New Pascal Titan X!

[Keysplayr]

but my last post wasn't actually hyperbole. mid range gpu is now 400$(that is just msrp, actual price around 450$) when it was 200$ just a few years ago. the last gold standard was gtx 460 1gb. even adding 10% every year per your post, it still wouldn't double. :thumbsdown:

Do you remember the price of a 6800GT? I sure do. Bought one for myself at the first day of availability at COMPUSA. 399.00. That was a 2nd tier card. Thee was the ULTRA above it. Somewhere above 500 bucks wasn't it?

At any rate, GTX 680 was 499 at launch. 780Ti was how much?
Look, you can complain about pricing all the live long day, but in the end, there will be a product that suits your price and performance. I don't see a problem. It all depends on what you wish to spend and what you're happy with. Being unhappy because one cannot afford the highest tier is a waste of time. Otherwise, everyone would be driving mercedes when a kia would do.
Just buy what you want to at a price you're able to live with. That's all there is too it. Mourning over new pricing is a waste of time when there isn't anything anyone can do about it EXCEPT the actual act of mourning. Gets nothing done.

 

[boozzer]

nope, nv hasn't release a product that I wanted since 660 ti. they are all either too expensive or not enough performance. 6800 gt? the card that was rebranded into 770? mid range?

and when did mid range = top tier in your book?

 

[Squeetard]

GEForce 3 cost me $599. But it killed on Max Payne!

 

[Aristotelian]

It's easy to have that attitude when you are paid by nvidia. You get free video cards so obviously you don't care how they are priced.

Meanwhile consumers are asked to pony up another 20% every couple of years, for no reason. Imagine if every new pc component went up by such amounts. There wouldn't be any fun unless you can justify thousands of dollars every couple of years.

New nodes are supposed to be the herald of big changes and lower prices for performance. Instead we get price hikes!

It's easy to have "this attitude" when some people here earn enough that buying flagship GPUs is comparable to the price they pay for lunch, or a suit. It's also easy to have "this attitude" when you prioritize enthusiast gaming as a hobby. In short, I do not understand all this complaining about price hikes, as if you're entitled to have something for a price that YOU dictate. Yes, it's better to get more performance for less money, all things being equal. But all things are anything but equal.

 

[boozzer]

It's easy to have "this attitude" when some people here earn enough that buying flagship GPUs is comparable to the price they pay for lunch, or a suit. It's also easy to have "this attitude" when you prioritize enthusiast gaming as a hobby. In short, I do not understand all this complaining about price hikes, as if you're entitled to have something for a price that YOU dictate. Yes, it's better to get more performance for less money, all things being equal. But all things are anything but equal.

no one is asking for more performance with less money. just not double the money or more for the same class of gpus in less than few years.

 

[swilli89]

nope, nv hasn't release a product that I wanted since 660 ti. they are all either too expensive or not enough performance. 6800 gt? the card that was rebranded into 770? mid range?

and when did mid range = top tier in your book?

haha no.. He's talking about the 2004 6800GT. That was my third video card and perhaps one of the most exciting cards I had ever owned, it was awesome!

 

[pj-]

but my last post wasn't actually hyperbole. mid range gpu is now 400$(that is just msrp, actual price around 450$) when it was 200$ just a few years ago. the last gold standard was gtx 460 1gb. even adding 10% every year per your post, it still wouldn't double. :thumbsdown:

People here get way too hung up on die size and number of cores.

Performance per dollar in GPUs has increased way more than CPUs or any other component in the past few years.

980ti performance was $650+ last year. Now it is $200 less and uses less power.

Can I get 6700k performance at 2/3 of its launch price?

 

[boozzer]

People here get way too hung up on die size and number of cores.

Performance per dollar in GPUs has increased way more than CPUs or any other component in the past few years.

980ti performance was $650+ last year. Now it is $200 less and uses less power.

Can I get 6700k performance at 2/3 of its launch price?

that is what happens when amd have no answer for intel cpus since core 2 duo. that is why you don't want monopolies. reason the gpu side is still somewhat ok is because amd is still releasing competitive gpus.

by your metric, do you add money every gen base on how much performance increases over the previous gen? :wub: that is idiotic, extremely.

 

[4K_shmoorK]

that is what happens when amd have no answer for intel cpus since core 2 duo. that is why you don't want monopolies. reason the gpu side is still somewhat ok is because amd is still releasing competitive gpus.

by your metric, do you add money every gen base on how much performance increases over the previous gen? :wub: that is idiotic, extremely.

I'm not sure why everyone seems to think they should be guaranteed anything in the free market. Sure you can complain about prices but year over year price increases are nothing new (see the 20th century)

 

[Alaa]

The card is not being marketed as a gaming card as mentioned in the update of Anandtech's article.

 

[boozzer]

I'm not sure why everyone seems to think they should be guaranteed anything in the free market. Sure you can complain about prices but year over year price increases are nothing new (see the 20th century)

when the yoy price increase is 100% or more in a few years? complaining is the least a gamer should do, they should be supporting the opposing side with money, amd.

and then we have "supposed" gamers supporting the exorbitant price increases........................

I wish gpu market is like the automobiles market, so many brands to choose from, not just 2. free market would work a hell of alot better than now with more competition.

 

[n0x1ous]

Update today from Nvidia to Anandtech confirming GP102, die size of 471mm2, neutured FP64. So basically GP100 physically stripped of the HPC stuff like NVlink, FP64 etc etc. If I remember correctly, that is about the size of G80 on 90nm

Update 07/25: NVIDIA has given us a few answers to the question above. We have confirmation that the FP64 and FP16 rates are identical to GP104, which is to say very slow, and primarily there for compatibility/debug purposes. With the exception of INT8 support, this is a bigger GP104 throughout.

Meanwhile we have a die size for GP102: 471mm2, which is 139mm2 smaller than GP100. Given that both (presumably) have the same number of FP32 cores, the die space savings and implications are significant. This is as best of an example as we're ever going to get on the die space cost of the HPC features limited to GP100: NVLInk, fast FP64/FP16 support, larger register files, etc. By splitting HPC and graphics/inference into two GPUs, NVIDIA can produce GP102 at what should be a significantly lower price (and higher yield), something they couldn't do until the market for compute products based on GP100 was self-sustaining.

Finally, NVIDIA has clarified the branding a bit. Despite GeForce.com labeling it "the world’s ultimate graphics card," NVIDIA this morning has stated that the primary market is FP32 and INT8 compute, not gaming. Though gaming is certainly possible - and I fully expect they'll be happy to sell you $1200 gaming cards - the tables have essentially been flipped from the past Titan cards, where they were treated as gaming first and compute second. This of course opens the door to a proper GeForce branded GP102 card later on, possibly with neutered INT8 support to enforce the market segmentation.

 

[crisium]

Exactly 50% larger die than GP102 and contains 50% more of everything (3840 Core uncut version, anyway).

 

[Grooveriding]

Buy the Titan X for gaming and watch nvidia release full GP102 as 1080ti with that INT8 feature disabled. Enjoy the premium black shroud afterwards.

 

[crisium]

There will definitely be a non-INT8 GP102 as a 1080 Ti. Whether it will be the 780 or 780 Ti of its generation remains to be seen, but either way it will cost less and perform around the same. They could do 1080 Ti as further cut and 3840 as Titan Black, worst case scenario.

Die size, and the Quadro rumours, both make it virtually certain there is a 3840 Core GP102. $1200 GTX 570 equivalent, folks. Best case scenario the GTX 580 equivalent is actually cheaper if you wait (3840 = 1080 Ti). Worst case scenario the GTX 560 Ti Core 448 equivalent is cheaper (1080 Ti is further cut; 3840 = Titan Black).

Even those that demand god-tier performance at all times, they should see this as a red flag. No one would actually go 1080->Titan X-> 3840 GP102 in a matter of months... would they?

 

[Arachnotronic]

Even those that demand god-tier performance at all times, they should see this as a red flag. No one would actually go 1080->Titan X-> 3840 GP102 in a matter of months... would they?

Already selling my 1080s to get my system ready for Titan X. If something beefier comes along, Titan X will be sold and I'll get that.

 

[Timmah!]

There will definitely be a non-INT8 GP102 as a 1080 Ti. Whether it will be the 780 or 780 Ti of its generation remains to be seen, but either way it will cost less and perform around the same. They could do 1080 Ti as further cut and 3840 as Titan Black, worst case scenario.

Die size, and the Quadro rumours, both make it virtually certain there is a 3840 Core GP102. $1200 GTX 570 equivalent, folks. Best case scenario the GTX 580 equivalent is actually cheaper if you wait (3840 = 1080 Ti). Worst case scenario the GTX 560 Ti Core 448 equivalent is cheaper (1080 Ti is further cut; 3840 = Titan Black).

Even those that demand god-tier performance at all times, they should see this as a red flag. No one would actually go 1080->Titan X-> 3840 GP102 in a matter of months... would they?

Haha, they clearly would.

Anyway, i have to wonder about availability of this Titan card, especially in the light of how the release of 1080 went, or is still going - i have been waiting for AiO version for 2 months now and still nothing. Hell, even lot of air-cooled versions are still not available here, like Evga FTW for example...

I wish they would bring on 1080Ti quickly, like another 2-3 months after Titan....even if its just further cut GP102 part and not full chip, i would prefer it over 1080 and Titan obviously...since i want new CPU and mobo as well, if they release it in early 2017, that will be once again close to the release of the Skylake-X, so my wait will be once again prolonged for another bunch of months... not to mention Nvidia may reveal some Volta chip at the time of Skylake-X release, tempting me to wait once again and in the end i lose all my enthusiasm and just give up and buy nothing...

 

[raghu78]

Update today from Nvidia to Anandtech confirming GP102, die size of 471mm2, neutured FP64. So basically GP100 physically stripped of the HPC stuff like NVlink, FP64 etc etc. If I remember correctly, that is about the size of G80 on 90nm

wow. The die savings are significant. 139 sq mm or roughly 23%. I think Nvidia will stick to a Gx100 / Gx102 model for the future. I surely expect GV100 for Tesla and GV102 for gaming.

 

[Arachnotronic]

wow. The die savings are significant. 139 sq mm or roughly 23%. I think Nvidia will stick to a Gx100 / Gx102 model for the future. I surely expect GV100 for Tesla and GV102 for gaming.

That's what a big R&D budget buys you -- the ability to do targeted products across the board.

 

[Headfoot]

My bet is the reason GP102 came so fast after GP104 is that they're reversing the situation from Maxwell. Instead of the big one taking longer to arrive, with the Ti variant shortly thereafter, I bet you nVidia will milk the Titan line for quite a while before dropping the 1080 Ti since AMD has nothing in the high end market until q1 2017 or later

 

[Headfoot]

I'm personally a bit sad that neither AMD nor nVidia decided to do a ~400mm2 die on interposer with 6GB HBM1 (contrary to popular belief there is no 4GB limit -- just difficult to fit more than 4x 1Gb HBM1 stacks on an interposer with a huge die. I have not seen and I doubt I will ever see proof that shows you can't do 6 stacks of 1Gb HBM1 on interposer) to gap fill 2016, with a xx5 revision using HBM2 upon economic availability.

 

[Arachnotronic]

My bet is the reason GP102 came so fast after GP104 is that they're reversing the situation from Maxwell. Instead of the big one taking longer to arrive, with the Ti variant shortly thereafter, I bet you nVidia will milk the Titan line for quite a while before dropping the 1080 Ti since AMD has nothing in the high end market until q1 2017 or later

I don't think there will be a 1080 Ti. Aftermarket GP104s already hit out of the box what you'd expect from a stock GP102-as-1080 Ti.

Next stop is Volta GV104, IMHO.

 

[Headfoot]

I don't think there will be a 1080 Ti. Aftermarket GP104s already hit out of the box what you'd expect from a stock GP102-as-1080 Ti.

Next stop is Volta GV104, IMHO.

Interesting prediction, Titan as the only high end chip until Volta. If AMD can't drop Vega before Volta then that might well be the case. I can't imagine AMD will be that late to the game though. And if AMD shows up with Vega I'd bet they spoil their party with a simultaneous released 1080 Ti a-la 980 Ti and Fury X

 

[Timmah!]

I don't think there will be a 1080 Ti. Aftermarket GP104s already hit out of the box what you'd expect from a stock GP102-as-1080 Ti.

Next stop is Volta GV104, IMHO.

And what would you expect from stock GP102 -as- 1080Ti? Me, at least 3200 cores, at potentially 1,7GHz boost (at that CC number, if more, than probably lower frequency). No aftermarket 1080 clocked to the max could come close to the level of performance provided by that. So i doubt there wont be 1080Ti.... well, unless GV104 is pretty much rebadged/polished GP102.

BTW, on czech server in of his recent reviews OBR says he is hearing GP102 is coming before Christmas. Make of it what you want. How many months between 980 and 980Ti btw?

 

[crisium]

I also think there's a real chance they will not release a GTX 1080 Ti if Vega doesn't spank GP104 and/or GV104 is really close.

But only to get more margins, and make Volta look relatively better. There's plenty of room for a card lower than Titan X and still easily faster than the 1080, to say otherwise ignores the present Maxwell. It would be closer to 1080 than the 980 Ti was, but that was already a very large gap. It didn't always look too big since reference 980 boosted higher than 980 Ti, but both cards reach the same max OC and it's quite a bit faster.

Nvidia is already more evenly spacing out its cards so far. 1070 is further removed from 1080. 1060 is closer to 1070 (although a 1060 Ti could still fit imho). 1080 Ti being a bit closer to 1080 wouldn't surprise me.

Say there's a 3328 Core 1080 Ti.
1070 -> 1080 = +33% Cores, +25% bandwidth, +33% effective ROPs
1080->1080Ti = +30% Cores, +50%? bandwidth, +50%? ROPs
980 -> 980 Ti = +37.5% Cores, +50%/+50%

NVIDIA Titan X has lower boost clocks, and 1080 Ti may too. But so did the 980 Ti. 980 Ti was 25% faster than 980 at launch: https://tpucdn.com/reviews/NVIDIA/GeForce_GTX_980_Ti/images/perfrel_2560.gif

It may only be slightly less than that. And if they actually give it equal boost clocks, then they could likely achieve the same 25% gap, and perhaps Pascal still loves bandwidth so maybe just having more GB/s per TFLOP will be enough to have a 25% spacing even with lower clocks. No reason not to expect a 20-25% gap at stock being rather easy to achieve.

But they won't bother if GTX 1080 remains king of the sub $1000 market long enough.