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

From the article:

" Nvidia’s current mobile GPUs have dominated the gaming laptop market for a while now, but that could easily change if AMD’s numbers turn out to be accurate. It will be interesting to see what Nvidia ends up announcing."

Whaaaat?

Edit: whaat? I ready Nvidia as AMD. Nevermind me... nothing to see here... move along.

 

Correct me if I am wrong, I am under the impression AMD at the moment only have 2 die based on polaris.
One with HBM2 thats for high performance desktop(big die, unlikely for mobile).
One that is demonstrated with the low power draw which makes no real difference for upper high end.
If I am correct, then it make sense for nvidia to rebrand since there is still nothing that challenges them in terms of high end mobile market.

 

AMD demonstrated thus far only entry level Polaris desktop gpu using GDDR5... they said that they also have a high-end version Polaris too, but we hadn't seen it in action just yet... and the high-end part will have the power enhancements of entry level too... so I wouldn't say that Nvidia will have no challenges in the high-end mobile market (Especially since AMD announced that they will release both desktop and laptop parts at the same time).
Now, whether laptops get HBM or not is unknown for now, and they said that they still have more power optimizations to do (the demo-ed part was just a showcase, and right now they demonstrated a gpu with half the power consumption of Maxwell producing same frame rates).
There's a possibility that after all power optimizations are complete and high-end mobile parts are released... a 100W mobile high-end Polaris (or one with higher TDP) might match 980Ti - or at least, Polaris architecture should theoretically be capable of doing that - especially if paired with HBM.

Polaris architecture was mentioned to have 2.5 times more performance per watt than Fiji.
Meaning, that if AMD decides to give us a mobile Polaris Nano, they could effectively cram R9 Nano Fiji level of performance into 70W (giving us desktop 980 level performance)... and I would imagine that another 30W to 50W might be able to push sufficient perfomance increase to get us to 980Ti level performance - though I'm extrapolating from desktop gpu that uses HBM as is... in absence of HBM, we might get between desktop 980 and 980ti level performance since power savings with HBM would be relatively small for the overall card (the biggest beneficial factor would likely be in reduction of the overall size of the GPU).

 

I will surprised if the other high performance gpu doesnt have a 250w power draw. It probably sacrificed power draw for performance. I dont remember the last time we saw a mobile card using top end flagship gpu die since 480m.

 

We use desktops at work, but I've only ever owned laptops for my personal use, so I don't have any experience building them, etc. - hopefully you guys can advise me on this.

If I buy a desktop with Z170 or X99 motherboard now, is it fairly certain that Pascal desktop cards will be compatible? (for example, the 1080Ti or whatever they will call the GP100 model similar to 980Ti )

 

Desktops are pretty universal when it comes to video cards. As long as it's PCI-e (which 99.9% of them are) you will be fine.

 

I'm guessing the mid-range desktop and laptop next gen gpus will be 4k certified some how, now that 4k TVS are becoming the standard.

 

If that happens then the GTX 1060M has to outperform the GTX 860M/960M by a significant margin. The current-gen mid-range GPUs run Witcher 3 at 4K resolution at 5-10 FPS. The new GPUs have to be at least as fast as the GTX 980M is today, and I don't believe it'll happen in one generation. I'm happy to be proven wrong, but I'm not getting my hopes up. I'll believe it when I see it.

 

If by "4k certified" you mean 4k @ min 30FPS, then mid-range would have to perform as good as a desktop 980 Ti. I don't see that happening. Top end cards yes, but mid-range, no.

Right. Mid-range might be closer to 980m which just can't handle 4k.

 

Can you explain a bit more how this will work with NVlink? It's not a new kind of connection to the motherboard, then?

 

Yes it is a new connection... It will be more for servers though initially..

 

Nvidia mentioned twice perf. Per watt gains on Pascal. So I would not get my hopes up for a massive performance boost (though we won't know until they actually release it) that would allow 4k gameplay on mid range mobile gpus. Volta might provide such gains though, bur Pascal... not so sure. Besides, Nvidia will use 16nm vs AMD's 14nm, so it remains to be seen if they will even be able to match, let alone surpass Polaris

 

Pascal will be compatible PCIe. NVlink is more for compute intensive tasks and for commercial type hardware. It may make its way on to some high end consumer multi-GPU SLI motherboards, but requiring a new motherboard would just sink that ship quickly.

 

Some rumors regarding Pascal Launch by end of 1H 2016 and availability in 2H 2016.
http://www.tweaktown.com/news/49815/nvidia-launch-next-gen-pascal-gpus-hbm2-2h-2016/index.html

While GCD (march 2016) seems to focus of VR and API (Directx, Vulkan).
http://www.gdconf.com/news/gdc_state_of_the_industry_most.html
http://hexus.net/tech/news/graphics...s-detailed-multiple-directx-12-vulkan-events/

 

Computex.

 

After reading a lot of news* releases and watching all of the CES videos, it's quite clear that NVIDIA is under immense pressure from car companies to design computing hardware capable of handling autonomous vehicles. Also, the 920MX, 930MX, and 940MX were the only confirmed GPU's re-branded with the "MX" (Maxwell). There is nothing regarding a 960MX, 970MX, or 980MX. Not sure where those rumors came from. The low-end cards are almost always re-branded. It means nothing in regards to gaming flagship GPU's. Pascal is still going to be officially announced 1H 2016 (probably at Computex). And when they announce new products, they release the first of them within days, not months. There's still a good chance for Pascal GPU's to be sold by Summer 2016.

*Official press releases from NVIDIA and other GPU manufacturers. Not speculative articles written by journalists looking to make a name for themselves.

 

Except the desktop cards will likely come first, and mobile to follow months later.

 

Yeah, of course. But I think they are returning to the original launch schedule for Pascal. The 700 series was the most recent "normal" launch. It was announced in April. The 780 was launched on May 23rd and the 780M was launched on May 30th. Desktop almost always comes first, but Mobile follows soon after.

http://nvidianews.nvidia.com/news/n...world-s-fastest-and-thinnest-gaming-notebooks

Funny thing is, they announced the first 700 series card on April 1st. Everyone thought it was a joke, lol.

 

The only thing we can do as consumers is wait. It does make buying decisions difficult though. But you know with 99% certainty that Pascal mobile will be out by end of 2016.

 

I hate to always be the guy that tries to extinguishes the hopes of new releases. Although 700 series mobile launch is completely a different scenario compared with this time. Probably the closest recent case we have is the jump from 580m to 680m.

If anything, the newest Pascal big die will be marketed towards the professional markets first in the forms of teslas. Why would you sell it to gamers for 1000 dollars when you can sell it to professional markets for 3000-4000 dollars? You sell the rejects to gamers after the yield had improved enough and the demand of professional market start to wane. We will probably see some forms of GP104 in July/August depending on AMD's launch schedule and their offerings. Which leads us to an Q4 launch schedule for mobile pascal. Dont expect GP100 for consumers until December.

 

EVEN BETTER THEN! The 680 was launched in March.

I'm not comparing products. I'm comparing launches, schedules, releases, announcements.

 

Which was after 3 month of amd launching their high end cards.

 

That's exactly why I referenced the 700 series being "normal." This launch schedule was meant to be the norm until AMD decided to re-brand because of Intel, which then screwed up every launch after. It's not really normal, that's why I used quotes.

 

For what it's worth, supposed updated Nvidia GPU roadmap for 2016-2017:



http://techfrag.com/2016/01/25/nvid...-in-april-gtx-1080-in-june-and-volta-in-2017/

 

Most of things in that article is believable except the titan pascal launch date. That is just so weird that it is silly. Also, the die size for GP100 is small, I cant remember the last time nvidia had a flagship die thats below 500m2.

Why I think the titan pascal in april is silly? Well, nvidia needs products to sell in 2017 and unless the initial titan is overly crippled, there is not a whole lot you can do with it besides a 1080 ti in early 2017. Launching it early will also reduce 1080/1070 sales. If you dont launch titan p in april, you will get a lot more sales for 1080/1070 since many people cant wait anymore. Then you bring in the titan p in q1 2017 to milk everyone again. Rebrand the said GP 100 as 1080 ti with some cut shaders in june/july, you have enough high end products to last you the whole 2 years. If you do bring in the titan in april, and rebrand it in early feb, what are you going to do in 2017 q3-q4 when you dont have volta out yet?

I do believe an HPC product based on GP100 launch in April is likely though. The only way we will see GP100 in consumer form this year is that AMD is hyper competitive with polaris and have over 20% performance edge over GTX 1080 and price it very aggressively.

Another thing, G80 is Nov 2006, GT200 is June 2008, GF100 is March 2010, GF110 is released in Nov 2010, GK110 was released in Feb 2013, GM200 is released in March 2015. We are looking a roughly two year time frame for big die launches barring refreshes(GF110, GF100), so it would be weird for them to pump out another big die in the span of one year.

Ofcourse, this stuff is all of my own opinion/ranting.

 

Yup, it does feel overly optimistic. For example HBM2 isn't supposed to come till H2 2016:

http://www.kitguru.net/components/g...-gpus-rumoured-to-launch-in-2nd-half-of-2016/

-------------

As for GP100 chip size - maybe 16nm process yields are so bad they have to go with smaller chips?

Kinda like what Intel is doing with Skylake, those CPUs are tiny:

 

Hey, that's ME!

 

LOL, it's like "Look Ma! I'm on TV!"

 

Yes Sir! You did a great job undressing those pieces of silicon

 

Nothing like some high quality silicon implants.

 

That soldered BGA view makes me go

 

Can't help it, can we? Just take a peek at the W230SS thread - easily 30-40% of the last 20 or so pages are people complaining about their NVidia GPU failing (including myself). The saving grace is that the CPU is rPGA 947 instead of BGA.

We need a smaller MXM form factor for mid-range GPUs like the x60M series, instead of just the x70M and x80M/x80.

 

BGA shouldn't have anything to do with those failures. GPUs are connected to MXM boards via BGA.

 

I think the point is that they wouldn't have to replace the motherboard if they didn't have BGA GPUs.

 

Exactly. And replacing an MXM GPU-only board is easier and cheaper than replacing the entire motherboard.

 

Its curious that so many of them are failing in the first place though. With the limited overclocking and the low heat output, it seems pretty shocking.

 

Indeed. I think Clevo just skimped on the W230SS quality control. This issue apparently shows up on the W230ST, too, since my friend's GTX 765M failed. The prevalence is a massive sore thumb.

All in all, it hasn't been a very good showing by Clevo for their smaller models, since they seem to be focussing on the desktop replacements. The P640RE has pathetic battery life, and an odd layout of the motherboard. It doesn't even have G-Sync, has worse internal connectivity than the W230SS and all in all, feels a bit half-hearted. Certainly, it's got Clevo's trademark good cooling, but besides the GTX 970M and a sturdier metallic chassis, it is somewhat of a downgrade from the W230SS.

Edit: there's yet another post on the W230SS forum about the GPU on the W230SS failing.

But I digress. What I really want Pascal to bring, or the GPU industry in general, is an update to the MXM form factor that allows for smaller boards, and hence opens up possibility for MXM in 13/14" laptops.

 

I'd probably be in the market for a small 14" pocket rocket with MXM modules.

 

So GDDR5X shouldn't come before mid-2016 either:

Starting to feel Pascal is actually blocked on memory - GPU chips seem to have been taped out already earlier but neither HBM2 nor GDDR5X seem to be ready (it'll not make sense to launch any Pascal GPU without either of those new memories).

Here AMD may have some temporary advantage if they could use HBM1 as a backup plan (so far they just told Polaris will come both with HBM and GDDR5 options, not telling subtypes: if it's going to be HBM1, HBM2, GDDR5 or GDDR5X).

 

HBM2 = 4096 bit bus. That's hilariously huge. Love it.

Also, since it was AMD and Hynix that actually came up with HBM in the first place, AMD should get first dibs on whatever is avaliable. NVIDIA can suck it.

 

Keep dreaming. NVidia has the cash and Samsung has the chips.

 

This is sadly true. Samsung has the most advanced fab outside of Intel, and AMD being the cash strapped company they are...

Sad that AMD and Hynix came up with it, but NVIDIA and Samsung are the ones that are going to profit big bucks from it.

 

http://www.kotaku.com.au/2016/01/rumours-nvidia-will-unveil-their-flagship-pascal-gpu-in-april/

http://www.pcadvisor.co.uk/feature/pc-upgrades/nvidia-pascal-uk-release-date-price-specs-3634003/

http://techfrag.com/2016/01/25/nvid...-in-april-gtx-1080-in-june-and-volta-in-2017/

There are a few more articles I didn't link with debatably better information. These are fairly speculative.

 

But what does this mean for mobile? Still sounds like a Q4 intro for mobile considering desktop 1080 isn't until June.

 

From what I've seen, and everything I've read, my gut is telling me they are returning to the 700 series launch schedule behavior:



We may see the 920MX, 930MX, 940MX, etc. early on, sometime in April with the first Pascal Titan launch. And then Computex is the official launch of the remaining desktop cards. And finally within a couple weeks mobile will launch. Almost identical to the 700 series. This launch schedule (what I've described) makes absolute sense when you consider market behavior. Waiting two fiscal years to release new products is unheard of.

Will mobile have HBM or HBM2? Possibly. Will it have GDDR5X? Maybe. This I can't tell with any certainty. Samsung began mass production at the beginning of this month. Give them six months (until the end of June) and it's possible we may see mobile cards shipping with HBM2.

Highly unlikely to be a Q4 launch for mobile. That makes zero sense.

P.S. I'm not saying it's exactly like 700 series launch, but similar.

 

Considering they just refreshed the 965M, I'd say that we aren't going to see Pascal for awhile yet on mobile.

http://www.notebookcheck.net/Nvidia-GeForce-GTX-965M-2016-Refresh-N16E-GR-Review.158192.0.html

 

Low-end cards are almost always re-branded.

https://www.google.com/search?q=gtx+960m+release+date&ie=utf-8&oe=utf-8

860M = 960M (overclocked a bit)?
965M = 965MX (overclocked a bit)?

 

965M switches from GM204 to GM206 so it's a bit more than a rebrand.

There was a lot of hope that they would release little Pascal as a x6x chip like they did with Maxwell, that doesn't seem to be happening either.

Pascal is not coming to notebooks any time soon. Q3/Q4 is what I've been saying and I'm sticking to that. Once we see a desktop Titan release, then we can start looking for a notebook chip.

 

Although that's a really short window for Pascal - only about a year before Volta comes in. I'm thinking they'll want to milk Pascal more than that - after all we're talking about NVidia here, they sure do love that dairy!

 

Yea, and releasing pascal without 5X or HBM would just create another step inbetween to be milked.