Pascal: What do we know? Discussion, Latest News & Updates: 1000M Series GPU's

50% over the 980M is 20% over the desktop 980. That's nothing. You're crazy if you think that's all Pascal will deliver. Have some faith.

The 1080M should (at the very least) be equal to or better than the Titan X or 980Ti. 980M SLI already performs better... So, it's definitely feasible.

 

It's gotta be at least as fast as the 980Ti. The desktop 980 mobile version is already 35-40% faster than 980m.

 

Well, since a guy stepped on my ethernet cord and busted up my Sager when it hit the concrete, I can see getting a new laptop when Pascal drops. The Sager still works, BTW.

Sent from my overpriced Galaxy S6 Edge +

 

4k gaming at 60 FPS should be possible with Pascal mobile flagship GPU(s). The 980Ti sees about 35-40 FPS on The Witcher 3 with hairworks off. This means the 1080M should be able to outperform the 980Ti by at least 30%. Of course, this is assuming it has HBM2 and is everything NVIDIA claims it to be. If the entire mobile Pascal line gets GDDR5X during 2016, we may not see such a drastic improvement until 2017. Wouldn't surprise me to see them intentionally hold it back.

@deadsmiley: Damn, sorry to hear. But at least you'll be happy with something new and improved.

 

Please show me proof that a mobile GPU will be 30% faster than a 980Ti. 780 Ti destroys a 980M.

 

40% ish above GTX 980 perhaps.
Which puts it 10% above GTX 980Ti.

We must remember that GTX 680M, which had a huge jump over GTX 580M, was "only" roughly the same performance as GTX 580.

There are some rumors from China that GP104 will arrive around April-May next year but I dont know how reliable these rumors are.
5 months will go pretty fast though

 

2x performance/watt should result in 100% more performance.

Sent from my SM-G925F using Tapatalk

 

2x perfomance in FLOPS...
The real truth is in the details Cakefish.

By that Nvidia mean SGEMM performance. Which is single precision.
Fermi is at 2 in Nvidia`s graph. Kepler at 6. GTX 580 (Fermi) score 1000 in SGEMM while GTX Titan (Kepler) score 3200. Which is 3x as much (2 * 3 = 6).

But how did those compare at gaming? 60% faster!
Not 3x as

My point is:
2x performance/watt is not 2x performance/watt in gaming

 

We should be seeing leaked benchmarks of new cards at least 30 days before the actual release date. We already see driver support for them (if I recall correctly). So, we're not far from leaks on Chinese websites and whatnot. We'll know how good Pascal is by then.

The 1080M better equal 980Ti or better.

 

desktop or mobile pascal will launch first?

 

Usually desktop launches first. The 700 series mobile cards launched like a week after desktop. You can look up dates on Google.

 

Yeah though 600-700 series were kepler so they are much easier to launch together. Some 700 and 700m were rebrands too. I do believe Pascal being a new arch will launch on desktop then followed on mobile.

It's bound to be more ground breaking for mobile users if they manage to keep high performance on a controlled tdp

 

ure all assuming that nvidia is gonna play nice and do exactly what we enthusiasts want them to....um, hello, have u met nvidia?!
how about... release of a 980M pascal successor that puts it in between 980M and 980 mobile performance wise in order to have the perfect excuse to release a 980 mobile successor later down the road and milk the laptop users for every cent theyre worth....now THAT sounds familiar with nvidia

Sent from my Nexus 5 using Tapatalk

 

AI isn't roses either. I heard from someone in my massively extended family that not only is AMD planning to refresh for mobile next year but apparently both of them have rejected performance vs watts.


You all realize that Pascal mobile will be the Haswell of...............
.. Well I can't say anything nice but expect 95C under load to actually go somewhere.

 

I'm not assuming they'll play nice. I'm assuming they'll continue their trend. More often than not, every new architecture has lead to massive improvements in performance. Among countless indicators of this, NVIDIA has clearly boasted about Pascal's ability to trump Maxwell. So, it's safe to say it will be much better.

You can't know where you're going unless you know where you've been. I study business analytics. We forecast company behavior in our classes. Closer you are, better the grade. No, it's not a perfect science... But you have a very good estimation as to what may happen. High risk, high reward.

My hope is more than 980M SLI performance in a single mobile GPU (or 30%+ over the 980Ti), but I'm expecting around 980Ti levels.

 

I am sitting here with rather toasty 880M. Usually, I buy new tech when the old tech no longer meets my needs. However, if Pascal is as "just" 50% faster than Maxwell it will be a good enough jump to convince me to buy it.

 

If it's not at least 50% better than the 980M, NVIDIA intentionally held it back. There's no way. Mixed precision, NV-Link, HBM2, etc. all perform more than twice as fast as current GPU technology. And not to mention, Pascal will have complete DX12 support and will be able to fully utilize it, unlike Maxwell.

 

Well AMD releasing mobile cards next year was sort of a given. They have been gone from mobile race for several years now and that was because they didnt have a efficient architecture like Nvidia had.

Care to elaborate on "rejected performance vs watt"?

Also one would think 16nm mobile Pascal would be around 680M/980M heat though. I understand it will be a more dense die, but one would think Nvidia still use 300mm dies with 100W limit?

Not sure why you keep everything secret. Nobody here knows anything where you got the info from and your contacts wouldnt risk anything because nodody knows who they are.

 

HBM2 is not going on first gen mobile. NV-link is for professional market only more or less. As a gamer, why does mixed precision even matter to you?
Pascal is build for pro market, where maxwell is build only for gaming. Hence the lack of FP64 units. It is a given nvidia will hold back on their first gen pascals.

 

2016 is certainly shaping up to be a great year for notebook enthusiasts.

 

AMD is going to continue to refresh their mobile parts on the high end and also continue to put Dell and Apple on the priority list.

Pascal testing isn't looking promising from the heat department. We all knew there would be a point where GPUs would slow down because of heat like CPUs. At the very earliest, there may be a Pascal mobile GPU at the end of 2nd quarter. NVidia doesn't want another disaster mobile flagship and they can still milk out even more power out of Maxwell

As for my being cryptic, I was just drunk. All of this is technically rumor mill stuff but as a mod I can't have my name plastered all over the net as if this is all definitive - I'm sure you know what I mean @Cloudfire

Anyway, both AMD and nVidia are going for the performance crowns, not the efficiency crowns. Makes sense for desktop but could be a disaster for laptops which is why laptops will probably be Maxwell refreshes until 2H or so.

 

They're not going to refresh Maxwell for 2016. If they were going to do that, they would have just released something during 2015.

They haven't released a "top tier" mobile GPU since October of 2014.

 

GTX 980 for notebooks says hi!

But I strongly agree that 2016 (at least the latter half) will be the year of Pascal.

Sent from my SM-G925F using Tapatalk

 

GTX 990 for Notebooks.

 

As long as it meets our expected performance (better or equal to 980m sli), i dont care what it is or what it is called.
It should not be difficult to acheive this. GTX 980 already comes close to 980m sli and the new flagship will have to be better than GTX 980.

I think the new lineup will have some form of pascal. We have been hearing rumors related to pascal consistently (e.g. pascal is taped out, some will be gddrx5 and so on). If the next line up is maxwell refresh, these rumors do no not make sense. It could be a mixmash of pascal and maxwell.

 

Do you really think that a full GM204 core needs 8 power phases?

Also, look at the cooling system the P870DM has... Then think about traditionally what x90 cards are....... Let's put it this way, 8 power phases provide enough power to run two cores.

 

Well.. emm.... I could answer if i knew what that means.

uhmm. Ok,...........i guess.

I deeply apologise for my noobness.
:goes back to lurker mode:

 

ethrem, ure making mouths water here....

 

No, thats more phases then the desktop version.

Yes, you could but it also lead to disasters like GTX 590. When nvidia thought it was a good idea to squeeze 2 power hungry GF110 on a single PCB with a 6+2 phase design. I am not a huge fan of dual cards.

I feel like cramming 2 x GM204 on a small PCB like that is not a good idea. Maxwell have pretty aggressive power management, so it might throttle a lot. And also, you need to fit 16GB of VRAM on that board.

 

Maybe we'll see the full GP100 chip on the P870DM next year, you never know.

 

With Clevo going all bananas with freaking 200W GPUs inside notebooks, anything is really possible now. lol

 

Hope we are surprised. It's clear that these companies have been holding out on us.

Less than five years ago, if you spoke of a machine like the P870DM, people would have said, "You're crazy. Keep dreaming."

 

Less than five years ago we had the Clevo X7200 (predecessor of the P570WM) with a 6-core desktop 980X CPU, triple channel ram, and dual 480M GPUs. (Those 480Ms had the GF100 chip, for comparison, today's big chip is the GM200, granted only 352 out of 480 shaders were unlocked).

It's all in the marketing Although I'm glad Clevo has been sticking with desktop CPUs even if they had to go to the mainstream socket instead of HEDT.

If we get a full GP100 based GPU in the P870DM then I'll be surprised.

 

AMD has already shown HBM2 doesn't increase performance 50%, not even close. And NV-Link is not for consumer, that's server purpose. Unless you can show me a link that next generation laptops will have NVLink instead of PCIe. Also show me link that Kepler will revert Maxwell's abandonment of DP and FP64 for consumer gaming.

Until there is proof that Pascal mobile can play Witcher 3 on Ultra at 100 FPS, all your speculation is premature. Please show me this... that will be the proof that Pascal will bring more than 50% increase performance. I'm still expecting 15-20% or about on par with GTX 980 @ 100W.

 

Ha! We wont even see full GP100 on desktop within the year, and maybe late next year if nvidia is generous.

 

You mean Pascal right? GP100 is most certainly a GPGPU card foremost. So I doubt it wont have FP64. GM200 is a cheap toss out to consumers on 28nm.

 

ROFL. You are not getting GP100 on mobile gaming, ever. Dream on. That will be Tesla or maybe Quadro. More likely Tesla with NVLink.

So please show me link that NVidia is bringing back double precision/FP64 for consumer gaming rather than just spewing speculation and hype.

 

I am not disagreeing with you on any of those two points. I was merely pointing out that you probably meant it as pascal and that GP100 will have those capabilities, but whether nvidia enables them on prosumer chips is another question. FP64 might be back on the titan class chips. NVlink probably not. And yes, GP100 not even close going on mobile.

 

Proof doesn't exist. But I'll let you know when it does. We always speculate on this forum and we're almost always on target. You are a very negative person. I can't show you physical evidence we went to the moon, either.

Just to clarify: I never said HBM2 or any of those other things will increase performance more than 50%. I was implying that because those things exist, the performance output of a Pascal core must require them, otherwise a bottleneck must appear. And if that's the case, performance is far greater than 50%.

 

Of GP100, which have no relevance with the mobile gpu. And with the bottleneck problem, every system/cores/whatever have a bottleneck. You cant have a system that have no bottleneck.

 

course u can! just make sure that ALL components are 100% balanced out and maxxed out in a specific software environment, then your WHOLE SYSTEM is the bottleneck, but no single component

 

The thing that makes me mad about Pascal is that nVidia is ready to hit production with cards that may as well be called Thermi.....

 

ure such a tease ethrem... pretending to be "suffering" from that inofficial info doesnt really help

let us suffer together! itll ease your pain so cmon, spill the beans already

 

Intel didn't mean for Haswell to run as hot as it did... The die shrink had consequences... This one happens to have them too. Volta will probably be the last chip that makes any more than a 5-10% gain until everyone gets graphene in place... Silicon has reached its limit in terms of heat dissipation. Top tier Pascal cards will need top tier cooling.

 

As long as Watts per mm squared stays the same then cooling shouldn't be an issue - maybe they'll just have to run more cores at a lower frequency.

 

That's why Pascal won't be out for mobile until 2H or later. I'd expect a March refresh of Maxwell.

 

Will they release a better card than 980 (notebook)? I was so sure about waiting for Pascal. But you made me think twice

 

There is no doubt that they will release something better than the current 980 but Pascal for mobile, I'll be amazed if they release one with how hot it's running right now. We're talking a temp target that's higher than the current Maxwell throttle point. This is the company that put big die Fermi in a laptop though... At this point I'll be amazed if their desktop cards they've promised can deliver the performance they've promised without them going the AMD "burn em to a crisp" route.

 

Do we expect these GTX 980 machines to be upgradeable to standard mobile Pascal cards?

 

sure that sounds crazy, but dont forget that the currently set "official" throttling point for maxwell mobile is rather low... only with prema vbios its set to a more reasonable 92C! so if, say, the temp target on pascal mobile tends to be in that range, totally fine by me doesnt mean i wont tweak it to stay below 90C for prolonged periods of time