P570MW GPU temp difference.

During Valley benchmark (just leave the program running for 1-2 hours), my secondary card (picture) gets almost 15 degree hotter (85 vs 70 degree) than my primary (wothout OC)

I did try to re-paste with Liquid Ultra and that just lowered both GPU's but I still have tje same temperature difference. Before I repasted the GPU's, the temp was 93 and 80'during Valley.



Is this a design issue that the middle card always will be much hotter?

okey, I see that the GPU I have posted a picture of, is the primary by looking at the SLi cable.

 

Stock or modded vbios? The modded vbios is much hotter than the stock one. If you don't need the speed or you're not benching, run the stock vbios.

Crazy that Ultra dropped it THAT much, hopefully it does the same in my 9377

15 degrees is a huge difference... Does that card happen to be next to the CPU? In my system it's the slave that's hotter.

Sent from my HTC One_M8 using Tapatalk

 

Im running modded vbios yes and I do recall you telling that you went back to stock because of this.

It happens to the card next to the CPU but the CPU is not gettinf hot during Valley..

Why does the vbios run much hotter than stock on the same speed?

 

you're using 880M cards? Hmm... the modified 780M vBIOS is actually COOLER than the stock one. If on 880Ms then that might explain things. But still so weird. My 780Ms stayed within 3 degrees of each other before I originally sent my machine to Mythlogic. Afterward, my master 780M would get 6-10 degrees hotter at times; though after raising the back of the computer the fan I use for cooling both myself and the laptop blows almost directly at the slave, and I never had the machine raised before, so that may explain things (though it varies greatly per-game).

If liquid ultra helped, that's cool, but yeah, what can I say? Try the stock vBIOS and see if it gets cooler. It shouldn't be hotter just because of a vBIOS change; that's so weird. I would understand if Ethrem's machine is an anomaly (which it is, since his slave is hotter than his master), but those temps are crazy. *I* don't even get those temps in Unigine Heaven 4.0 with everything turned all the way up AND 950/6000 clocks on both cards, and your cooling is superior to my P370SM3

 

Keep in mind that the modded vbios forces the cards to run at 993MHz all the time while the stock vbios uses different degrees of throttling depending on the workload.

I don't know why my machine is an anomaly - it only makes sense that the slave card will be hotter since its further in the chassis and has heatpipes to transfer the heat to be dissipated by the fan while the master GPU is right next to its fan so it has shorter distance for the heat to travel.

Anyway, the increase in heat is because it removes the auto-throttle that these cards have. Heaven runs 954MHz on one card, 918MHz on the second with the stock vbios because of the strain. I don't remember what Valley was running but I have yet to find a benchmark app that will boost to the 993MHz clock with the stock vbios. Its a bit *too* aggressive in my opinion but it is what it is and its certainly better than 93C...

 

You can test your theory by swapping the cards/heatsinks over but it could be down to paste job and sample variations.

 

If your second card is at 918MHz then your SLI is at 918MHz. This means that you aren't even seeing the 954 benefit of your main card. Secondly, the slave GPU almost always runs cooler due to the main GPU having more workload (due to memory only being copied to the slave cards but CPU-fed to the primary one I believe), it's actually a pretty common thing. A slave card getting hotter is cause for concern for most people that something's off with the slave card; especially if it's a large temperature difference.

Also, in a large number of your posts you claimed that in games your cards didn't throttle at all when gaming and only in synthetic benchmarks... have you tried using the modified vBIOS and then subsequently downclocking your cards to 954MHz on each card and seeing if the same heat issues take place? Because if your slave card is throttling a lot then your main card will never actually be using its full 954 or 993 clocks, as SLI simply adopts the effective clock/memory speeds of the slower card. In which case, the stock vBIOS is indeed actually harming you; if you could simply downclock via nVidia inspector or something on the modified vBIOS but still get a purely consistent experience.

 

Funnily enough in my P370SM the slave always ran 1-2C cooler when I used the stock 780M heatsink. But since I started using the 880M heatsink for the master 780M the slave now actually runs 3-4C hotter!

But that's because the 880M heatsink is quite improved and cools the master 780M better, as the absolute temperature of the slave 780M still remained the same. But after lapping down the slave 780M heatsink both cards now run at about the same temperature (within 1C of each other).

 

Well that's true, but your 780Ms aren't running at 993MHz on the core, I think. Though, I still never saw my slave card get hotter than my main. I think Ethrem does indeed have an anomaly.

 

Well I've been considering swapping the primary and secondary in the machine for some time now but I'm almost out of IC Diamond. I have GC Extreme coming tomorrow and when I get my Liquid Ultra early next week I'll put the liquid ultra on the CPU and then swap the primary and slave cards and repaste them with the GC Extreme and see what happens.

Sent from my HTC One_M8 using Tapatalk

 

The heatsinks are interchangeable so no need to take the cards completely out

 

Ethrem is talking about a P377SM-A though (or maybe a P370SM-A... one of the two; same difference really)

 

Yeah I was talking about my 9377.

Pidge is looking into the problems I've been having with 880s, I just gave him a whole boatload of evidence to take back to investigate so we'll see.

I'm going to swap the master and the slave when I put the Liquid Ultra on my CPU and see if they run the same. I have a feeling it will be the same behavior but it can't hurt.

 

The slave running a bit hotter is normal I think, given how long the heatpipes are (2x the length?) compared to the master heatsink.

 

Well I never had it, and I never saw anyone (including yourself, pre-modding) arrive at a hotter slave GPU... so it strikes me as odd. Even in desktop scenarios it's usually hotter master GPU.

 

Modded vbios, confirmed the slave card is the hotter of the two by setting single GPU in nVidia control panel and running it again.

Spoiler

It would appear that the reason the P377SM-A has throttling with the 880Ms is because of Clevo's engineering. There's a 10C difference between the slave and master. The only reason it doesn't exhibit that behavior stock is because of how aggressively the vbios throttles with temperature. Its easy to see that happening here where the slave card is throttled to maintain the same temp the master has.

Spoiler

This is a picture of the heatsink assembly in my machine.

Spoiler

The way its designed, it doesn't surprise me the slave is the hotter card.

 

Lapping the slave heatsink should improve the temps. Everything stock the master was hotter by 1-2C, with the master 880M heatsink the slave became hotter by 3-4C. Lapping the slave heatsink now makes it so the master and slave are within 1C of each other.

 

I don't have the patience to sit there and lap a heatsink and the slave heatsink is already fragile (the heatsink block will actually separate out from the rest of the assembly, the heatpipes turn). Hopefully its decent enough for Liquid Ultra then I can just call it a day.

 

Well I can say something is DEFINITELY wrong with your slave heatsink then... 10 degrees difference between master and slave where the slave is that much hotter is just crazy. I guess the 880M heatsink does indeed help the master GPU a bit... 10 degrees is a lot. At the rate you're currently at, you're not even getting full SLI benefits, because your slave isn't even hitting the 954 mark and thus your SLI functionality will dynamically operate at the slave GPU's speed at the time. I hope your Liquid Ultra works out. I asked Mythlogic to apply Liquid Ultra for me when they do the motherboard change (as they'll need to repaste anyway) but I didn't get a reply yet. I hope they do; that way I can get right into working with the CPU temps when I get it back. Honestly, the temps I was getting was pretty ok before I sent it up; especially at night. I can handle a little OC at those temps, and it'll help immeasurably with my streaming and games like bugfield 4.

And hey, if it helps my GPUs as well, I could crank up the settings a bit and still get a 120 constant in BF4 or similarly demanding games *glares at Far Cry 3* =D.

 

Well I have done more testing and my slave GPU is also consuming more power than the master. Perhaps switching them around is a good idea.

Spoiler

That's a run of Valley until the slave hit 90C. The master maxed at 87C but it was only at 81C when the slave was at 90C and I ended the test (had it on auto fans).

It shows the TDP being higher on the second card across the board though at the same voltage. Weaker silicon?

I don't know, I'm tempted to buy the base Alienware 18 off of Amazon and put my Clevo parts in it. I'd lose an mSATA SSD and my Bluray writer in the process but I'd gain much better cooling out of the box.

We'll see what the CLU does.

 

Check the ASIC quality with GPU-Z and see if there's big discrepancy between the master and slave 880Ms.

I honestly think lapping is the best option in your case. The contact plate must be pretty warped to get temps like these, and if you lap it flat you could see a drop of 10+C. I know deadsmiley went from low 80s to high 60s while gaming after he lapped his GPU heatsink.

Be careful with the AW18, it had issues with overheating because this time around Alienware decided to favor "quiet" over "cool", so the CPU and both GPUs run hotter than they should. Although with the newest BIOS that allows manual fan control you might be able to create your own fan curves -- check with Mr. Fox to see if that's the case.

 

How am I going to lap something so fragile though? The heatsink plate already will snap through the assembly with light pressure. It rotates. I can't really describe it but either way, it's not solid so I don't see how I would be able to even keep it steady to make sure it comes out flat?

Sent from my HTC One_M8 using Tapatalk

 

I know what you mean, the 780M slave heatsink is exactly like that as well. There's nothing to support the plate once the heatsink is removed, so it just kind of dangles there.

Maybe two pictures will help: (I used a 680M heatsink but the exact same procedure applies)


The basic idea is to use your index and middle fingers to press down on the contact plate to make it sit level with the flat surface (machinist block in my case), then use your other 3 fingers to gently lift the heatsink assembly so it clears the flat surface and doesn't run into it when moving the contact plate back and forth across the sandpaper.

Just make sure you don't use too much force and dent the heatpipes. I think the plate does dangle a bit lower after the 2 hour ordeal, but nothing the heatsink screws can't take care of. Initially I was also worried and tried to inch it back in place by bending the other way, but in the end it made absolutely 0 difference.

 

Also, a little hint about Clevos. "Auto" fans will *NEVER* under any circumstances, arrive at "max" fanspeed. Even when recording paranautical activity using playclaw with 70%+ usage on all cores on my CPU getting ~80 degrees and 99% constant scaling on my GPU getting 90 degrees (doesn't use SLI though), switching to "max" fans will take me down to 86 degrees or less on the GPU and I won't come close to 80 on the CPU. Max fans while gaming should have been something you were doing the whole time; auto-fans shouldn't be used for any kind of heavy workload as the most it ever hits appears to be something like 80%.

EDIT: This holds true for both my previous D900F and my current P370SM3. Auto fans never kick up to maximum on either machine. One was from XoticPC and the other from Mythlogic, so it's a direct BIOS fault from daddy Clevo.

 

I just don't know if I have skill and patience to pull it off, I've never lapped a heatsink before. Really this machine is my first time doing hard mods to anything. It gets addictive but it's also expensive. Lol

Yeah I noticed that the fans never go on max but the stock vbios does a decent job balancing the temps with power. When the fans are maxed in this machine, they are so loud they can be heard over the speakers on max and the sound is just horrible. I've actually had a few anxiety attacks from the noise getting to me so I try to avoid using them on max. The sound is so much more worse than my M17x ever got, even when doing a BIOS update (which was the only time the fans went in max), its like a high pitched squealing noise and its rhythmic to boot.

That's why I stayed on the stock vbios is because while it will throttle with too high of a load, the temps are managed quite well in the 80-82C range once the fans kick up at 87C. I wasn't aware until now that the throttling was actually taking both cards down to the speed of the second card though. It doesn't change the performance all that much in most circumstances - Watch Dogs is the only game that I play regularly that doesn't maintain well over 90FPS when maxed out.

Oh and master asic quality is 80.5% while slave is 86.3%

Sent from my HTC One_M8 using Tapatalk

 

Interesting, my master 780M ASIC is 72.9% while slave is 73.5%. If your 880Ms aren't outliers, then this lends credence to the theory that the 880Ms are simply higher binned 780Ms.

If it makes you feel any better this machine is also the first time I'm doing hard mods to anything (or even using a Dremel for that matter). Was definitely hesitant about lapping the heatsink, and I did botch my first attempt, but I redeemed myself with a good lap the second time.

And auto fans only come on when the CPU is >95C and GPUs >92C. Basically fans only go turbo when the hardware has already reached the throttling point lol

 

By the time the fans go max, its too late...

And yeah, this machine will be my first time with a dremel. I'm going to have to contact Xotic about getting a bottom cover.

I really think that this should be a warranty repair though since the video card is getting so hot it can't maintain its advertised clock speed. I mean 21C on a table with the back raised using a pill bottle cap on each side and still throttling? Even after multiple IC Diamond repastes???

Of course if I send it off, I'm without my machine for probably 2 or 3 weeks and they may or may not fix it...

 

Well, firstly, if the noise bothers you, you're in a really quiet room without using a headset. I suggest using a headset. It's not a full fix, but hey, if it keeps your stuff cooler and you get better audio (especially if you buy a good headset like a logitech g35 or corsair vengeance 1500 v2) then I don't see why not. You are right, the downclocking probably doesn't make that much of a difference in games. The only games that benefit from higher speeds like that are stuff like BF4, Watch Dogs, Titanfall, Sleeping Dogs, Sniper Elite V2... basically any game that has a super demanding graphical setting and in which you would... possibly... benefit fr... well moo, you have a 60Hz screen... well then.

Titanfall is buggy, you could benefit from a constant 993MHz on that, but only the master GPU would be used. It *NEEDS* 60fps constant though, or your aiming goes to what you'd find under the boots of a cross-country trekker. Sniper Elite can maintain 60 constant with SSAA on 4x at 1080p (read: 4k res supersample) with the 850/5000 clocks of the 780Ms... BF4/sleeping dogs/etc won't benefit for you like it does for me because of your screen, you could lock those games to 60fps without problems.

Honestly bro, I think the best thing for you to do is probably:
1 - Invest in good headset.
2 - Possibly get fan near you to drown out laptop fan noise if headset isn't enough, though it should be.
3 - Acquire Jessica Nigri wallpaper. Good to look at.
4 - Use modified vBIOS for clock stability but downclock 880Ms to ~900MHz on core; giving you throttle-free, stable usage.
5 - Attempt at least some cooling mods; even if you buy spare heatsinks. Money isn't a problem for you as you've stated, so some spare parts won't hurt; e-bay might actually work for a couple heatsinks even.
6 - As temps go down with improved cooling mods and such, you could re-adjust your clock speeds to 954 on each card or whatever.
7 - ??????
9 - Profit!
8 - Change Jessica Nigri wallpaper as at this point it's probably been there for a few weeks and you want a different JN wallpaper to look at

 

I have the 120hz screen and it drives me NUTS that I can't get even get a smooth 60FPS on Watch Dogs. At least not with settings that look nice. Its maddening how poor the graphics are in that game in comparison to its GPU demand...

1) I will think about it. The problem is that I'm a live in aide to my roommate and I can't very well hear him when I have a headset on, he and I have been over that before.
2) The A/C is louder than a fan, its just the whine from these things is overpowering. I'm much more sensitive to those things than most people because I suffer from an over-excited system. Glutamate toxicity thanks to my doctor. Long story short, I'm lucky I don't have seizures but I have all the other NMDA overload problems including panic attacks and agoraphobia so any repetitive sound can get in my head easily. This is nothing new, I couldn't use the M17x anymore which was sufficient for my gaming needs at the time (emulators) because one of my GPU fans was off-balance and squealed. Its a non-stop balancing act living my life.
3) LOL!
4) I actually am thinking about doing that. I'm also curious what would happen setting the temperature target to 87C - the stock vbios seems to have an 81C or so target and that usually yields 954MHz. I need to grab Precision X and play with it until I find a setting that works for most things that I can set on boot and then have alternate presets saved for games that need different settings.
5) Already working on it. Money *IS* an object, it just happens that I have money I have to spend in the near future. Once that's gone, its 1300 a month to live on for the foreseeable future so I'd rather not blow all of it for no benefit. CLU will be here next week, I got a tube of GC Extreme because I have heard its better than IC Diamond and its much cheaper, and the converter box to use two power supplies will be here tomorrow.

 

Try and ask Xotic very nicely if they could you send a slave heatsink somehow, otherwise you may just have to pay out of your pocket for one.

I'll report back on the Shin Etsu once I receive it (it's being shipped from HK so it's gonna take a while to get here).

 

Honestly, if your roommate needed you THAT much, he could just yell. With the two headsets I mentioned you DO hear people if they try to talk to you (though I do have really good senses, so I might be an anomaly). Honestly, if it comes down to your equipment or your roommate telling you something (probably) useless, I would pick your equipment. But yeah, if you have enough for a barebones AW18 to throw your parts in, seriously, I think the backplate and heatsinks would be far far cheaper. I'm going to look up GC Extreme as well; we'll see how that goes.

 

The 570wm does kick into full if required.

 

I live with a roommate as well but for the most part we just each do our own thing so I can't complain.

Any experience with Shin Etsu pastes Meaker?

 

No just mx-4 and ic diamond for me.