[Guru3d] Nvidia Big Pascal GP100 To Debut in April - Mid Range GTX 1080 in June (?)

[xthetenth]

Folks can keep hoping for a $250 card using 14nm and HBM2 to hit by June, but they shouldn't hold their breath.



Care to look up when the 5850, 470, 7950, 670, 960, Fury and all the other second-tier cards hit?

Hint - it wasn't at launch.

This is a launch like none other. New micro-architecture and new memory standard rolling out together. It will be complicated, low yield, and ultra-high-end.

Well for starters, the 960 was the third tier chip (or fifth if you're going by card rather than chip, which is what everybody else in the discussion is talking about). The 980 was the second tier chip. That's the entire point you're missing. Fiji was either the only chip of the generation or a massively late follow on to the 285/380. Again, you've defined top tier in terms of what's released, and of course they're going to have released the top released card, it's literally not impossible to. If you're discussing the top chip in the family rather than the top chip in the release cohort, then suddenly we're back to pointing out that the 680 released before the Titan and the 980 released before the Titan X, so why are we expecting a much more ambitious card to suddenly break ranks and release without a pipe cleaner?

 

[Headfoot]

Was the GTX 680 mid-range? Did it or did it not hit the market at $500, and was it or wasn't it the fastest card on the market at that time?

There hasn't been a single new-node or new-arch card released <$500 in the past 5 years. Are people seriously holding out for a mid-range card on 14nm in Q2?

Yes, the GTX 680 was midrange. It had a high end price and was marketed as high end, and apparently some people fell for it. But it is undeniably the midrange chip in the kepler family. There was GK106 below it and GK110 above it. That is by definition "in the middle." It had a middle die size of the dies in Kepler, GK110 > GK104 > GK 106 > GK107. Especially when you consider the lowest 2 die sizes were more laptop focused.

Nvidia did this again for Maxwell. It is fairly clear they prefer to release the midrange chip first as "high end" then sell another chip to the same person later in the same architectural family and node size for the big one

 

[poofyhairguy]

Folks can keep hoping for a $250 card using 14nm and HBM2 to hit by June, but they shouldn't hold their breath.

Your wisdom, it's blinding.

It is pretty funny looking at old threads on this forum from say a year ago for people looking at midrange cards who are told to "wait for Pascal." Seems like GPU consumers have been told to wait and wait and wait for the next generation that has just been around the corner since 2014. I also think last year the dam finally broke on people getting fed up with waiting which is why the GTX 970 is the number one GPU in Steam.

I think it is much better to be pessimistic, then you aren't let down by Nvidia's release schedule of doom. GTX 970 replacement? 2017. 960 replacement? 2018. 980 Ti killer for $300? 2020.

 

[Silverforce11]

This is from a roadmap that was presented in 2014, then updated recently.

Roadmaps could be real, if everything works out.

 

[DooKey]

Yes, the GTX 680 was midrange. It had a high end price and was marketed as high end, and apparently some people fell for it. But it is undeniably the midrange chip in the kepler family. There was GK106 below it and GK110 above it. That is by definition "in the middle." It had a middle die size of the dies in Kepler, GK110 > GK104 > GK 106 > GK107. Especially when you consider the lowest 2 die sizes were more laptop focused.

Nvidia did this again for Maxwell. It is fairly clear they prefer to release the midrange chip first as "high end" then sell another chip to the same person later in the same architectural family and node size for the big one

AMD set the tone for the GTX 680 and NV thanks them for it.

 

[tential]

Yes, the GTX 680 was midrange. It had a high end price and was marketed as high end, and apparently some people fell for it. But it is undeniably the midrange chip in the kepler family. There was GK106 below it and GK110 above it. That is by definition "in the middle." It had a middle die size of the dies in Kepler, GK110 > GK104 > GK 106 > GK107. Especially when you consider the lowest 2 die sizes were more laptop focused.

Nvidia did this again for Maxwell. It is fairly clear they prefer to release the midrange chip first as "high end" then sell another chip to the same person later in the same architectural family and node size for the big one

The people love it though.... so keep doing it right?

I mean Nvidia are amazing at stacking money.

On a side note, my little puppy looks so cute tearing things apart right now.

 

[jpiniero]

Well, nVidia could sell the cut GP100 as Titan and keep the full dies for Tesla only. Then in the second round, sell the Titan as full die.

 

[railven]

The people love it though.... so keep doing it right?

I mean Nvidia are amazing at stacking money.

On a side note, my little puppy looks so cute tearing things apart right now.

I'm back on annual upgrades. If GP104 is the fastest card when my window opens up, I'm buying it. If GP100 is the fastest card when my next window opens up, I'm buying it.
EDIT: price ceiling of $650. For AMD and nvidia. Screw Titan and if Fury decides to stretch into the tier of pricing.

But I seem to be some kind of anomaly. Not many people upgrade with every release.

If the "super rich" (tehe) are willing to open their wallets, why would Nvidia (and it seems AMD will be following suite) not want to take their money?

 

[AdamK47]

Want

 

[Techhog]

We haven't' seen a single 14/16nm GPU but yea this must mean AMD is dead. Again.

They need to release first to have a chance. If they're faster but late, it won't matter.

 

[GodisanAtheist]

Nvidia might simply be diluting their "Titan" brand name a bit more here: nothing dictates that the GP100 HAS to be a 600mm2 chip just because prior Gx100 chips were 500-600mm2 (unless I missed some major piece of information that everyone else is in on).

NV might pump out a ~400mm2 GP100, still costlier and "unyieldy" than they would like, and call it a Titan so long as it leads GM100 and the AMD competition by a decent margin. They then wait for a refresh or two when yields improve and costs drop to unleash the 600mm2 kraken.

Short Version: Just because NV calls it a Titan does not mean its their 600mm2 monster chip.

 

[DarkKnightDude]

Are they actually calling it the GTX1080? Maybe it'll be able to run every game maxed out at 1080p at 60 FPS?

 

[dacostafilipe]

So many articles that use ascii.jp's roadmap as "source" ... kinda scary :/

 

[beginner99]

If it is releases which I doubt it would be as tesla card only. They can easily sell all of them for 5x times the price of a consumer titan. This won't happen.

 

[JoeRambo]

It would make a lot of sense for Nvidia to release Tesla big boy first. They are under epic attack (understatement) from Intel's new Knight stuff, so showing up with new products is imperative for them.

Customer "Titan" -> i will believe it when i see it.

 

[guskline]

Want

Who has dibs on your 4 TitanXs ?

 

[beginner99]

It would make a lot of sense for Nvidia to release Tesla big boy first. They are under epic attack (understatement) from Intel's new Knight stuff, so showing up with new products is imperative for them.

Are they really under attack? At least in scientific research NV is so entrenched with CUDA I hardly think that will change just due to prior investments made.

 

[ShintaiDK]

Are they really under attack? At least in scientific research NV is so entrenched with CUDA I hardly think that will change just due to prior investments made.

They are under serious attack in the HPC space from Phi.

 

[Headfoot]

AMD set the tone for the GTX 680 and NV thanks them for it.

What?

 

[Kenmitch]

fake roadmap = no info available on potential product(s) release cycle/dates in the end.

Deep pockets for kickbacks = info on the other teams potential performance/products.

Above = we know nothing at this time....Other than the Polaris teaser and the mocked up dummy card JHH held up.

 

[TestKing123]

GTX 1080 is official name? Don't think it is. Nvidia "reset" back to a low number the last time this happened:

Geforce 9800 series --> reset to Geforce GTX 270/280

 

[Unoid]

Both AMD and Nvidia always release the highest-end first when a new node or arch arrives.

5870 Sept. 2009
480 May 2010
7970 Jan. 2012
680 Mar. 2012
Titan Mar. 2013
R9 290X Oct. 2013
980 Sept. 2014
Titan X Feb. 2015
Fury X June 2015

Sole exception was 750 Ti because Maxwell Gen. 1 couldn't scale up.

Folks waiting for midrange cards will be waiting a long time. Q2 is all about cards $700 and up.

9600Pro disproves that statement. Maybe that's too old for your trending. I know a thing or two on the 9600pros

 

[Termie]

9600Pro disproves that statement. Maybe that's too old for your trending. I know a thing or two on the 9600pros

The market was different in 2003.

 

[moonbogg]

All I know is as soon as Pascal is released, my two 980ti's are going straight into the trash bin where they belong. Hot, slow, overpriced and underperforming junk!
Day 1 of Pascal release, Nvidia sends out a "game ready driver". Moonbogg downloads it. Suddenly Moonbogg's 980ti's are getting 30fps in BF4.
I predicted it. You read it here first.

 

[n0x1ous]

All I know is as soon as Pascal is released, my two 980ti's are going straight into the trash bin where they belong. Hot, slow, overpriced and underperforming junk!
Day 1 of Pascal release, Nvidia sends out a "game ready driver". Moonbogg downloads it. Suddenly Moonbogg's 980ti's are getting 30fps in BF4.
I predicted it. You read it here first.

We'll have to start a recycle bin for them lol