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

It'll cost 50% more than Titan X SLI if Titan Z was any indication. One thing's for certain, it won't be cheaper than two Titan X cards, not with the engineering and binning required. We may even see a new cooler.

 

Sorry to ask probably molesting question. Do you guys think, so at least some Pascal card could be exist in current MXM format? Just may affect on choise 970M/980M for my new laptop purchasing.

 

From what i have been hearing, there should at least be an MXM version of it.

Is anyone wondering how they are gonna do NVlink in laptops and home pc? the last i had checked they where using these big MOLEX connectors or someething. looked like a 8pin connector on steriods.

Sent from my SGH-M919V using Tapatalk

 

You guys do realize now that NVIDIA has put a GTX 980 in MXM-B form factor inside a laptop, they must be extremely confident in Pascal being able to top that performance. It would be beyond reprehensible for NVIDIA to have their new Pascal lineup beat by previous generation hardware... What am I getting at here? It's practically guaranteeing that the 1080M will be more powerful than a GTX 980 at this point.

Let's just hope next year we see a GTX 1080 in a laptop. 4k 120Hz may become a thing by 2017. Oh, and Micro$oft... Fix the damn scaling.

 

Ugh, I just hope they don't design the 1080 with laptops in mind. Desktop chips should be designed for desktops, and I the lat thing I want to see is Nvidia cutting down a flagship desktop GPU just so they can shove it in a laptop.

 

So what you're saying is, you wouldn't blame Nvidia if Pascal is slower?

 

nVidia does not cut down flagship for mobile.

The GTX 980 was not flasgship it was GM204 based which is a smaller chip. GM200 was the flagship and we never have seen that chip in mobile. Titan X is the flagship with all 3072 cuda cores and 980Ti is the runner up with 2816 cuda cores.

We never really see flagships in mobile, we did see GF100 (480M) but look how much slower that was from a desktop 480. Things could change however with Pascal.

If we saw a 980Ti based GPU in laptops well then that would be interesting. Those GPUs normally have a 250w TDP. The new Clevo P870DM has a 200W tdp heatsink now so there is a small possibility if nVidia cuts power consumption down further without hurting the GP100 performance, maybe we can see a 1080Ti in laptops. Not a full GP100 chip like the Titan Y or w/e they want to call the new Pascal based Titan but the next model down like a 1080Ti. 250W TDP to 200W TDP seems doable with 14nm and slight reductions in clock speed. And if HBM2 makes it to MXM next year, I can definitely see it happening.

 

Haha, I think J Dre meant to say "reprehensible" rather than "irreprehensible"!

 

sure, based on the assumption that nvidia always delivers what the fans want, right? seriously, wouldnt even surprise me if their pascal mobile flagship does NOT beat the 980....rather: "hey look, its as fast as the 980, just with HALF the power consumption!"

Sent from my Nexus 5 using Tapatalk

 

Nah, they're not that dumb. Why go around telling everyone it's 10x faster than Maxwell only to have it perform the same? Doesn't make sense.

It's actually impossible for it to not perform better than a GTX 980, given HBM2 and NV-Link. I'd expect it to perform like 980M SLI.

 

I understand that; I never said they were. I'm just worried that they might start doing just that so they can continue putting flagship desktop chips in laptops.

 

Continue? Sounds like you didn't understand ssj92's reply. Flagship desktop chips are the big dies (codenames ending in -0), not the medium dies (codenames ending in -4). Except for the one-off that was GTX 480M (an extremely crippled GF100 mind you), they haven't put flagship desktop chips in laptops since G70, and that was before the big die era.

 

By flagship I meant flagship at launch (like the GM204 980). Sorry about that.

 

Nvidia used to always launch a new architecture with the big die, but starting from Kepler they switched to leading with the medium die. Also when laptop 980 became a thing, GM204 was longer Nvidia's fastest.

 

Allow me to clarify further: I meant I don't want them launching the first Pascal desktop flagship, the 1080 which would be based on the -4 chip (not the 1080 Ti -0 chip which would come later down the line) with one of the main intentions being to shove it in a laptop, and holding out until the 1080 Ti to give us a worthwhile desktop GPU.

 

You're afraid the Pascal medium die will be gimped because Nvidia has to shove it into a laptop? I don't recall that happening with Fermi, Kepler, or Maxwell, and we had the full medium die on mobile in all 3 gens. Bro, efficiency and building GPU architectures from the bottom up is Nvidia's strategy. That's a good thing for mobile users, and it's why Nvidia is so far ahead of AMD here.

 

Was the 680M a completely unlocked GK104? I don't recall that being the case.

 

No, but 680MX, 780M, and 880M were.

 

Right, but the desktop counterpart to the 780M (780) was based on GK110. With Pascal, we're looking at a new architecture.

 

Like I said, except for the one-off that was 480M, that has been the case. 8800 GTX was G80, 8800M GTX was G92. 280 was GT200, 280M was G92. 580 was GF110, 580M was GF114.

Also GK110 was delayed into the 700 Series. It was supposed to be GK100 and 680 was supposed to be based on that instead of GK104.

 

Mobile has always been behind. nVidia tried shoving the flagship (480) into mobile package (480M) but had to reduce clock speeds so much that it performed worse than a 465.

Fermi already ran hot and used plenty of power, that's when nVidia decided to reduce power consumption as much as possible with kepler. Although I gotta say all of my fermi based GPUs were monster overclockers and loved voltage.

They didn't gimp desktop kepler for mobile. Even with kepler, we got the 680m which was just a lowered clocked 670.

nVidia has been saying they're closing the gap between mobile and desktop(and they actually have) but they're been comparing them to midrange desktop chips (GXX04) not flagship (GXX00).

I'm hoping with Pascal that gap will actually be smaller and we get closer to flagship performance in a 200w MXM 3.0c package with HBM2 in late 2016.

 

http://wccftech.com/nvidia-pascal-n...ked-memory-1-tbs-bandwidth-powering-hpc-2016/

NVIDIA will be officially presenting Pascal GPU's at GTC on April 4th, 2016. The launch date will follow soon after.

 

Hi J.Dre, I skipped through the article but couldn't see where it said that it will be officially presented on April 4th 2016 - does it actually say that in the article, or are you getting that info from elsewhere?

 

actually kinda crazy if u think about it: before, it wasnt a biggie putting the full medium die into the laptop, and that at regular MXM TDP values. now its a big PR stunt of doing exactly the same thing, only with DOUBLE the TDP

nvidia PR machine at its best....(or worst, if ure an enthusiast )

Sent from my Nexus 5 using Tapatalk

 

Yep. Basically because they have no way to step up their "A" game until Pascal hits, they figure this will sound brilliant to most. "Wow, desktop 980 in a laptop" even though it will only be in a select few laptops, yet everyone else trudges along with two year old technology for another year.

 

"This year at GTC 2015, NVIDIA shared quite a lot of details about Pascal GPU but we have yet to get a glance of the architectural improvements implemented inside Pascal GPU and all the juicy details would have to wait till next year’s GTC in April 2016 which will be around the time Pascal makes it into the market."

Same article. They have a large booth at Computex 2016, according to the event list. They'll be launching there most likely.

Exciting times ahead. Back to the old (original) launch schedule from the 700 series!

 

personally as long as it works im happy, and i wouldent complain too much if nvidia wants to continue putting desktop parts into laptops.

Though, if someone has the time can they put unified memory into laymans terms for me. from what im understanding is that instead of mirroring the data to each components memory its going to allow them to work off the same memory.

and can i get a realistic expectation of what the impact would be please?

Sent from my SGH-M919V using Tapatalk

 

realistic expectation? sure, lemme grab my pascal gpu real quick and do some benches for ya.... oh wait....

nah, but in all seriousness, way too early to speculate on this. itll depend on the memory dependencies of a specific application and how much that dependency is based on cpu / gpu performance, as well as system ram, vram and cpu L caches...

Sent from my Nexus 5 using Tapatalk

 

Yea, I figured I'd ask. you guys seem to have your fingers on the pulse more than I do.



Sent from my SGH-M919V using Tapatalk

 

Cool, I missed that! It doesn't sound set in stone from that quote though...but would be good to hear sooner rather than later!

 

I'm sure more details will come soon... We always know before hand. They haven't released any new GPU's for 2015 to replace the 970 or 980, so that means they'll plan to launch a bit earlier in 2016. Gotta keep the shareholders happy!

 

exciting times ahead

 

What do we know? ****ing NOTHING!!!!

 

Pretty much what I mumble to myself every time I see the thread title.

 

NBR when Pascal launches:

 

That's NBR everytime a new rumor shows up

 

speaking of NBR going crazy does anyone think the power and cooling solutions from the 980 for notebook are going to leak into the mainstream?

i think we might see beefier cards than what we normally get in laptops.

Sent from my SGH-M919V using Tapatalk

 

GTX 980 requires 17" laptop, right? I suppose there is not a lot of demand for the larger size. Will the 120W version fit in 15" models?

 

I think it was aimed at 17" and above, though i can't see why 15" cant benefit from improved cooling, or improved power phasing.

as for the second part im not sure ill go peek about. im fairly sure clevo had some 15 inchers hanging about.

Sent from my SGH-M919V using Tapatalk

 

No, it's not even standard MXM. The board is longer and has an extra cable for moar powah.

 

Funny, that article is from March 2014. So it means NVIDIA had Pascal ready for 1.5 years now??

What do you guys think: will the next mainstream mobile GPU (meaning the xx60m series) still be a Maxwell rebrand, or a real Pascal one?

 

pascal is going to be a big leap in performance from what im hearing, so i'd imagine most if not all of the new GPU' should be pascal.

Sent from my SGH-M919V using Tapatalk

 

Yeah, they'll all be Pascal. NVIDIA is done with mobile Maxwell in 2016.

 

Some new Pascal rumors:

- there may be also variants with GDDR5X (instead of just HBM2)
- GDDR5X will be improvement over GDDR5 (still on 256-bit bus but with 448 GB/s bandwith)
- Nvidia is already testing also smaller GP104 chips (not just the flagship GP100)

http://www.tweaktown.com/news/47976/nvidia-rumored-use-gddr5x-next-gen-pascal-based-gpus/index.html
http://www.3dcenter.org/news/hardware-und-nachrichten-links-des-12-oktober-2015

 

Looks like GDDR5X will be for entry-level cards. HBM2 will be for mainstream flagship cards (i.e. 1070, 1080, Titan).

http://wccftech.com/nvidia-pascal-g...ns-hbm2-mainstream-performance-highend-cards/

I wonder if mobile will see both as well.

 

wouldnt surprise me if they cheaped out on mobile at first and went either the gddr5x or even regular gddr5 route. but then again, that would mean a higher chance of regular sized mxm 3.0b cards

Sent from my Nexus 5 using Tapatalk

 

GDDR5X at 448GB/s is still a pretty good improvement if it comes to mobile instead of HBM. Still practically double the bandwidth!

 

That's true. It is still a substantial improvement.

 

GDDR5X is actually confirmed to arrive for Pascal (most likely GP104, the mobile chips) if you know where to look
https://in.linkedin.com/pub/tejendra-pratap-singh-chauhan/18/b42/300
So yeah, this is actually great news because it mean no redesign of MXM cards since its just better GDDR5

It will have 2x the bandwidth of the upcoming GTX 980. More than enough for us and only slightly slower than HBM1. Couple that with new architecture and new node, and it will be amazing for sure even without HBM