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

With nothing on both fronts, for now, I guess it can't be otherwise, but to enjoy what we have.

What you want to know about my point? You want me to do a list why I would never trust this particular guy? And no, you are the "typical" one, because you started to get personal, making fun of my believes and now you're right there in the corner having nothing else to say, but to play the "touched girl" card.

 

Well since this is a speculation thread after all who wants to bet on what kind of performance increase we'll see?

I'm going to say 35% on the next flagship card over the 980m.
Even if Pascal can do a bigger jump, pretty sure they won't give us a doubling.

PS. For what it's worth I don't trust a word that comes out of Jensens mouth either. I'm sure if you asked him how much VRAM a 970 has, he'd happily respond with "4gb at full speed" with a sh#t eating grin.

 

trying to predict when itll launch is dumb. i remember early last year, nvidias ceo popped out of nowhere onto another companys staged, walked up to the ceo and said, '' heres a gift. i call it the titan x. its your gift'' drops mic then walks out.

 

That's because the default SLI profile is borked and scaling is atrocious. Here's the fix.

Charlie's rants aside, his research is usually quite solid. I definitely wouldn't dismiss him as "some journalist on a website". This is his self-professed track record. Yeah ok sure it's self-aggrandizing, but the man has industry connections. When you read his pieces, just filter out the emotionally charged filler and stick to the facts. Many times you end up learning something new.

And his R&D team gave him wood screws for the Fermi mockup. Seriously.

Yup who could forget those awesome wood screws. And yes you're right about the bold part. G80 (of 8800 GTX and 8800 Ultra fame) was the first monolithic die nVidia made. However it was not their first time making 90nm based dies, as they had 90nm based G71 and G72 before that. Likewise before making 65nm and 55nm based GT200 big dies, they had multiple smaller dies (G92, G94 G98, and likely others I'm missing) out on those two processes.

Fermi was indeed the first time they came swinging out the gate with a big die on 40nm based on a brand new architecture. Also didn't help they updated the memory from Tesla's GDDR3 to Fermi's GDDR5. So the combination of new arch on new process previously unvalidated, along with the update in memory caused the perfect storm that gave Fermi.

If you really think deep about it, one could draw chilling similarities between what happened with Fermi and what will happen with Pascal:

Fermi: 55nm to 40nm shrink, new architecture updated from Tesla, memory change from GDDR3 to GDDR5
Pascal: 28nm to 16nm shrink, new architecture updated from Maxwell, memory change from GDDR5 to GDDR5X/HBM

So given past precedence, I wouldn't be the least bit surprised if Pascal is indeed having some major issues right now. Although nVidia likely learned their lesson, and won't release another GTX 480 again (hopefully, but Jen-Hsun's arrogance and ego knows no bounds).

Ain't that the truth! Seriously TSMC has screwed up EVERY SINGLE NODE starting with 40nm, which gave rise to the infamous GTX 480. Then they completely scrapped 32nm HKMG, which was what the HD 6000 series was designed around. This of course forced ATi to retool HD 6000 to 40nm, causing them to lose the performance crown to GTX 580 (and personally I think this event marked the beginning of AMD's GPU division's long spiral downwards). Then of course they had initial yield issues with 28nm, and 20nm was pretty much an epic fail and complete writeoff.

And don't forget, TSMC has NO prior experience with FinFET either. Given TSMC absolutely ABYSMAL track record over the past 5 years, I'm always amazed at how much faith people have in them actually being able to deliver on time and in the necessary quantities.

I mean to put that in perspective, if you thought AMD was bad in terms of delays and empty promises, TSMC is even worse on those two fronts. Just let that sink in for a second. I'm too lazy/bothered to go dig up their old roadmaps, but if you have the time and will, as you start digging you'll very quickly see how their roadmaps have ummm, evolved over the years and how timelines got magically shifted around.

 

First of all, I'm not dismissing anyone. Let's make that very clear. I'm only saying I'd take the word of a company over a writer/journalist (who may have contacts?) if the two are going head to head. Public companies are extremely predictable. Individuals are not.

Didn't know about the fermi mess. Seems unlikely to happen again, considering how long ago that was, and how many launches they've had since.

All that aside, I was reading through 2015's shareholders meeting summaries and all of the deals NVIDIA made total into the billions of dollars, a majority of them relying on Pascal's ability to handle deep learning. It's extremely unlikely they'd purposefully lie and set themselves on a path toward bankruptcy. They've invested so much into this architecture, and have been developing it since Kepler was launched. More than 50% of the Research and Development budget was for Pascal in 2013.

If the CEO says 1H 2016, it's because he can't offer a specific quarter to investors, and is prepared for delays. So, this time frame accounts for that.

My understanding is that Comptex offers them a perfect time to both announce the gaming product line and launch the first of the desktop cards. April is when we should see the first of the "car" products hit the market, with Tesla first in line for the new tech. Not to mention, they have a booth reserved for the event.

 

Eh personally I'd rather trust an independent tech writer than the CEO of a company, whose only aim is to please shareholders and maximize profits, and thus would be willing to say and do anything and distort the truth to make it seem favorable to his narrative.

As for the Fermi mess, only reason I highlighted it is because of the many parallels we can draw between what happened with Fermi and what is going on with Pascal. Namely TSMC troubles, come swinging out the gate with a big die on a new process with a new architecture using new memory. Do I think they're going to make another 480? Probably not, if only for PR reasons. But there's no denying Pascall on 16nm FF+ is the most challenging design they're going up against since Fermi. (both Kepler and Maxwell still used GDDR5, and Maxwell didn't even involve a process shrink)

Yes nVidia's main source of income is from the professional/HPC market, which is what Pascal is supposed to be good at (remember Maxwell can't DP/FP64 to save its life). Remember how nVidia got the Oak Ridge contract, and almost couldn't deliver because TSMC was such a screwup? They had to scrap GK100, then harvest every single good GK110 die they could get their hands on. So I could probably believe 1H 2016 for Tesla based GP100 to hit the market, but I don't think we'll see GeForce parts for quite a while after that.

 

You make some solid points there. The fermi issue does raise a flag for me. Sorry if I came off offensive or like a jerk, wasn't intended. Actually wrote that last post on my cell (mod can check IP to confirm). Just trying to get to the point.

In my opinion, it would be a big mistake to release Pascal any later than the end of Q2. But that's just me.

 

No worries, it didn't come off as rude to me, you're just stating your opinions.

To be honest I'm also just speculating as much as everybody else is, but I guess I'm trying to bring some additional facts to the table and keep the hype in check a little bit.

IMO with the marketshare nVidia has, they could afford to sit on their rear a little bit. Unless AMDs surprise launches Polaris and it blows the pants off of Pascal, then we might see them scrambling. (oh that reminds me, ATi did have HD 5870 out 6 months before 480, and 5870 absolutely slaughtered GTX 285, so nVidia was extra desperate to rush Fermi to market)

 

The other thing is that DP cores running much more hotter/power hungry than SP cores. GK110 on full DP have to throttle down to 700mhz core from 1000mhz. One of the other reason why maxwell can do what it does on a larger die size. Fermi just shows nvidia CEO is willing to lie.

 

May I kindly ask then,

do you think the MXM3.0 type B, (the universal format that Clevo is using, like 980M, 970M, not the current newer L-Shaped MXM board for the 980) will still be enough to support pascal based hardware?

 

Unknown for HBM depending on die size and HBM placement but GDDR5X definitely should see standard MXM cards.

 

HBM should need special heatsinks i would imagine. Otherwise probably MXM 3b or MXM3c for large die.

 

That's what I'm thinking as well but depending on the die size and HBM layout, it may need a new screw layout as well.

 

You'll almost certainly need new screw layouts if going HBM. Die+HBM+interposer = one gigantic package

 

My guess is that if they HBM the second-to-top-tier (the chips that usually end-up in notebooks) they would fit in the current layout, because new process The thing is that HBM would most likely be top-tier exclusive this year as well.

 

Lol.. I love it, 122 pages and suddenly dead, but we now know...Pascal isn't really close

Apologies for the Necro.

 

necro after 4 days? cmon, lets not exaggerate

Sent from my Huawei Mate 8 NXT-AL10 using Tapatalk

 

Around here that's definitely a necro lol

 

As for the release date of the NVIDIA Pascal GPUs, some observers are saying that it might launch early this year, since it was already showcased at a major tech event. It has been rumored that the new NVIDIA GPUs might arrive as early as April, and will reportedly replace the Titan-grade series of graphics cards.

early april according to some site??sounds good even i still hope they release it somewhere in march

 

That's some mighty wishful thinking.

 

One could even say imaginative.

 

showcased where exactly? havent seen any functioning pascal gpus anywhere yet...

Sent from my Huawei Mate 8 NXT-AL10 using Tapatalk

 

Yeah.. We've seen nothing.

I think he's referring to the Maxwell prop dressed up as Pascal.

 

"prop"..

 

Lol, these posts.

 

Heheh... this is my type of 'prop':

 

twiddling thumbs and passing the time with "stuff", wonderful

 

So, I've been reading up on the "Fermi incident" back in 2009 and 2010. On the previous pages, I was told to, "Look at Fermi, Pascal won't be ready in April. Things don't go as planned." But every article I read about this incident says that even though NVIDIA did only show a "mock-up" design with wooden screws, the complete model was in full production at the time, and they even launched in March of 2010 (on schedule), despite facing these challenges.

What does this suggest to me? It's entirely irrelevant. Nothing has changed in my opinion. I've known from the beginning that there would be challenges and delays, but schedules are kept. As I've said many times before, a delay can be as little as a day or a week, and can later be countered by a furtherance in the production schedule. Sometimes there's a delay, sometimes they're on time, and sometimes they're ahead of schedule. It balances out in the end.

TL;DR: NVIDIA showing Maxwell in the Tegra (Pascal) model means absolutely nothing. They just wanted something functional to show as opposed to having a non-working mock-up to avoid the nightmare of Internet overreaction. For all we know, production is still on schedule, or maybe even ahead of schedule.

Guess we'll just have to wait and see. We're only a few months away. A lot can happen in 90+ days.

 

This NEVER happens in ANY production environment.

 

You mean for NVIDIA, AMD, etc. or everything in the world? Because it sure does happen. Surely not often but it does, hence "sometimes."

Just making a point about production in general.

 

Pretty much in general. Usually timing is so tight with anything that it's darn near impossible to meet the expected date as it is.

 

Ah.. Well, I'd agree with you straight up if you said "in computer hardware manufacturing." I've never witnessed something ahead of schedule, at least nothing of significance. But for other products, like cars, batteries, etc. they do present ahead of schedule, and more often than you'd think. For example, Tesla's production has been ahead of schedule several times ( source). Google has also launched multi-million dollar initiatives on products ahead of schedule.

It really depends on the industry. NVIDIA being ahead of schedule will likely never occur because of AMD and Intel fiddling thumbs. I guess if you take into account the release schedules of AMD, Intel, and NVIDIA... NVIDIA is actually "ahead" of them. But that's technically wrong.

 

Fermi was 6 months too late and still had massive heat/power issues. I dont think nvidia is producing working fermis on any decent scale at the time of the demo.

Not sure why wont nvidia just state its a mockup instead of lying and breaking SEC rules, unless you know, they are hiding something.

 

Anything can be ahead of the schedule as long as you publish a different external/internal schedule.

 

Exactly. A public announcement of availability =/= meeting timing. Internally they could have had multiple delays and just just posted a public availability date. I've worked in automotive engineering for over 20 years and not once have I ever seen something done ahead of time.

 

Personally i think that it is not necessary that the only reason behind Nvidia not showing pascal GPU of PX2 is that its not ready or they are far behind (it could very well be the case). It could just be that they didnt want to show pascal ahead of its release. The demo was for their autonomous drive system not GPU. So it makes sense to hide you brand new GPU that you plan to launch few month later and use Maxwell as a placeholder.

And yes, they should have made clear that model being presented this is not using pascal GPU.

I am not saying that everything is ok. Lack of any leaks and credible information is definitely suspicious, but i have not yet lost all hope

 

I don't read too much into the Maxwell posing as Pascal incident either, but with regards to Fermi, do you not remember how hot and power hungry the 480 was? Worst part is 480 wasn't even a full chip, it had one set of SM disabled, and even then it still ran hotter than the sun and guzzled down power like crazy. The 480 was basically "failed Fermi", and 580 is what Fermi should've been like had nVidia took the time to do it properly. Bottom line is nVidia may have launched 480 "on time" but they did so at a huge cost.

 

Back when that was going on, I was 16-17 years old. Didn't own any gaming rigs at the time. Just built regular computers. Joined the gaming world when I turned 18 and got my first credit card, lol. That was a huge mistake, by the way (financially).

 

LOL. I think I am younger than that when fermi rev 1 came out. Cant believe how long ago was that.

 

lol you whipper shnappers! That was just 2010 wasn't it?

 

lol, u spoke my mind wingnut

Sent from my Huawei Mate 8 NXT-AL10 using Tapatalk

 

It is crazy how fast time goes by. Enjoy it!

 

Well when Fermi launched I was a poor student just finishing up college, so obviously missed out on all the good stuff.

(I was still using a single core AMD Turion laptop into the first half of 2011 LOL. Looking back I'm really not sure how I survived)

 

In other news, GDDRX5 enters mass production this summer and early components looks promising:

https://www.micron.com/about/blogs/2016/february/gddr5x-has-arrived

 

hey wingnut, theyre AHEAD of schedule!

Sent from my Huawei Mate 8 NXT-AL10 using Tapatalk

 

So I guess this means this is yet another piece of the puzzle showing that we probably won't get any mobile products from NVidia with Pascal before the Summer (probably well after), because didn't we think GDDR5x was the most likely candidate for mobile Pascal. Good link Mr Najsman.

 

http://wccftech.com/nvidia-pascal-graphics-cards-spotted/

NVLink? Sounds interesting!

Sent from my HTC One_M8 using Tapatalk

 

Old news, pictures, and schtick, but one fun comment:

Pause
And
Stop
Could
Arive
Late

 

lol im still paying monthly bills on my 880m with a credit card from almost 2 years ago and i hate this crap machine now lol. sucks