Nvidia launches the Pascal GPU-based TITAN X

[skubi2808]

Hi, i think 2x1070 SLI will be £780 free ship and more faster then this X
Regards,
https://www.overclockers.co.uk/inno3d-geforce-gtx-1070-herculez-twin-x2-8192mb-gddr5-pci-express-graphics-card-gx-068-in.html

[DanceswithUnix]

I've not researched it extensively but from a quick Google they are different, the 1080 is around 9 TFLOPS, this new Titan is around 11, I've not got the time ATM to look into if those numbers are half, single, or double TFLOPS numbers though, apologies if I've got it wrong.

That isn't different though, that is just faster.

When the original Titan came out it was a clear FP64 monster which had some clear uses compared to the top GTX of the time.

So the old GTX vs Titan was like comparing a Ford Focus and a Transit, there is stuff you can do/fit in a Transit that just isn't practical in a Focus, but sometimes the Focus will be the right tool for the job as well.

Last gen was more like Focus vs Focus ST, one is a bit quicker, but they otherwise look & feel the same because they work under the same limitations.

[Corky34]

Faster is different.

But yea you're probably right and like i said i based what i said on past iterations of Titans (probably outdated thinking), I've not looked in depth at Titans as even if i had £1.2k i wouldn't buy one.

[scaryjim]

Who's to say there won't be a 1080 Ti around October for Battlefield 1 with HBM2? ....

I'd be absolutely stunned if nvidia have managed to make a hybrid HBM2/GDDR5X memory controller, given how different the two technologies are Titan X using GDDR5X pretty much guarantees that the only nvidia cards that will be getting HBM2 in this generation will be GP100-based compute cards - and they've got so much extra hardware in for FP64 that I can't see them trying to do a gaming derivative of GP100. Clearly nvidia don't think gaming is bandwidth constrained currently - they must be pretty confident in their DCC implementation.

[LSG501]

Seriously.... $1200.... Nvidia really needs to get some competition in the high end, even if it's aimed more at small businesses that can't afford quadro and tesla parts that is still overpriced for what you're getting.

With the way 3D/Video/graphics software (what this is really aimed at in my opinion) is going it won't be long before the combined gpu memory is utilised as a single pool rather than individual pools and that makes it a much better buy to buy 2 1080s....

11 TFLOPS FP32 is single point precision which is pretty much fine for everything but hardcore mathematical/science studies where errors can make a huge difference. It will likely support double precision but it's going to be lower performance as usual, not sure on how bad though.

[CAT-THE-FIFTH]

Its going to be another E-PEEN waste of money show off product for PC Master Race(tm) people. All the websites will give it top marks and then Nvidia can use it to up price all the cards below it. In the past the cards in that price range were mocked - not anymore.

Wait a few months for the aftermarket GTX1080TI cards at least!

[watercooled]

Who's to say there won't be a 1080 Ti around October for Battlefield 1 with HBM2? The Ti usually follows 3-6 months behind the same generation's Titan anyway.

As scaryjim said the two technologies require different memory controllers - GDDR5X is used on this more expensive card so I think the chances of seeing HBM2 on a possible 1080Ti are very low. Personally I doubt GDDR5X and HBM2 variants would share a die because of the vastly different pin spacing for PCB vs interposer, but even if they somehow managed to do that, and fit that many pins on the die, a HBM2 version would take a different manufacturing route because of the need to attach it to the the interposer. And then they'd need completely different PCBs to attach that interposer too. With all that in mind I think the chances of seeing it are close to zero.

I've not researched it extensively but from a quick Google they are different, the 1080 is around 9 TFLOPS, this new Titan is around 11, I've not got the time ATM to look into if those numbers are half, single, or double TFLOPS numbers though, apologies if I've got it wrong.

Those figures are both for single precision (FP32) - just repeating what DanceswithUnix said the draw of the original Kepler Titans was their good double precision (FP64) performance - 1/3 of FP32 at the time. Since Maxwell they no longer have that to differentiate; they share the 1/32 ratio of the Geforce cards. So unless there are some driver differences beneficial to professional use (there could be, I haven't checked), they're just bigger Geforce cards architecturally.

Its going to be another E-PEEN waste of money show off product for PC Master Race(tm) people. All the websites will give it top marks and then Nvidia can use it to up price all the cards below it. In the past the cards in that price range were mocked - not anymore.

Surprising what a bit of elitism will have people spending! It would be funny if people continuing to give in to spending increasingly silly amounts of money didn't consistently drive up prices like it has been...

[DanceswithUnix]

Surprising what a bit of elitism will have people spending! It would be funny if people continuing to give in to spending increasingly silly amounts of money didn't consistently drive up prices like it has been...

Well with a bargain like $1200 I just hope they re-enabled quad sli

Faster is different.

A Ducati is faster than a Transit, right up until you need to transport a sofa

[CAT-THE-FIFTH]

Don't worry a lot of the press will say "its multi-brilliant" but "pricey" then mark it 9.8/10 instead of giving it a crap score for being stupidly expensive,and the questioning the whole point of these stupidly expensive cards which are quietly pushing up prices for all cards. I expect they will toe the line you want the best will need to pay the price - just like good old Rollo used to say.

Except the cards will look old hat in less than a year or so in most likelihood and crash in value.

Its bloody hilarious when the GTX580 was $500 and the top Nvidia card and now its $1200 for the top one. Even AMD is getting in on the action.

[Tetsuo Ryuu]

I don't really understand why Nvidia feels the need to push an even faster card when the 1080 is just barely appearing on the shelves.
Wouldn't have been smarter to first push the mobile versions and then a few months down the line go back to the desktop segment again?

[DanceswithUnix]

I don't really understand why Nvidia feels the need to push an even faster card when the 1080 is just barely appearing on the shelves.
Wouldn't have been smarter to first push the mobile versions and then a few months down the line go back to the desktop segment again?

I suspect the reason for these cards is to make the 1080 look cheap when it really isn't.

[CAT-THE-FIFTH]

I don't really understand why Nvidia feels the need to push an even faster card when the 1080 is just barely appearing on the shelves.
Wouldn't have been smarter to first push the mobile versions and then a few months down the line go back to the desktop segment again?

Cause you forget the allure of halo marketing,and them using E-PEEN marketing,etc to get people hooked into these things.

Then,they push the whole price-range up,as there are wider pricing tiers. it means on the lower pricing tiers they can get away with lower performance,meaning people will have to upgrade quicker.

It also makes the £500 to £600 cards look "better value" and so more people will buy them.

The press will say its "pricey" but make sure they cover that with a 1000 accolades of brilliance.

Its a win-win scenario.

[Tetsuo Ryuu]

I suspect the reason for these cards is to make the 1080 look cheap when it really isn't.

Cause you forget the allure of halo marketing,and them using E-PEEN marketing,etc to get people hooked into these things.

Then,they push the whole price-range up,as there are wider pricing tiers. it means on the lower pricing tiers they can get away with lower performance,meaning people will have to upgrade quicker.

It also makes the £500 to £600 cards look "better value" and so more people will buy them.

The press will say its "pricey" but make sure they cover that with a 1000 accolades of brilliance.

Its a win-win scenario.

I suppose that you both are right under that point of view.
The sad thing in this whole rush to upgrade is that all this power will never be really used, because unlike a dozen years ago, with such a frantic upgrading process, developers feel at ease being sloppier and not push the optimization as far as they could.
It seems to go arm to arm for both of the parties.
We need games that require more powers so that we can sell our cards.
We need cards that provide more power so that we can produce games with less effort.

[watercooled]

I suppose that you both are right under that point of view.
The sad thing in this whole rush to upgrade is that all this power will never be really used, because unlike a dozen years ago, with such a frantic upgrading process, developers feel at ease being sloppier and not push the optimization as far as they could.
It seems to go arm to arm for both of the parties.
We need games that require more powers so that we can sell our cards.
We need cards that provide more power so that we can produce games with less effort.

Quite. I've been saying for years I've yet to see a game that convincingly beats Crysis 3 graphics hands-down. Oh but there are plenty of games harder to run!

[Tetsuo Ryuu]

I don't want to be nostalgic (I am not that old, 30, but I've been using pcs and consoles since really really young) but things back then were better.
I am also a developer, so I always appreciated how smart devs were back then, when they didn't have the computing power, the memory, and had to make full use of every single resource they had.
Their ability really shined through all the hacks and tricks they had to come up to realize something that was in their mind against the limits.
Compared to today where they mostly rely on middleware and pre-made graphic engines, it's like their minds got numb, or just sloppy, even when the shared knowledge through internet should allow for futher improvements.

And I don't want to even start with today's business models and zero-day patches...

[shaithis]

The press will say its "pricey" but make sure they cover that with a 1000 accolades of brilliance.

The problem is, you can't expect anything else.

If they score "card A" 92% and "card B" is faster in every way......then it seems harsh to rate it lower due to price....unless of course it's a VFM score (which does seem to be lacking from many reviews these days)

Of course, if you know what you want the card to do and what you expect from it, you can weight that score yourself but a reviewer cannot. For instance, if you had a 1280x768 display the Titan X VFM would be near-zero. If you wanted it for a compute-cluster, then suddenly VFM is high. The price is highly subjective from the view-point of a purchaser and it should be left to them to decide what is too much to pay....not a reviewer who will always weight it from their perspective and then slant the review/score accordingly.