M15x & 6970M / 6990M - What you need to know - upgrading, problems, solutions

Spent another night working on the card and came up with some weird results. I also have a better theory for what might be causing the shutdowns for some people.

I pointed out the last big post that I didn't get shutdowns, but the card would reset to 680 @ 1.1v at random times during Crysis 2. The weird part was I've changed that setting in the vBIOS, there's no 1.1v setting available anymore. I've now found that it does this reset in OCCT as well, but only when the CPU is also being used somewhat heavily.

This leads me to think it is a power draw problem and not temperature or stability related.

I locked my CPU to 1200MHz and was able to get up to 820 @ 1.1v stable, with no downclock or reset in OCCT. Temps maxed at 96C (Too hot for everyday use.) When I set the TDP back to 65, it reset the card in less than 10 seconds. I then tried to 665 @ 0.95v and ran prime95 in the background of the OCCT test. Something weird happened, the voltage reset to 1.1 after a few minutes, but the downclock stayed. This happened all the way down to a 35w TDP on the CPU.

I think what is happening for the people who undervolted to stop their shutdowns, is that the voltage is reset after a time, the motherboard sees a total draw that's more than 45w + 75w, and it shuts off. The XM's may have more leaway because the board expects 62w + 75w. It's just a theory, but it fits pretty well with I've seen and heard. I'll keep testing.

 

so my local tech was right all along, now i get what he was trying to explain with the shutdown issue, either heat or power.

someone in the forum much a while ago tried reducing the clocks and it does prevent them from shutdowns.

anyways i ordered the last resort, a 920xm coming in a week's time, hopefully it does solve the issue and confirms this theory.

thanks for the write-up again johnny, you deserves a beer.

 

Wow, great work Johnny!! That makes sense especially when you know that Skyrim is cpu limited.
Does that mean that the card must run at 1.1v to avoid shutdowns?

 

Thanks! Hopefully it gets to you quickly.

I think it resets to 1.1v thinking that's a lower power setting than what's being used, but it's actually higher draw, which then trips the machine off.

I tried flashing all the clocks in the vBIOS to 710 @ 1v, then disabling powerplay in windows when I wanted to game. I hoped that would force the clocks to stick... it doesn't. I think all this might be related to the 460m throttling issues. I'll do some more work tomorrow, still have 5 days of vacation left.

 

Interesting read. Goodluck with finding out more and have a good vacation!

Sent from my SCH-I405 using Tapatalk

 

@JohnnyFlash: Can you post the VBIOS you're using?

 

Interesting theory so you think if the CPU and GPU use a combined power draw of 137W with XM installed then it shuts off.

You might be on to something however one problem remains. Slv7 tested even a heavily overvolted 920xm and reported the m15x can and will draw as much power as the 920xm requires or until it becomes thermally unstable even when that results in a 200+ W power draw at the wall!

I guess it is more sensitive is the 6970/90 pulls more than the 75W at the wall. As soon as this happens it will crash?

I had just one shutdown recently and that was when I ran my 6990M at 750/900 at 0.96V and the cpu at I think it was 21 or 22x the machine shut off after 30 mins of folding.

On the other hand I can use the cpu at up to 25x (very good chip) across 4 cores for video encoding 100% for hours and even using the 150W it won't skip a beat.

So I think we are definitely GPU power limited which is a shame. Perhaps if the new series of cards doesn't work I will downgrade to a 6870M lol

 

The systems definitely does not shutdown when the power draw increases(except the PSU trips, which you would notice). I also overvolted my 6970m a while ago to 1.2V and more and I had no shutdown issues, only throttling.

 

The problem comes when you hit more than the expected power draw I think. You also need to add draw from the LCD, HDD, board and other peripherals to that total. On my system it's what the CPU is doing that determines whether the GPU throttles or not.

I haven't had any either.

If anyone wants to test this: Just load up Prime95 to 8 threads, then run OCCT's GPU test at the default settings. Watch your clocks with GPU-Z and the voltage with Hardware Monitor.

Remember, stressing can always be dangerous. I'm not responsible for anything that happens as a result.

Here's my vBIOS, stock and modified.
View attachment 6970m.zip

 

I guess the next question is, how do we deal with this power draw without upgrading to the 920xm or 940xm?

 

Upgrade to a 28nm GPU in the future or use a 6870M which has lower power requirements for now.

I will try this test when my 240W power adaptor arrives.

 

I have the 240w adapter already, the machine acts the same on either for me.

Some more information:

Using Prime95, OCCT and 715 @ 1.0v as my test setting I have the throttle settings worked out.

With 8 threads loaded in Prime, the anything higher than a 13x multipler makes the card throttle.
With 4 threads loaded and affinity set to each physical core, anything above 14x throttles the GPU.

Playing Crysis 2 I can go up to 24/24/20/18 @ the stock 62TDP.

 

Would this behaviour continue with a lower power GPU realistically.

The 5870M was a really low power part vs the 6970/90M. i hope the 7900M series of cards can take the power consumption back down to 75W area.

100-150W is too much for a laptop anyways. Which is what a 6990M OCed will suck from the wall. The 580M will do that without even trying lol.

edit: I have just started a test at 15x on the 920xm and the GPU at 0.96V and stock clocks and no throttles apparent.

Using Furmark and Prime 95. Core temp and GPUz 0.5.9 to monitor things plus the temps in furmark show the clocks haven't throttled at all.

Shall I test further and push the processor up and see if throttling occurs.

Ok I did just that! I upped the tdc and tdp from 50/50 to 50/60 and stuck the multi up to 17x and the PSU tripped off turned the laptop off too.

Had to reset the PSU to get things going again.

Mind you pushing the CPU up to 25x at 87/60 100% load across all cores + HT for hours the laptop can do no sweat (92C ) as I do this often when encoding. This time hower the GPU was running with furmark.

So am I suffering from the same problem or just exceeding what the PSU can supply.

Wow three posts in a row. I recieved the 240W power supply tonight and continued to test. I set up Prime 95 and furmark again with exact same settings as before then after 5 mins when temps had loaded I increased the tfp to 60 again and increased the multi to 17x. No crashing and more inportantly no throttling or switching off. Also I then increased to 62W which is enough for 18x and kept going still no throttling on GPU or other issues.

My temp in furmark however isn't good. 86C on hottest sensor after barely 5 mins! Need to address that soon!

 

I hope there is a solution that doesnt involve new parts x.x I would prefer to not have to keep buying new things.

Sent from my SCH-I405 using Tapatalk

 

Looking at this issue perhaps it is more a question of luck than parts.

 

I just looked at the modified vbios you posted, according to the flag in the file you used RBE. However, there are still two entries which read 680/900 @ 1.1V.
So it should be possible to circumvent this resetting with editing these entries.

 

How can you see those? I'm using RBE 1.28 and I can't see them. Do you see an entry for 250/900?

I really wish they made 220v killawatts, then I could test and tweak actual draw.

 

Just upgraded from the 260m to the 6970 and so far so good.

Stock clocks
Idle temps are 47-50c (I think this is on the higher end of good, but still okay). My gaming temps are 70-80c which, from what I've read, is normal. I'm using the 260m backplate with no issues. I have the 6970 backplate but I can't figure out how to get the screws off of the heatsink in order to replace them with the screws for the 6970 backplate. This doesn't seem to be an issue though so I'm not worrying about it.

 

if you have it, try Alan Wake. The game will shut down almost every time within the first 5-10 minutes of playing from the beginning. I've also had it shut down on Skyrim. Other than that i haven't had shutdowns however I never play BF3 but i hear that's another game that shuts down.

 

I played Skyrim for awhile without any shutdowns. How long does it typically take for the shutdown to occur? I do have the i540m instead of the i7 that most people seem to have. What kind of temps should I expect when running Skyrim?

Just played about 30 minutes of Skyrim on Ultra (with AA off) and didn't have any issues. I've attached the GPU-Z Sensor Log.

Without going through all the pages here are a few links regarding thermal paste application that may be useful:

Arctic Silver Application Methods


General Thermal Paste Application

 

I'm using a hex editor, but I can't seem to find a 250/900 entry, at least not on a quick look... are these the clocks when connected to a second display? Might be implemented differently... but that would be weird, hmm.

There are 220V kill-a-watts, I have one at home myself...

 

Ya, 250/900 is the dual display setting. I would change it to 250/150, which performs the same in benchmarks; maybe I need to contact the RBE guys.

What's the model number of your kill-a-watt?

 

just received the copper shim but i have no idea how to install it. can someone give me some good instructions. i checked the site it told me to go to but im still confused

 

Just apply paste to the die the normal way, place the shim on top and push down just a little. Put more paste on top of the shim, then put the heatsink on. Be sure to start tightening closest to the vent first, one turn on each screw at a time.

 

Man, so it's just luck on whether or not our cards work? Is there any way for the CPU to draw less power, or am I not understanding the power draw issue correctly?

 

Lock your CPU lower with throttlestop; that should work for now. The next step is figuring out how to fully edit the vBIOS to set the lower voltages across the board.

 

Do I just lower clock modulation/clock mod, or how would I go about doing this, and how low should I take it? Sorry for all the questions >.<

 

Which driver are you using? GPU-Z can't read voltages on the newer versions.

Set multiplier: Something less than turbo. Do it in another profile so you still have full CPU at the desktop. Just switch when you go to game. In my tests 14x seems to be the mark to aim for.

 

Hm, thanks. I'll try that tomorrow and see how it goes. Got a paper due in 7 hours haha..

 

Yep, no worries. I pistol-whip my students if I get anything late.

Also, I made a post in the RBE forums. We'll see if anyone responds.

 

ok thx for clearin that up for me johnny

 

I'm using the 12.1 drivers that were released in December. Should I upgrade the one released in Feb?

 

No, they're fine... I'm just shocked it could read the voltages.

 

I went ahead and upgraded to the 12.3 drivers and GPU-Z can still read the voltages. I did notice that when I disabled PowerPlay in CCC, it wouldn't allow me to make voltage changes... for whatever that's worth.

I do have a question though - my idle temps are about 47c (.800v) which is a little higher than I like. I've repasted twice to make sure I had good contact but it still doesn't seem to be dropping any lower. I've got thermal pads in all the sweet spots (except for the big pad underneath the card). I tried undervolting down to .700 and it only dropped the temps down to 46c . Any ideas on getting the idle temps down to the low 40's??

The only other thing I've noticed is that my thermal diode sensor is usually a degree warmer than the MemIO sensor when on idle clocks. Under heavy load with 3dMark11 the thermal diode won't get over 75c (even when OC'd to 800 on the core), but the MemIO will briefly touch 91c. On stock clocks, the temperature on the MemIO will peak at about 85c during gaming. I guess my question is... I dunno.. is that normal? I thought the MemIO would always be warmer than the other sensors?

 

Which version of GPU-Z do you have?

As far as temps go, those are really good, idle temps aren't going to go much lower than that; mine idles around 50. The MemIO is almost always the hottest, that's fine. If it hits 90c at stock, then maybe it needs a cleaning or repaste.

 

I'm running GPU-Z 0.5.9

Alright, I feel better about those temps.. I got it in my head (from the first post in this thread) that 40-50 is normal temps for idle so I wanted the lower number Yeah the memIO doesn't get that high on stock clocks, so I guess I'm good.

 

After i repaste my temps will be about 38-42 but after about a week they will settle around 48-50, and memI0 is alway higher sometime up to 10c higher while @ 100% load. I have never seen voltage on gpuz hwmonitor for gpu voltage reading

 

I've attached a screenshot of the voltage reading in GPU-Z and ATT.

After screwing around with undervolting, I've gotten temperature down to 44C, and I'm happy with that.

Although, after 2 hours of Skyrim, the memIO temp reached 95c a few times. That seems high, but the other temp sensors were all high 70's to low 80's. It didn't seem to affect performance at all though... still was getting 60fps.

 

ah, so i5's is in the clear of shutdown ? thanks for the info

 

32nm 35watt TDP, big advantage.

 

I'm continuing to undervolt - chasing numbers at this point, but I'm starting to not trust what I'm seeing. Is this undervolt actually possible or am I just getting false readings?

 

These are wrong readings. Lowest voltage available in the vbios is 0.8V.
Besides, 0.65V wouldn't work anyway.

 

I'll take another look at at the vbios when I come around to do so.
The kill-a-watt is a "Brennenstahl" model "PM 230".

 

@JohnnyFlash, so for my CPU, the setting right below Turbo is 20, is that still too high or is it better to just run tests/go lower?

 

You mean multi? Are you running 20x?

What are your temps if you run prime 95 or do some video encoding. Most XMs ought to be stable at 19,20x

As I type I running 20x to do some encoding which luckily for me is well below my own chips limits however you need to test things in tiny increments as you stress test to see how high your XM can go and at what tdp/tdc.

For example my chip is currently running 20x rock solid across 8 threads at 100% load and through testing I know I just need to dial in 67W tdp. My temps with fans at full speed are under 80C.

Try 67 TDP and 65 TDC yourself and then see if your chip can do 18x on all 4 cores. Then while the test or encode is still running if stable push it up to 19x and so on. If the multi starts fluctuating downwards bump the tdp upwards a little until it becomes stable again. Don't at any point just dial in a really high tdp and tdc and then take the multi up way high. This will just mean you will throttle instantly or shut the machine off and you will be none the wiser of what your chip can actually do and what it actually requires for a certain OC.

Use HWinfo to put fans to full speed before you stress test so your cooling has a head start and run core temp in your sys tray and make sure your temps stay under 95C.

I have run encodes for several hours at 25x which for me demands 87W tdp and 60W tdc but the temps hit 93C on hottest core. I think therefore 95C is probably still just about safe. Especially as this chip is basically a desktop 920 which can handle alot of abuse...as it runs at a higher voltage then its mobile cousin.

 

Well, in Throttlestop the next choice after Turbo is 20x and then on and on; I have i7 620m also so I'm not sure how much your things would match with mine.

 

Oh wow, that's an important piece of info. That's only a 35w chip... and you're getting shutdowns. Set it to 18 and see if you still get them. It might just take some testing.

 

johnnyflash. I have the same set as you 6970 and 940xm both are es chips.
I had had my laptop stable with the 6970, but once i installed the 940xm it started with the shut downs again. Kind've hard to keep both 6970 and 940 from building up so much heat and causing shut downs. What's the secret to success lol?

 

i tried under volting my 6970 it helps a lil but notice my card looses performance.

 

When i run 3dmark11 on 1.0v at 680/900 one of the test keeps getting video driver not responding.