Upgrade MXM or wait for new 1000 series cards?

Hi all,

got myself a laptop with a GTX970m and wondering whether it would be better to spend the money on a GTX980m or wait and see what they reveal for 1070/80 laptop cards and just get a new laptop?

I've got a Clevo p750zm with the aforementioned 970m I'm guessing even though it's MXM the 1070/80 new releases probably won't fit this chassis

 

A 980M will give you a noticeable jump in performance (as far as 3D applications go) over the 970M and will cost less than a brand new system. It's really up to any budget constraints you might have and your need for performance.

 

Yeah I thought there would be a big increase between the two cards however sellers are quoting 600gbp which seems ridiculous however it is the only part in the laptop I'm interested in upgrading at the moment as I'm happy with my SSD's 2x 512 Evo 950's and 32gb ram as well as an i7 4790k. If it was going to cost 600 without the extra powersupply which will be 40-50gbp I think that could be a sizable chunk from buying a new system,

 

You should also consider waiting. A 1080m should sell for around 725 bucks if its like the 980m at launch, and could be far far superior.

 

do you think they'll fit? because if so that's the ideal solution for me I'm super happy with this laptop I just want a bit more oomph in the gpu

 

Well, MXM is standardized, so it'll fit. What you should be worried about is whether you can find a heatsink assembly that matches the new card since memory and power delivery may be moved around.

Also, Pascals chips should be far more power efficient, that's the real kicker for a mobile application there.

 

That's good to know. I thought the new chips were much better with power efficiency? wouldn't that mean less power need? I'm not clued up to how they'll launch the mobile versions and the mxm of the pascal series.

 

We haven't seen any real specs for the new graphics cards as of yet, so it is still hard to compare. Still no official dates as of yet. But if you were to compare a pascal graphics card and maxwell with the same wattage, then you would see more power from the Pascal card (comparing desktop cards basically).

 

In terms of buying a card now it's not worth it IMO unless you get a good price for your current one.

 

seen a lot of people speculating they'll be full versions of the desktop cards only ramped back to avoid high temperatures but I don't see it being that good

 

didn't even think of selling this one? figured it wouldn't be worth anything without the other parts plus I've not changed one in these laptops before. I was hoping to just order the 980m kit and do it keep the 970 as a backup. But if there's a chance they release the mxm version and it'll work in this system I'll definitely wait

 

Just wait for the 1080M.. It will fit hopefully and will be much better then either.. Just wait a bit.. 2-3 months tops..

 

yeah I hope so. I'm just hoping not to get gouged badly buying an upgrade kit. Rumours are saying the 1080 laptop version will be on a par with the dt1070? if thats the case I'm happy. a lot of the reason I got this laptop was for mxm and I was worried they would stop doing it for these cards as people had said. Just in time for xmas hopefully

 

even if they are the same chips but scaled back (power or cooling related), I'm guessing a scaled back 1080 is still going to be better than a 980 by a pretty decent amount.

 

I'm waiting for a Pascal (or Polaris for that matter) card as well. I only opted for the 970M knowing full well how "late" in the Maxwell architecture I bought in (Nov 2015). The Pascal architecture is significantly more efficient, which means the mobile cards should be similarly quicker than the Maxwell Gen. I'm also pretty surprised at just how well the 970M overclocks. It's a monster. You can easily pull another 15-20% out of it without much work.

One other side-effect is battery life. At idle, the GP104 parts appear to use anywhere from 5-10W LESS power at idle. Even if that only translates to 3-6W difference in a mobile platform, that's actually quite significant. For the P750DM (using NotebookCheck's idle power draw figures) that represents at worst a 10-20% increase in battery life and at best 15-30% increase. That could be as much as an extra hour of battery life in low usage scenarios which is definitely a good thing.

Whatever part they put out at the 100W TDP skew (980M successor on standard MXM or Integrated) should hopefully perform around the range of the full fat GTX 980 which is nothing to sneeze at. I imagine they'll use it as a huge selling point for VR laptops as a 75W TDP skew should perform about the same as a GTX970 (ie VR spec).

 

I would advise against assuming that all new cards will be compatible or operational in the current chassis. This isn't a statement of fact, just advice - I can't say any more at the moment.

 

EDIT - double post, sorry

 

What he says!

 

Older cards do tend to hold their value for MXM boards. However unless you are into modding firmware and are a very advanced user I would wait and see how it pans out.

 

Anything we can do to swing odds in our favor? I'm assuming Nvidia want to do away with MXM because they're greedy.

 

Nope, nothing you can do to swing anything - I wasn't implying anything at all really, just advising against assumption ;-) For example, there are photos from Computex (and posted here/techinferno) which proove that next gen GPUs will be available in the Clevo models which will be superseeding the current discrete GPU range. But this in itself doesn't guarantee that current gen and next gen GPUs will be hotswappable in current gen chassis.

All I can do atm is keep my lips sealed regarding specific details...

 

Even if it's not "officially" supported, there's always amazing people out there like Prema.

 

Correct, but even Prema liked my original post where I said that no one should assume that next gen cards will slot right into current gen chassis.....there's "officially supported" and then "physically compatible", not everything is always possible.

 

Yeah.. I'm worried about this. The Alienware m18x was a fine machine but Maxwell didn't really play ball with it. Here's to hoping we don't get screwed into another artificial EOL with a machine that deserves a long life.

I just picked up a Clevo P870DM-G so I've got alot riding on it. Worst case scenario I'll very reluctantly get a 980 or just give up (again) on the enthusiast mobile scene all together. Its a real shame. This scene could be bustling if there wasn't so many parties all trying to pull screw jobs on their customers.

 

my left-click finger is itching to press the buy button on a P870DM-G.

really want to put it through in july as a self-b'day gift.

 

Haha it is tempting I know, even with my current setup.

 

Videocardz has an article with what is apparently a 1080M pictured.
http://videocardz.com/61256/nvidia-geforce-gtx-1080m-to-feature-2048-cuda-cores

Things you can safely derive from those picture:

• The pics on that site are woefully low quality. The actual source on Chiphell does have much more readable shots clearly showing GP104 in HWInfo and
  
• The hardware pic appears to be a MSI GT72 Dominator
  
• It's using the cooling from the full fat 980 system (3 heatpipes to the GPU block, whereas the 970M/980M variants only get 2 heatpipes there).
  
• The aluminium heatsink for the memory and heatsinks isn't anodized/painted yet so it's probably a pre-production test.
• It appears to be a regular sized MXM card (good news for P750/P770DM users) and no sign of the secondary power connector. The MSI expanded board for the 980 looks like this link HERE. Note the bit jutting out on the bottom left.
• On the Chiphell source, the shader/rops/tmu counts are all the same as the GM204 GTX 980.

 

The heatsink design always looked like it should have been that way from the start IMO. However as a pre-production unit you can never be sure.

 

I think it may have been built into the original GT72 design where the 880M actually broke MXM spec and ran at 130W TDP. But with how Maxwell chips ran much cooler they could save some cash by dropping it to 2 heatpipes.

What's more impressive from the Firestrike image is the fact that the card is pulling a 16893 GPU score with what appears to be regular MXM. This is <10% away from a reference 1070. Entirely plausible if they intend to use that 130W TDP again though. Better yet, it's around 980Ti territory which is exceptional for a laptop.

 

If we get 980ti levels of performance in an MXM I'll be very happy. Gosh.. the thought of jamming 2 in this machine is getting me giddy. Shame SLI is always taking a back seat in game development.
What an awesome machine this thing could be if things turn out like they should and we can upgrade.

 

Those photos are actually from NBR here not Chiphell

(hope I don't get anyone in trouble with this comment)

 

That would be a pretty nice jump in performance in laptops. Considering the 700 to 800 series, and 800 to 900 series of mobile wasn't anything quite that big of a jump, so it would be a welcome increase.

 

Maxwell touted a nice little hop in performance compared to previous iterations, but it really seemed to shine in heat (or lack of it) by comparison.

 

The jump to the 980M was huge while sorting out the thermal issues too IMO. Certainly considering it was on the same manufacturing node. The 970M made a lot of 870M based machines far more practical too.

The 7 to 8 series was a re-badge on the high end (860M and below got a big boost as they were maxwell) so there was never going to be much of a change there.

 

Take a closer look at the picture that Chiphell hijacked...

 

I know I am in minority, but i am not so impressed by the Pascal architecture. I mean the wonderful performance increase is due to Finfet. Clock for clock it is not much of an improvement over maxwell. In fact if Nvidia had gone ahead with their original plan of having maxwell on 16nm, it would have had the same performance as Pascal.
And 1080m roumored to be slower than 1070, adds to the disappointment. I hope either 1080m has variable TDP (like GTX 980 for laptop) or they release GTX 1080 for laptop.
Keeping the fingers crossed

 

Watt-for-watt, Pascal is about 60-70% faster which is nothing to scoff at. That translation pretty much directly goes into mobile chips which is pretty fantastic. I don't really see how people find that disappointing.

I still get the feeling that the "1080M" we keep seeing is a regular sized MXM card so we're looking at 100W (or maybe 130W if they intend to bend MXM spec again like the 880M) and thus simply a 980M replacement. Not a 980 replacement. There may yet be a "full" GP104 1080 card using the expanded MXM designs that the current 980 uses in the works. AFAIK Nvidia is the controlling factor in the MXM spec so who knows if the current mobile 980 is simply an aberration to get on the VR bandwagon or if they actually intend on updating the spec itself.

As long as desktop cards target much higher TDPs, there will never be parity. As long as mobile chips have a lower TDP than desktop chips, prepare to be permanently disappointed. Any parity we get in this generation is purely because Nvidia haven't released a 250W variant yet and we're in a strange situation where we have the option of up to 200W with these expanded MXM 980 cards.

I fully expect there to be a full GP102 (rumoured to be basically a GP100 with FP64 units cut to 1:32) desktop card down the line which will simply reset everything again.

 

I said i was in minority

My point is that Pascal an architecture has failed to impress me. Now if Nadia had kept Pascal to its originally planned specs, things could have been different. The Pascal we have today is mostly the originally planed Maxwell with some improvements.

Yes this is likely to happen, but the question is that will it happen now (at the same time as 1080m) or later.
And to be honest this is expected, they are creating a tier above top tier in mobile market. Similar to what they did in desktop. I am jutting waiting for the time when they would start calling these cards “1080mTitan” or “1080mTi”.

AFAIK Nvidia has done controlling the MXM and will leave it to the discretion of ODM to implement it. This would mean a bigger mess of MXM variants then we had for GTX 980. But I am not sure about this rumor.

I never said I expect parity. All I am saying is that I expected them to do more than that for 1080m.

 

Competition in the segment would be nice I think.

 

Original specs? I think you're getting mixed up here. Here's the NV blog post: https://blogs.nvidia.com/blog/2014/03/25/gpu-roadmap-pascal/

These are all targeted at HPC usage and all of those things are available on the Tesla P100.

I suspect most people got caught on the "Pascal = 10x Maxwell performance" bit at GTC 2014. But as usual, most people didn't read properly, as that was for double precision performance. They absolutely blew that out of the water anyway. The GP100 is 20x faster in double precision than the GM200. Most people also forget that GTC is a GPU compute conference, not a gaming conference.

The "Original" Maxwell was supposed to be a drop to 22nm which never happened. Judging Pascal by a process change that didn't even happen is pretty unreasonable.

Nvidia took their highest standard TDP envelope, built a GP104 chip around it and call it the 1080M (their highest badge). How exactly can they "do more"? You can't just create better performance out of thin air.

 

Its clear that that we have a different point of views on Pascal, and i dont think either of us can convince the other to change it. Nevertheless I will try to clarify my point one last time.

Can you kindly inform where is it mentioned in this article or in any other article of that time.

Well most people I know didnt confuse DP performance with gaming performance.
Also this extraordinary performance gain in DP further strengthens my point. If we have had something even close to this in gaming performance, I would have been impressed.

Again this strengthen my argument regarding not so impressive architecture. The performance increase is not more than what kepler was to fermi or even maxwell to kepler.

Having said all that please note that i dont imply that Pascal is a failure by any means. I am just saying that it is not as special as some may believe it to be and I personally think it could have been much more.

 

Have a look here: http://wccftech.com/nvidia-pascal-gpu-gtc-2015/
The fact that this is a GTC conference about compute should be be evidence enough these are HPC features.

Many people assumed that all of this news from GTC would apply to Pascal GeForce consumer cards, which is not the case. Particularly the assumption the 1080/1070 would get HBM. To be fair, if you're into HPC and looking at GP100 Teslas then the performance gains really are that big. Maxwell was only useful to HPC groups if you used FP32 heavily. Anyone doing FP64 still had to hang onto Kepler gear. Again, GTC is a compute conference, not a gaming conference.

As for the official P100 specs: http://www.nvidia.com/object/tesla-p100.html
Some of the names change, but each of the big 4 features from GTC are indeed there with HBM, NVLink and Page Migration Engine (unified memory).

Most importantly, Nvidia always talked about their GP100 chip at GTC (see the "2016 update" at the bottom) and everybody drew the wrong conclusions that it meant that GeForce cards would also get said features. Almost everywhere people were getting giddy about HBM2 on their GeForce cards...But that was not the case at all, that was for the Tesla card which is completely different beast.

That being said, HBM would be absolutely brilliant for mobile cards. Especially from an efficiency standpoint. I'm guessing it's just too expensive to roll out in large/consumer quantities.

Just google "Pascal Maxwell 10x" and look at how many publications regurgitated the "Pascal is 10x faster than the Titan X" as a headline without further thought. Furthermore, being 10x faster than Maxwell (TitanX, if they just said "Titan" then that 10x number becomes a much less impressive 2.5x) is actually not a big feat at all, primarily because all Maxwell cores were hamstrung in FP64 and had no FP16 hardware at all (which was counted in that "10x" number).

The huge double precision performance gain comes from the GP100 actually having the FP64 modules on the actual chip in the same 1:2 configuration and not 1:32 which is the configuration of both the GM20X series and GP104. This is precisely why the GP100 Tesla has more that double the number of transistors as the GP104 1080/1070!

I similarly don't believe it to be as special as some but I'm not of the opinion that it could've been any more than it is. I think there's also a level of underappreciated for tech, both red and green team. People think "oh it's going from 28NM -> 14NM therefore everything should be twice as fast if not more!". It's just not that simple. When it comes to games which are largely FP32 reliant there's only so many ways you can make that faster and we're fast running out of ways to make that process more efficient. This is why you start seeing some "out-of-the-box" thinking when it comes to optimizations like multi-projection/time-warp/memory compression/etc.

People talked it up and drew all the wrong conclusions from GTC information which was only ever setting them up for failure. Me, I'll take that performance increase as a bloody miracle (and possibly the last similar sized jump for a very long time) because we're basically trying to squeeze blood out of a stone when it comes to performance these days.

 

Pascal offers a huge performance gain, a 1080 > Titan X, enough said.

Now, could Nvidia made Pascal better ? OF COURSE, but if they want customers to buy Volta (which will focus on Vram) they can't put everything in Pascal.

It offers features for Vr, Dx12.1, and again a HUGE performance gain, how can someone think it's a failure ? Commercially it's a big success already and needless to say that when the mobile versions will be revealed (july normally) it will be even better.
You're just seeing the nm reduction from 22 to 14nm...but you seem to forget a whole part about pascal : software features, new algorithms, optimizations, new drivers, new bios, new power management....

Nvidia did a GREAT job period.

 

I think the 1080m will be a nice upgrade to the 980m but i dont see it as a "necessary" upgrade if you have a 1080p screen and are on a 980m. The 980m is pretty capable of vr also.

 

Try the test @ CPU 3.6ghz and GPU @ stock

The 980m is nice, but it's no 1080, and the 980m isn't even a 970.

 

Thats why i said its a nice upgrade. At 1080p reso on games a 980m pretty much chews through every game at it. Maybe with next gen games it will show its age.

 

The 980 mobile desktop is the only GPU certified for VR from the Maxwell generation of laptop GPU's.

The 1080/1070 mobile, "M" or full mobile desktop, should both be certified aok.

Games are taking more GPU power all the time, and many want to graduate from 1080p to 1440p at least, if not 2160p.

A single 1080 can do 1440p at higher FPS, but games that can scale with SLI - 2 x 1080 - show great promise on 2160p - a single 1080 isn't quite up to 4k 60 fps yet.

You might as well upgrade to a 1080 mobile on your 980m laptop while you can.

 

Idk. Last time they said a 970desktop is enough for vr and a 980m is close to a 970desktop. Did the test over at steam and passed the test. Ill get the vr soon and if it works, it works. I believe its just a ploy to get everyone buying the "new stuff" ofc thats how marketing/advertising works . Take more money from people by misleading them. Ill do my due dilligence before i do some serious spending.

 

They set the desktop 970 as the vr floor officially so anything below no matter how little does not get recommended. Tweaking can of course fix that.

 

Multi-Projection on Pascal could alter the "VR requirements" as well. If it's as effective as claimed (and software correctly takes advantage of it) then the requirement could actually drop when a Pascal GPU is in use. ie, something like a 1060M (whatever that may be) while it may be weaker than a 970 in pure performance, it may edge it out by simply not being penalised as hard due to lens warp and how SMP deals with it.