New 980M right performance?

Yo everyone. I recently popped a 980M in my M15x.
Is this firestrike score about right for 980M at stock with 920xm?
980M at 1206mhz/1250mhz svl7 vbios,
920xm at 3.5ghz (underclocked for now)



The 980M is one hot card! FAR hotter than the 680M that preceded it! Even with liquid ultra on it the temps are barely under control in this old laptop with only 2 heatpipes!

What are safe max temps for this card? 85C is ok for gaming?

 

mid 70s is better.

 

I undervolt my GPUs (and CPUs) so I don't like seeing anything over 70c, all of my GPUs (and CPUs) are generally under 60c when gaming. Generally, you want to stay under 80c, preferably under 70c.

 

The 980M should be a cooler card than a 680M as far as I know. Maybe your heatsink just isn't too amazing with it? I don't know.

 

I have similar results with stock 980M paired with stock i7-3740QM, so I guess it's all good.

Isn't 980M a 125W card, compared to 680M 100W?

 

Don´t know about the CPU but the Graphics score is good. I get everything between 9500-10100 depending on drivers stock.

But as the others said something´s off with the temps, 980M is a cool card. I don´t pass 80C and am mostly sub 70C (stock speed and voltage, just repaste). Presuming a proper pastejob I´m with D2, sounds like a heatsink issue in some way. Is it the same you had before so you know it´s not warped or faulty in any way, or is it a new one that came with the card?

I´m gonna stick out my neck and say 85C is "OK" for gaming. It might degrade a little faster but I wouldn´t worry unless it was at 90C+ for longer periods.
That said 980M is capable of lower temps and sub 80C is preferable.

 

The heatsink is fine. The problem is that it was originally designed for 75W cards aka 5850M and 260M (yes it is that ancient) the twin heatpipes lead to a fairly small radiator.

Modern laptops almost always have minimum 3 heatpipes and a larger radiator.

The cooling was able to handle the 680M albeit with load temps in the mind 80sC with an OC.

The 980M is MUCH hotter as the card hit 75C before I even got in the game and then within 2 minutes shot up to nearly 90C under normal paste! If you compare the two cards the core on the 980M is quite a lot larger.

I now have it under Coolaboratory liquid ultra and the temps took about 15 minutes to reach 85C and more or less stabilised at that temp. Might eventually hit 90C I guess during prolonged gaming.

The 980M is running at 1V 1206/1250mhz with nearly 80% asic value. Really hot. Unfortunately I can't see a way to undervolt Maxwell myself!

What I might do is lock it to the secondary 3D clock value if possible which runs at a lower voltage? Why can't I lower the voltage?

 

You need a modded vbios to undervolt usually.

Sent from my VS980 4G using Tapatalk

 

I guess with maxwell there's no undervolting possible with user, the vBIOS should be coded like that and flashed.

 

i overvolt and overclock and most games don't go into the 80s. has to be a cooling issue.

 

Like I said the M15x heatsink has a small radiator connected by only two fairly long heatpipes. Your 17 R1 features 3 heatpipes to cool the 980M and a larger radiator. So yes it is a cooling issue.

 

Maybe there is a chance to get a better heatsink?

 

@Prema...

 

Yep! Put up the Bat Prema signal!

 

PM Prema for a custom undervolted vBIOS, he does a great job with them. Don't forget to donate if he does help you out.

 

Ah thanks for the PM link. I have done so! I will happily donate before he has made the vbios.

Surprised he tells of people that don't even pay him for his work!

 

Did I seriously just see a card with an 80% ASIC? nV must be unloading Maxwell cores... because 80 is about 20-25% higher than what would be optimal for temps in a laptop... an 80% ASIC means high leakage at low voltage so your stock VID is actually probably close to 1V... I can't recommend using that vbios and would instead recommend using the Alienware vbios from Prema. Bottom line is that svl7 disappeared after release... while Prema has made multiple optimizations.

 

I am rather curious about this because I'm not all that familiar with the average ASIC quality among laptop cards. What's the ASIC quality for your 980Ms at at the moment, @Ethrem?

 

72.7 and 74.1 - and yes my cards overheat even after Prema knocked them down to 975mv.

Well max fans get me down to 82C........

 

Thanks! I didn't know any of that so this is new knowledge for me. Hopefully Prema will come back to me with the same recommendations and extra advice

My asic is 78.4% I believe. Not quite 80% but close enough. I will check tonight when I return home. Even with CLU the temps slowly but steadily rise with fan on max to 85C and more under full load at full clocks. The stock heatsink just cannot dissipate the heat as there is too much..

I will be running a 200mm fan shortly under the machine without the access panel on which will control temps fine but still. The 980M in its current state is incredibly hot and comparable to the 880M that I never experienced and avoided at all costs....I had to OV the 680M to 1.1V and to nearly 1100mhz to get similar thermals!

 

Score seems fine, here is a benchmark from my old ASUS G751JY laptop using a single 980M with driver 347.25:

 

LOL, my cards are 79.3% and 81.3%. I like high ASICs maxwell over low ASIC though. I can cool them fine.

 

My GeForce 980 GTX is only 70.1

 

For where you live(I think dubai is hot??), that might be a good thing. Higher ASIC cards OC better but they have more leakage and runs hotter.

 

explain this ASIC thing to me bro. I thought that the higher the ASIC the better the quality of the card and thus, the better the overclocking

 

Ok, so this is how it works. Higher ASIC means higher leakage and lower voltage needs to maintain clock, so higher temperature at the same voltage compared with a lower asic card but it needs less voltage to maintain a certain clock. However, higher ASIC cards dont "scale" well with voltage. As in, higher asic cards hits a voltage wall due to temp/etc way earlier. Before maxwell, higher asic cards were great for OC on air but since they didnt scale well on voltage, the super higher end cards(lightning, classy) was binned for lower ASIC cards for LN2/water. However, with maxwell, we have physical voltage scaling limits,as in after a certain voltage, the cards refuse to scale at all. That means since at the same given voltage, the higher ASIC card is likely to clock higher, higher ASIC cards are preferred this generation.

Personal experience, I had a 7970 and 780 ti with over 80% ASIC. My 7970 had a 1.05v for core with most others people's on 1.175v. It overclocked great on stock volts but it simply refused to scale at higher voltages. Same thing on my 780 ti, it was a great card on stock volts but refuses to scale with voltage. Generally, now, you want higher ASIC cards period.


Edit:

here is a paragraph from KINGPIN himself describes this issue:
http://forums.evga.com/STATEMENT-BY...LE-WITH-VOLTAGE-NO-MATTER-BRAND-m2362059.aspx

 

This video made me understand it really well:

 

This. FInally someone gets it. Any chip over 70% should not be used in a laptop that doesn't have the cooling available...

I haven't looked at the voltage tables but I can tell you from my experience with the 880M.

1v (ALL 880Ms were high ASIC chips because nVidia was trying to squeeze performance over 780M out of what would be considered lemons...) could run any 880M on the market... But the consequence was seeing 90+C on a regular basis.

Clevo changed the sBIOS when the 980M came out and refreshed their entire line (sans-the P570WM because the demand wasn't there and cooling Haswell-E when they already couldn't get a 4.3GHz 4940MX while Dell did their last hurrah and 4.3GHz was nothing for an Alienware... except the InsydeH2o sBIOS they use(d) made a huge mess all the way around...

ANYWAY... All 880Ms were chips destined for desktop 680s that nVidia was scrapping... on top of that, the 880M with its dynamic voltage was their test for Maxwell because all Maxwell is ------- Maxwell is Kepler wrapped up in a new package that has not only had compute capability taken away but on top of that, has rapid dynamic voltage scaling (which is why you don't get a vBIOS that can undervolt even on the desktop - it will crash unless you have someone like Prema to fix it for you - and please everyone, stop bothering him - if you really need his services, you can buy a machine from one of his partners. There's one partner that wants a 6 month exclusive in order to be a Prema partner and I won't say who because that would throw my impartiality into question but I feel I need to say this - Prema has partner shops and they get everything first. Prema signed a contract as far as I know (and I may be wrong so @Prema) can you please chime in...

As far as the high ASIC chips... Let me explain it this way. You are you and you have a sister. You were both born on the same day but she just always seems faster and smarter and you don't know why. Well in the silicon world, there's a quality number affixed to every chip - its based not just on internal testing (which neither nVidia or AMD do once they have the baseline number) - but its based on where its located on the silicon wafer using a complex algorithm that is proprietary info.

High ASIC chips are only ever put into desktops unless someone either has excess to get rid of or they're desperate to keep up with demand. The latter is the case with the 980M because the 980M is made from desktop 970s that failed QA... So nVidia (or AMD) has no use but either the trash can or laptops for the chips.. This is a good sign for everyone except the one that gets a high ASIC chip... because it means nVidia has a mobile surplus which means Pascal is coming. Based on what I know now... 3Q/4Q especially with the FinFet problems.

Anyway, ASIC is a measure of the quality of the silicon wafer. You ask yourself "why is high a bad thing," High ASIC chips have a severe issue with electron migration which increases heat....... Basically your 80% chip probably has a voltage close to 1v while a 65% chip uses 1.062v (which is the average Prema makes his bases from)... The issue is that while that 1 volt sounds great, you now have a chip that has electrons crossing transistors and in turn, heating the chip up. Average silicon quality won't be able to hit those crazy high frequencies that you hear about... and they take more voltage... but they run cooler. Electrons like cheap silicon. A chip close to 80% is a chip close to the center of the wafer and would make an amazing desktop chip (you could probably do 1450MHz ---- after you fixed nVidia's power phase changes...

Why do you think the 980 Notebook Edition is posting scores the desktop can't touch? They're binned for performance. That lovely feature all of us hate that nVidia calls "boost?" Well these chips might as well not have a boost because at stock, they're already boosting to the max.

I probably got some details wrong. Maybe @Meaker@Sager or @Prema have more to say but the basic gist is that you can probably pop that 980M frequency incredibly high... Way past the GTX 970 its based on, especially with the memory advantage. BUT... the trade off is higher temps. On top of that, the higher the ASIC, the lower the default voltage, the less likely undervolting would actually work because these chips already consume less power at boost and without boost out of the box.

 

@Ethrem i think pretty much all your points on right on except for 980m die binning.

I for one, dont actually believe 980m are binned from failed 970s. A 970 have around the same working portions of the die and really, 28nm yield are really good at this point. If anything, 980m are likely binned for lower power consumptions. Maxwell's power trick can only help so much when you need more or less a 970 in an 100w tdp thats scalable to a variety of cooling solutions. You need to reduce power consumption by at least 45%. How you do it? Lowering the memory speed but using a bigger bus is a move to reduce power. Just from looking at numbers, the extra disabled 128 cores barely affected performances. Nvidia probably did testings and find disabling the CUDA cores improved power consumption and didnt significantly affect performance.

The truly messed up GM204s goes to the 970m and 965m. Thoses are the cards I wont touch with a 6 foot pole.

The GXX are probably binned seperately. Its would be some of the absolute best GM204 chips. Also remember the cards thats are on the P870DM-G have a better VRM setup and higher power limit than actual desktop cards!!. That card is a complete overkill. You could slap the same design on a titan x and probably be fine on stock. I wonder if different 980 gets different binning as well.... that would be interesting.

I feel people are overtly concerned with temps and etc. As long as my cards aint throttling due to temperature, is fine. GPUs runs hot, deal with it....

I dont undervolt. I see no need to, as I said, if the cards aint throttling, its cool enough.

 

I believe people are worried about GPU temperatures for good reasons. Look how many dead 7970m posts there were in 2015, there was literally like 3-5 new threads a week about it. Cards that run hotter will die sooner. I try to undervolt every device I have so I can keep it running for as long as possible so I don't have to worry about replacing any parts.

 

Temperature is an issue to a point. If it's running in the high 70's, low 80's dropping a few C more really won't help matters much. It's when it's near it's max temp limit that it becomes a problem. And underclocking isn't as good a solution as is undervolting. I frequently undervolt components when I can because higher voltage is even a bigger killer than heat, but undervolting gives you the best of both worlds by dropping temperatures as well. If you underclock and maintain same voltage you're not helping matters a whole lot. You'd probably be better off undervolting at stock clock speeds than underclocking at stock voltage.

 

It depends on how much lower temperature we're talking about here, though. If you're running a GPU at 80-90c or more for hours every single day for years, it could certainly reduce the life of a hard drive. You're talking about max temp limit being the only thing that matters for temperature, but I disagree entirely. Any and all heat reduces the life span of any electronic devices (yes, voltage does too). I would hate seeing one of my GPUs running at over 80c, and I personally have seen it, but with reseating my heatsink, using new thermal paste, and undervolting a ton I dropped the temps to literally like 60c or lower which makes me a lot happier. I would find it really hard to believe that temperature had little to no effect in all the dead 7970ms that kept popping up here.

I love undervolting, and I meant undervolting, not underclocking. I did leave that comment at 4AM so that may be why I said underclocking instead of undervolting. Sorry for any confusion, lol!

Although, underclocking helps too for adaptive voltages, and most devices these days use adaptive voltages by default, you would have to manually specify a static voltage for clockspeed to not determine voltage or matter for voltages.

 

Its probably more like, extreme temperature drops frequently applied over a long period of time will kill GPUs. It lead to solders cracking...

I find hard to believe 80ishC is bad for the GPU as many desktop cards run at that temperature.

 

Of course 80c or higher is bad for a GPU, about 60c is the point where an object is too hot to hold onto, and that's a temperature that is considered GOOD for a GPU or CPU. Remember, whenever a component is on, and voltage is travelling into/through it, and it is producing heat, it is literally reducing its life span. It's definitely not too far fetched to believe that temps 80c or higher played a part in the untimely death of so many 7970ms on this forum. From what I have heard, those bad boys ran HOT.

 

7970M doesnt run that hot.

Actually the 980M runs MUCH hotter.

Problem with 7970M was that at stock voltage they didnt OC well. Many people therefore OVed to over the stock 1.05V in order to run 1ghz+.

This is what killed the cards. I ran my 7970M mostly undervolted with a very small OC on core and vram clocked high. It didnt die and I sold it 2 years later.

At least on that card I had control. This crappy 980M cant be undervolted. A well managed 680M easily beats it in my laptop at present as there was no thermal runaway with that card even when OCed!

Not impressed considering how much better this card is supposed to be. All smoke and mirrors just like its creator... heat to power ratio has not improved whatsoever IMHO since 680M.

Before you say I am wrong compare the gpu cooling solutions used for modern 980M mxm laptops compared to 7970M/680M generation laptops.

Sent from my SM-A500FU using Tapatalk

 

Wow, i am amazed. When did 80-90 become too hot?
Is this for new gen cards? As far as i remember, 6-7 years back, the cards could easily run in 90s without any issues.

 

80-90 is hot but not too hot. The thermal protection should kick in anyways. For maxwell I believe it was set to 85 or so, but since i am running Prema's, it's 92 for me. But it's rare for a game to push beyond 70s for me.

My GT60's 780m ran between 80 to 90 though, and is still alive and well. My CPU worries me, because on a hot day it does game on the low 90s, while on winter it's on the 80s. I need to repaste and mod the CPU area.

Oh yeah, my older 920xm ran in the 90s when heavily overclocked and it's still alive today. Older GPUs like GTX260 for me ran always in the 90s hahaha, and they did last, but those were very worrysome temps and times!

 

Doing anything to its reducing lifespan.....

The fact is, you dont need a 10 year lifespan on most components.

 

I lose my temper when I see cards amd CPUs running 80s... I'm sorry but I've had cards fail and I have a cell phone that has a dead Exynos core (3rd core on the A57 cluster)... Note that my phone hits 85-90C in a matter of seconds (and they said the Snapdragon 810 ran hot...) and this is Samsungs major flagship, the Note 5.

High temperatures degrade transistors and the smaller transistors get, the lower the safe temperature goes.

Remember... Ivy Bridge was happy to go to 105 and never failed or throttled. Try that with Haswell.

There's no excuse for a 5k laptop running 80+C on the CPU, sorry.

 

IMO the 680M is the best mobile GPU up to date, however, the 980M is also very good it runs super cool @1.000v (1202/2910) while using the same cooling solution that came with the stock 680M (3-pipe heatsink + stock fan). Never exceeds 73c under the toughest circumstances.

 

Now I like you more than I already did.

I disagree entirely. 90c is DEFINITELY too hot, even WAY too hot, depending on how much you use it, etc.

 

80 and under for prolonged use has always been where my comfort levels are at.
I agree, 90 degree's is on the verge of a shutdown too. It's too hot for prolonged use. I couldn't game away carefree knowing I'm on the raggedy edge.

 

You're gonna like me even more... I get furious whenever anyone tells me "oh you can push it, your chip will throttle when it hits 95C to protect itself - you don't have to worry, Intel has your back."

Can someone tell me when in history that Intel has anyone's back except their own? Dude they made Haswell's throttle a /marketing feature!/

I'm sorry but if you don't get what's wrong with that statement... Even after being told that the temp we see is the core......... You need to do a lot more research.

 

One last thing... Go look up GK104s that BSOD'd at 80C and tell me high temps are normal...

I'm going to bed now. But seriously... GK104s, scalding hot chips, went unstable at around 80C so the high temps being normal goes out the window.

 

Are the modded undervolted Maxwell vBIOS files available for download somewhere? XCOM 2 brings my stock 980M into mid 80's and if I remember correctly with the Prema modded vBIOS you can increase the voltage, but not decrease it?

 

You need Prema to make you a custom modded vBIOS... Otherwise you can only increase voltage and not decrease it with Maxwell..

 

It's also a lot of work because of Maxwell's voltage handling... I thought it was nVidia drivers causing my undervolted mod to crash but it was because Maxwell was throwing 2D voltage way down the tube... Maxwell is a mess with voltage adjustment. I didn't realize that the vbios was dropping to almost 800mv between power states, Prema had to make me a fix when I was testing v2 because 3DMark was throwing a TDR every time the CPU physics test loaded. No matter the driver. For some reason windows 10 was running the voltage lower than it should have.