voltage circuits. Doing this first allows you to then play the game fluently and even at an OC without throttling as the voltage circuits are given a more gradual increase in load. A sudden increase causes the throttling problem.

even 1.1V at 1000/1500 worked fine. However the M15x doen:t like overvolting. It shut off after a short while during crysis 3 so I settle for stock voltage. Which is plenty anyways.

 

Im kind of confused. I already knew about the 3d voltage in crysis 3 starting when the game starts and the gpuz thing being less of a load. Why does the throttling happen the whole time you play if the voltage jump is only at the beginning? Shouldn't it level out or something?

Anyways, I don't think I'm even having a throttling issue because I look at the gpuz or msi afterburner and it says the clocks have been stable, but in game the fps is not stable at all. Also, the gpu load when this is happening is 30-40%. When I have played crysis 3 and it not been having this fps issue I look at the msi afterburner and it says the load is 70-80%, which reflects the higher fps. My confusion, now, is what is the problem if the clocks are stable but the gpu load is so low? I can't tell any difference between the times I played and had good fps vs the times I played and had bad fps, other than the gpu load.

 

There definitely is still throttling even though it may not show up through your clocks. At least for me.
I'll be in a new load spot in C3 with 25fps and it won't budge from 25. I restart the game, load the same spot, ramp up the fans and temps with render test and jump into that same spot and Ill have steady 40fps. But it doesn't last long for me. It's just a matter of time before it gets unstable again.

 

Maybe you are rushing the process. Please do in this order.

1.) Ramp up fans
2.)Load Crysis 3 until you are ready to launch into gameplay BUT don:t enter into gameplay.
3.)ALT tab out and in gpuz run the render tool for 30 secs.
4.)Then quit the render tool and alt tab back in and hit enter to start gameplay.

The game wont throttle.

 

Yeah that's exactly what I've been doing.
It seems to work for about 10-15 mins. This is with Low settings. Once I start messing around with anything higher then it seems to throttle.

 

Hmm that is odd. I also find if I do the above from a cold boot because temp is lower stability is higher. Either way though I can always defeat throttling this way.

I am surprised you are having issues. Stock clocks is enough for 1080p as very high everything at 30-40fps. Why are you running low?

I have got so little time but I intend to give BF3 a blast again and see how my technique works on that one.

What clock speeds and voltage are you running lambda? I have a feeling you have it undervolted too much so the card isn:t fully stable. Try the next increment up in voltage and re^try you might be surprised what you find ;D I say this as in my experience if the voltage was set too low for the clockspeeds set then gameplay slowed to a crawl even though no throttling in gpuz was apparent. Upping the voltage allowed the card to take the power it needed to run properly.