UpdateNvidia GeForce GTX Titan Z postponed again

[3DVagabond]

Newegg seems to have 52 295X2's in stock, with 35 of those being $1500 and the rest between $1530 and $1600
:sneaky:

Too expensive. They need to come down a fair bit, IMO. Gotta love the Sapphire O/C for $100 more cuz it's 1030MHz, a 1% Overclock.

$3000.00 ?!?!?! The strangest part is the people actually trying to justify it. It's not even the typical, "paying extra for the best", argument (Which makes me laugh as well). We're being told to pay extra for #2. Two times as expensive as an already over priced card that kicks it's butt. You have to love a duopoly.

 

[Magic Carpet]

Too expensive. They need to come down a fair bit, IMO. Gotta love the Sapphire O/C for $100 more cuz it's 1030MHz, a 1% Overclock.

$3000.00 ?!?!?! The strangest part is the people actually trying to justify it. It's not even the typical, "paying extra for the best", argument (Which makes me laugh as well). We're being told to pay extra for #2. Two times as expensive as an already over priced card that kicks it's butt. You have to love a duopoly.

Agreed. And 3dfx isn't coming back :/

 

[3DVagabond]

Take your pick:

Option #1:
GK210 (Titan Z was GK110) is already being built....
EVGA rep said earlier that a new Kepler card was coming soon (Not Titan Z)...

Option #2:
Things are going so well with GM204 they said "the hell with this" and just pushed Maxwell launch forward....
The water cooler kit mentioned here, could that be a similar approach as AMD chose to use with R9 295X2....?
https://www.zauba.com/import-gm204-hs-code.html


:sneaky:

Option #3:
They got caught with the 295X2's 500W TDP and didn't make a card to compete and now they have to go back to the drawing board. My bet is "The new nVidia card" is a 780 ti X2 variant. Water cooled. While AMD doing water cooled is fail to some, when nVidia does it those same people will hail it as awesome.

So are we to believe that the titan-Z never existed? They planned to release a GM200 (Too early to be a respun 210 model, isn't it?)? They must have been tryng to bait AMD with that card JHH was waving around.

 

[KaRLiToS]

This pic makes me cringe. Very frustrating to see him so confident with this silly price tag.

I bet when he announced the 2999$ MSRP, he was saying to himself in his head: "Yessss, you will all get screwed." :twisted:

 

[3DVagabond]

This pic makes me cringe. Very frustrating to see him so confident with this silly price tag.

I bet when he announced the 2999$ MSRP, he was saying to himself in his head: "Yessss, you will all get screwed." :twisted:

He seems a bit intoxicated with power to me. He thought he was looking at a 375W TDP card from his competitor and figured he could easily beat that and would charge $3000 to rub AMD's noses in it. I'm sure he figured the 295X2 would sit on the shelf while the Titan-Z sold out at twice the price. He could then text Rory (surely they have each others phone number?) all day long gloating and taunting.

What I originally thought was a marketing interfering with engineering fail, with the 2x 8pin connectors on a 500W TDP card, night well have been the best stealth marketing move in history.

 

[Cloudfire777]

Option #3:
They got caught with the 295X2's 500W TDP and didn't make a card to compete and now they have to go back to the drawing board. My bet is "The new nVidia card" is a 780 ti X2 variant. Water cooled. While AMD doing water cooled is fail to some, when nVidia does it those same people will hail it as awesome.

So are we to believe that the titan-Z never existed? They planned to release a GM200 (Too early to be a respun 210 model, isn't it?)? They must have been tryng to bait AMD with that card JHH was waving around.

This discussion is a little silly.

It all depends wether the professional or the gamer customers are the preferred choice of buyers.

For those who need CUDA, run algorithms in their workplace or software that is optimized for CUDA, R9 295X2 is a non option, which means Nvidia can charge pretty much what they want. Well there is probably a limit here and R9 295X2 might have lowered that limit some, but I don`t think anyone here knows this threshold, because 0.0001% (yes I know Im just making a number up) of Anandtech`s userbase are these users, and you don`t see them complaining.

For gaming, R9 295X2 is obviously the best choice, no doubt. But that other group, those who need the CUDA Flops....It really depends on how much Nvidia intend to sell of this chip, and how much demand there is from that group...
That Nvidia decide to wait could just be a change in the vbios to allow some higher clocks to beat R9 295X2 to market the Titan Z accordingly, and Nvidia may have run a bunch of stress tests to ensure that the die`s and cooling system and power supply can withstand this increased strain. You can`t just push out a $3k card and cross fingers that it will be ok. They gotta be sure. Could you imagine the bad publisity if pictures of burned cards along with the title "Look what happened this $3000 Nvidia card" popped up? Say bye bye to a lot of future buyers.
Of course this is just a theory, and it may have been driver issues, but we will see if we even get to see this card.

 

[3DVagabond]

This discussion is a little silly.

It all depends wether the professional or the gamer customers are the preferred choice of buyers.

For those who need CUDA, run algorithms in their workplace or software that is optimized for CUDA, R9 295X2 is a non option, which means Nvidia can charge pretty much what they want. Well there is probably a limit here and R9 295X2 might have lowered that limit some, but I don`t think anyone here knows this threshold, because 0.0001% (yes I know Im just making a number up) of Anandtech`s userbase are these users, and you don`t see them complaining.

For gaming, R9 295X2 is obviously the best choice, no doubt. But that other group, those who need the CUDA Flops....It really depends on how much Nvidia intend to sell of this chip, and how much demand there is from that group...

This is a Geforce card. It's a gaming card. It is not a Tesla card as people are trying to make it sound like. Geforce = Gaming.

 

[Cloudfire777]

This is a Geforce card. It's a gaming card. It is not a Tesla card as people are trying to make it sound like. Geforce = Gaming.

So?

Titan`s user group wasn`t just gamers.
GTX Titan, GTX Titan Black, GTX Titan Z, they all have the same FP64 compute performance you get from a Quadro/Tesla card with the GK110.
Not everyone need the ECC memory or the support that comes along with the professional cards, and since the Titan series is 1/4 (approx) of a Tesla/Quadro, it will be very tempting for those who build clusters of these cards and need massive computation power for the simulations they run.

GTX 780, GTX 780 Ti, they are different. Nvidia have disabled the majority of the FP64 cores

 

[Obsoleet]

Titan Z just doesn't have what it takes to compete. They wanted it to be a $3000 card intended for the "professional market", yes, sure. But missing the Quadro branding and with a Geforce sticker on it.

 

[iiiankiii]

So?

Titan`s user group wasn`t just gamers.
GTX Titan, GTX Titan Black, GTX Titan Z, they all have the same FP64 compute performance you get from a Quadro/Tesla card with the GK110.
Not everyone need the ECC memory or the support that comes along with the professional cards, and since the Titan series is 1/4 (approx) of a Tesla/Quadro, it will be very tempting for those who build clusters of these cards and need massive computation power for the simulations they run.

GTX 780, GTX 780 Ti, they are different. Nvidia have disabled the majority of the FP64 cores

I don't know. If the intended audience are professionals, that support that comes along with it is pretty important. Since this is not a tesla card, there won't be any support if things goes wrong. If you're doing research, you can't afford to have errors and failures, that would invalidate everything. That ECC memory, along with the Tesla driver support, are key components to a professional grade card.

 

[Cloudfire777]

I don't know. If the intended audience are professionals, that support that comes along with it is pretty important. Since this is not a tesla card, there won't be any support if things goes wrong. If you're doing research, you can't afford to have errors and failures, that would invalidate everything. That ECC memory, along with the Tesla driver support, are key components to a professional grade card.

I guess it depends on what they are running.

Important data harvested from a ton of code, you are probably right. Any incorrect data there and it might make the result invalid. Tons of data wasted.
Graphic/3D modelling, the simulations might skip the errors and proceed.

I don`t know. Im just saying don`t write Titan Z off just because R9 295X2 can beat it in price/performance in gaming.
We will see what happens with Titan Z. The best of course would be a price reduction down to $1.5-2k to also cater for gamers. But atleast gamers have options, like SLI.

If price reduction do happen, well there is our answer, gaming was their priority and we should thank AMD for pushing them.
If the card is removed from the market alltogether, one can`t do much but wonder if AMD gave their plan a massive blow.

 

[VulgarDisplay]

I guess it depends on what they are running.

Important data harvested from a ton of code, you are probably right. Any incorrect data there and it might make the result invalid. Tons of data wasted.
Graphic/3D modelling, the simulations might skip the errors and proceed.

I don`t know. Im just saying don`t write Titan Z off just because R9 295X2 can beat it in price/performance in gaming.
We will see what happens with Titan Z. The best of course would be a price reduction down to $1.5-2k to also cater for gamers. But atleast gamers have options, like SLI.

If price reduction do happen, well there is our answer, gaming was their priority and we should thank AMD for pushing them.
If the card is removed from the market alltogether, one can`t do much but wonder if AMD gave their plan a massive blow.

Nevermind price/performance, the titan z can't even beat the 295x2 in performance regardless of price. That is the only reason it has been scrapped. Their best just wasn't good enough.

 

[Silverforce11]

So?

Titan`s user group wasn`t just gamers.
GTX Titan, GTX Titan Black, GTX Titan Z, they all have the same FP64 compute performance you get from a Quadro/Tesla card with the GK110.
Not everyone need the ECC memory or the support that comes along with the professional cards, and since the Titan series is 1/4 (approx) of a Tesla/Quadro, it will be very tempting for those who build clusters of these cards and need massive computation power for the simulations they run.

GTX 780, GTX 780 Ti, they are different. Nvidia have disabled the majority of the FP64 cores

All your argument is based on is what the potential users will maybe use it for, this differs from what NV is selling it as.

1. It's using Geforce drivers, not Quadro.
2. On NV's own site, they boast about its superior gaming performance.
3. On the listing on Amazon, they clearly boast its gaming performance, no mention of anything else.

This is why we consider it a fail at $2,999 (or was it $3,999 listed? lol).

If it had Quadro driver support, ECC memory, then $2,999 would be justified, and any gamer who buys it will be clearly shown not to care about money.

This is why its DoA and probably canceled. They just can't expect much out of ONE-fan air cooling if they are planning to clock it higher to match the R295X2. It would be a laughing stock to release a halo card thats slower and twice as expensive.

Given in 6 months we'll probably get GM204, all of this is moot and its actually a good decision to scrap Titan Z.

 

[dn7309]

So?

Titan`s user group wasn`t just gamers.
GTX Titan, GTX Titan Black, GTX Titan Z, they all have the same FP64 compute performance you get from a Quadro/Tesla card with the GK110.
Not everyone need the ECC memory or the support that comes along with the professional cards, and since the Titan series is 1/4 (approx) of a Tesla/Quadro, it will be very tempting for those who build clusters of these cards and need massive computation power for the simulations they run.

GTX 780, GTX 780 Ti, they are different. Nvidia have disabled the majority of the FP64 cores

So what you're saying that the Titan is basically the Quadro or Tesla and there is no reason to spend more on Quadro or Tesla.

 

[chimaxi83]

So?

Titan`s user group wasn`t just gamers.
GTX Titan, GTX Titan Black, GTX Titan Z, they all have the same FP64 compute performance you get from a Quadro/Tesla card with the GK110.
Not everyone need the ECC memory or the support that comes along with the professional cards, and since the Titan series is 1/4 (approx) of a Tesla/Quadro, it will be very tempting for those who build clusters of these cards and need massive computation power for the simulations they run.

GTX 780, GTX 780 Ti, they are different. Nvidia have disabled the majority of the FP64 cores

Irrelevant. Nvidia markets their Titan Z as a gamer card. The failed justification attempts are mind bottling.

 

[Lonyo]

Nvidia sells the GK110 single GPU based K6000 for $5000. They aren't going to release a product for $3000 which outperforms it for workstation purposes.

 

[RussianSensation]

So?

Titan`s user group wasn`t just gamers.
GTX Titan, GTX Titan Black, GTX Titan Z, they all have the same FP64 compute performance you get from a Quadro/Tesla card with the GK110.
Not everyone need the ECC memory or the support that comes along with the professional cards, and since the Titan series is 1/4 (approx) of a Tesla/Quadro, it will be very tempting for those who build clusters of these cards and need massive computation power for the simulations they run.

Right, which means this card has no real world relevance to anyone but perhaps 1 person out of 10,000 on these forums. Why? For $1,000 more, you could put 4 Titan Blacks into an X79 motherboard. Therefore, 4 Titan Blacks > 2 Titan Zs for less $.

Users who are running heavy compute with CUDA cores aren't going to stuff a $3,000 Titan Z in some small case because 2 Titan Blacks would fit and cost $1,000 less. Besides, you could fit 2x 295X2s for gaming inside a Corsair Air 540, but this is impossible with 2 Titan Zs. For gaming or compute in small cases, Titan Z fails again against Titan Blacks or 295X2s.

That leaves a group of buyers who will be buying at least 3 Titan Zs for their system or spending $9,000 USD.

NV should just release a water-cooled GTX790 with 780Tis and move on from the Titan Z PR disaster.

 

[Stormflux]

As someone who has a heavy interest in the GPGPU capabilities for 3D Graphics, the Titan-Z is simply and plainly over priced. They got caught with their pants down in the wake of the 295x2. But I'm not one to go out and buy a Quadro 6000 12gb for $5000. I'm not doing engineering and critical compute that would require ECC ram so Quadro's are overpriced in general for my purposes.

It is unfortunate there is not a lot of quality support for OpenCL and CUDA is the main attraction in the major 3d software choices, cause moves like the Titan-Z are showing arrogance on nVidia's end. According to some, the poor adoption of this is indeed down to AMD Drivers and their compilers.

I'd rather buy 3 Titan-Blacks for the price of one Titan-Z. I'd arguably get 1.5x the performance for the same price. But if OpenCL was up to snuff in more software, I'd rather buy 5 290x's...

 

[KaRLiToS]

As someone who has a heavy interest in the GPGPU capabilities for 3D Graphics, the Titan-Z is simply and plainly over priced. They got caught with their pants down in the wake of the 295x2. But I'm not one to go out and buy a Quadro 6000 12gb for $5000. I'm not doing engineering and critical compute that would require ECC ram so Quadro's are overpriced in general for my purposes.

It is unfortunate there is not a lot of quality support for OpenCL and CUDA is the main attraction in the major 3d software choices, cause moves like the Titan-Z are showing arrogance on nVidia's end. According to some, the poor adoption of this is indeed down to AMD Drivers and their compilers.

I'd rather buy 3 Titan-Blacks for the price of one Titan-Z. I'd arguably get 1.5x the performance for the same price. But if OpenCL was up to snuff in more software, I'd rather buy 5 290x's...

In order words, there is no excuse for a product such as the Titan Z and its price.

 

[parvadomus]

In order words, there is no excuse for a product such as the Titan Z and its price.

ofc there is no excuse. I would never buy Titan for DP, and even less for dying stuff like CUDA. Intel is just destroying Nvidia @ HPC and with that its propietary things. I predict 2 years for CUDA to die.

 

[cytg111]

What I originally thought was a marketing interfering with engineering fail, with the 2x 8pin connectors on a 500W TDP card, night well have been the best stealth marketing move in history.

Its been a while since we've seen amd pull a fast one on its competitors.. nice to see its still alive.

 

[3DVagabond]

So?

Titan`s user group wasn`t just gamers.
GTX Titan, GTX Titan Black, GTX Titan Z, they all have the same FP64 compute performance you get from a Quadro/Tesla card with the GK110.
Not everyone need the ECC memory or the support that comes along with the professional cards, and since the Titan series is 1/4 (approx) of a Tesla/Quadro, it will be very tempting for those who build clusters of these cards and need massive computation power for the simulations they run.

GTX 780, GTX 780 Ti, they are different. Nvidia have disabled the majority of the FP64 cores

So the Titan-Z user group is researchers that are on a tight budget and can't afford proper Tegra cards and support, don't require accurate calculations, whose workflow requires CUDA but doesn't benefit from Tegra drivers, who are savvy enough to not require any support from nVidia for their software stack, and whose tight budget makes it worthwhile to spend 50% more, an extra $1000, for the same performance as 2x Titans. Even if this user group does exist, which I can't imagine it would, how small would it be? Would it be worthwhile designing a card like the Geforce Titan-Z for?

How would it's performance vs. the 295X2, a gaming card, delay it's release. Changing it from, "It's being released tomorrow." To, "We'll have it out next week." To, We would like to insure our investors that we will be releasing it sometime in the next 3 months." It's supposedly a poor man's compute card. What difference does it make how well it performs gaming?

The answer is obvious, it's not a poor mans compute card. It's a gaming card. They miscalculated and thought it would be faster than the 295X2. Even if it was only 10%-15% faster they felt they could charge their faithful $3000 and they would buy it. I really hope they've overdone it this time and people say, hell no! We'll see, assuming they ever get it released at all.

 

[3DVagabond]

NV should just release a water-cooled GTX790 with 780Tis and move on from the Titan Z PR disaster.

Except for the egg on their corporate faces (one or two people might have to fall on their swords for the company to save face) this is the only outcome that makes sense. I'm not sure how they reconcile that with the recent promise to investors, but I suppose that won't matter if the "790 ti" drops and is a success.

 

[Cookie Monster]

ofc there is no excuse. I would never buy Titan for DP, and even less for dying stuff like CUDA. Intel is just destroying Nvidia @ HPC and with that its propietary things. I predict 2 years for CUDA to die.

Think we've been hearing this one for the last 4 years. CUDA won't disappear anytime soon because its much more robust than its nearest competitors. The Intel Phi is a threat for sure but its perf/w is laughable.

Plus, its not the theoretical maximum FLOPs that matters but how you achieve it. There is a reason why AMD based cards dont jive well with GPGPU programmers outside a few applications. I mean, what is luxmark and how does this relate to the real world apps? Its also based on OpenCL which nVIDIA gives minimal support/performance optimizations.

If everything was based on pure theoretical numbers alone, the Tesla/CUDA/Quadro market would cease to exist. Except in the real world, this is far from the truth.

However the 3K price tag is a little too extreme.

 

[parvadomus]

Think we've been hearing this one for the last 4 years. CUDA won't disappear anytime soon because its much more robust than its nearest competitors. The Intel Phi is a threat for sure but its perf/w is laughable.

Plus, its not the theoretical maximum FLOPs that matters but how you achieve it. There is a reason why AMD based cards dont jive well with GPGPU programmers outside a few applications. I mean, what is luxmark and how does this relate to the real world apps? Its also based on OpenCL which nVIDIA gives minimal support/performance optimizations.

If everything was based on pure theoretical numbers alone, the Tesla/CUDA/Quadro market would cease to exist. Except in the real world, this is far from the truth.

However the 3K price tag is a little too extreme.

4 years ago there was almost no competition. Now we are talking about intel, a much bigger competitor which will kill nv HPC marketshare sooner or later like it did to AMD in servers.