NVIDIA Announces the GeForce GTX TITAN X Pascal

[efikkan]

GP100 isn't faster than GP102 for rendering, since GP102 is essentially GP100 with FP64, NVLink and HBM removed. A graphics card with GP100 will just use ~300W instead of 250W with no benefit for rendering. I'm very happy Nvidia made non-compute version of GP100, this way we can get graphics cards with better efficiency and availability.

Huh? So you are saying the "ti" will use 50 more watts but no additional rendering "benefit"?

I'm saying GP100 will use ~50W or so performing the same as GP100, or perhaps even more. The amount of CUDA cores are the same, but GP100 has a lot of extra features which have no extra benefits for rendering.
Just simply look at the Tesla P100 (GP100), 1328 MHz, 300W TDP.

R9 290X was faster than the GTX TITAN and the GTX 780, forcing NVIDIA to launch the GTX 780 Ti with the full 2,880 CUDA cores of the GK110 silicon.

R9 290X, GTX 780 and GTX Titan were all performing within a 2% margin, so roughly the same. Even in OC mode R9 290X didn't displace GTX 780. GTX 780 Ti and Titan Black was released because the yields of GK110 improved a lot, and R9 290X didn't even come close to the 10% advantage of GTX 780 Ti.

It was launched in response to the R9 Fury X, AMD had to readjust its pricing down to $650. Their performance was within 1% of each other, so NVIDIA designed the GTX 980 Ti as a response to the R9 Fury X. NVIDIA sacrificed the GTX TITAN X and GTX 980 to make GTX 980 Ti successful.

You once again get the facts wrong. GTX 980 Ti was released before R9 Fury X, so it's impossible that GTX 980 Ti was a response to R9 Fury X. We know that GTX 980 Ti was planned several months ahead, and we know it was sent to testing a couple of months before release.

If yours made sense, TITAN X Pascal would have 3,840 CUDA cores.

That have no relation to anything I said.

NVIDIA doesn't want GTX 1080 to suffer the fate of the GTX 980 (market irrelevance, unsold Lightnings and AMP Extremes). There won't be a GTX 1080 Ti till Vega is on the horizon.

GTX 980 was "irrelevant" because it only performed 12% better than GTX 970(because GTX 970 was "too good"), leaving GTX 970 and GTX 980Ti as the only sensible choices in the upper segment. GTX 1080 is way better positioned vs GTX 1070, so that's not the case any more. If Nvidia releases a "GTX 1080 Ti" performing ~30% or so over GTX 1080, Nvidia actually have a perfect scale ranging from GTX 1060 to "GTX 1080 Ti" with nice increments.

I've still seen no confirmation that "GTX 1080 Ti" is coming anytime soon, so it's probably three or more months away, if it's coming at all. But I do see two problems; (I don't expect you to answer, these are just general questions)
* Titan X offered more memory over GTX 980 Ti, which matters to it's target semi-professional customers doing CUDA and professional graphics. A "GTX 1080 Ti" will obviously not have just 6 GB, so what will be the configuration?
* I would argue that a "GTX 1080 Ti" should rather be called "GTX 1090", to position it better, granted there are no dual-GPU products scheduled to use this name.

 

[Captain_Tom]

I'm saying GP100 will use ~50W or so performing the same as GP100, or perhaps even more. The amount of CUDA cores are the same, but GP100 has a lot of extra features which have no extra benefits for rendering.
Just simply look at the Tesla P100 (GP100), 1328 MHz, 300W TDP.


R9 290X, GTX 780 and GTX Titan were all performing within a 2% margin, so roughly the same. Even in OC mode R9 290X didn't displace GTX 780. GTX 780 Ti and Titan Black was released because the yields of GK110 improved a lot, and R9 290X didn't even come close to the 10% advantage of GTX 780 Ti.


You once again get the facts wrong. GTX 980 Ti was released before R9 Fury X, so it's impossible that GTX 980 Ti was a response to R9 Fury X. We know that GTX 980 Ti was planned several months ahead, and we know it was sent to testing a couple of months before release.


That have no relation to anything I said.


GTX 980 was "irrelevant" because it only performed 12% better than GTX 970, leaving GTX 970 and GTX 980Ti as the only sensible choices in the upper segment. GTX 1080 is way better positioned vs GTX 1070, so that's not the case any more. If Nvidia releases a "GTX 1080 Ti" performing ~30% or so over GTX 1080, Nvidia actually have a perfect scale ranging from GTX 1060 to "GTX 1080 Ti" with nice increments.

I've still seen no confirmation that "GTX 1080 Ti" is coming anytime soon, so it's probably three or more months away, if it's coming at all. But I do see two problems; (I don't expect you to answer, these are just general questions)
* Titan X offered more memory over GTX 980 Ti, which matters to it's target semi-professional customers doing CUDA and professional graphics. A "GTX 1080 Ti" will obviously not have just 6 GB, so what will be the configuration?
* I would argue that a "GTX 1080 Ti" should rather be called "GTX 1090", to position it better, granted there are no dual-GPU products scheduled to use this name.

Ok I need to call you on your BS. The 290X soundly defeated the Titan by 10% so just stop right there. Then the 780 Ti launched in response with only 7% more performance on average. Within a few months the 780 Ti had already begun performing worse in most of the latest games (A 7970 comes close these days more often then you probably realize).

Sure the 980 Ti came out before the Fury X, but stop playing dumb - it isn't crazy to believe that they saw the Fury X coming just like the 290X and decided to pre-empt it with the 980 Ti once they realized it would roughly match the Titan X (Also note that in DX12 games the Fury X destroys the TItan X).


Now let's move on to Nvidia's latest prank - the (new?!) Titan X. This thing (Based on their own numbers) will be a laughably small 25% stronger than the 1080 for double the price and only 50% more VRAM instead of the usual 300% increase in VRAM. It is very clear imo that Nvidia knows next spring will be VERY rough for them. All of the latest games will support DX12 and AMD will have a 4600 - 6000 SP Vega card out with HBM2. It will be a bloodbath.

So what would you do if you were Nvidia? Launch the ultra high-end ASAP no matter what so you can grab people's money in the short 6-month time span they will give it to you. Even with their early yields they can't sell them soon enough. Paper launches of paper launches is something I have never seen before lol. And then they even went as far as to nix the HBM2 on the Titan so they could launch it 4 months sooner (Even if it cost them 4GB and 20% performance).

Lastly, what will they call the GP100 gaming card? Easy, the 1180! Let AMD have some sales in January - March and then launch the HBM2/3840/1800 MHz 1180! Who cares if it screws over Titan buyers? That's what always happens to them anyways....

 

[MxPhenom 216]

Ok I need to call you on your BS. The 290X soundly defeated the Titan by 10% so just stop right there. Then the 780 Ti launched in response with only 7% more performance on average. Within a few months the 780 Ti had already begun performing worse in most of the latest games (A 7970 comes close these days more often then you probably realize).

Sure the 980 Ti came out before the Fury X, but stop playing dumb - it isn't crazy to believe that they saw the Fury X coming just like the 290X and decided to pre-empt it with the 980 Ti once they realized it would roughly match the Titan X (Also note that in DX12 games the Fury X destroys the TItan X).


Now let's move on to Nvidia's latest prank - the (new?!) Titan X. This thing (Based on their own numbers) will be a laughably small 25% stronger than the 1080 for double the price and only 50% more VRAM instead of the usual 300% increase in VRAM. It is very clear imo that Nvidia knows next spring will be VERY rough for them. All of the latest games will support DX12 and AMD will have a 4600 - 6000 SP Vega card out with HBM2. It will be a bloodbath.

So what would you do if you were Nvidia? Launch the ultra high-end ASAP no matter what so you can grab people's money in the short 6-month time span they will give it to you. Even with their early yields they can't sell them soon enough. Paper launches of paper launches is something I have never seen before lol. And then they even went as far as to nix the HBM2 on the Titan so they could launch it 4 months sooner (Even if it cost them 4GB and 20% performance).

Lastly, what will they call the GP100 gaming card? Easy, the 1180! Let AMD have some sales in January - March and then launch the HBM2/3840/1800 MHz 1180! Who cares if it screws over Titan buyers? That's what always happens to them anyways....

Not really.

 

[efikkan]

Ok I need to call you on your BS. The 290X soundly defeated the Titan by 10% so just stop right there. Then the 780 Ti launched in response with only 7% more performance on average. Within a few months the 780 Ti had already begun performing worse in most of the latest games (A 7970 comes close these days more often then you probably realize).

You obviously have a hard time reading numbers, R9 290X didn't defeat the Titan(the first one), they performed the same. GTX 780 Ti and Titan Black performed significantly better than those, without any competition from AMD at all!

Sure the 980 Ti came out before the Fury X, but stop playing dumb - it isn't crazy to believe that they saw the Fury X coming just like the 290X and decided to pre-empt it with the 980 Ti once they realized it would roughly match the Titan X

Graphics cards' details are finalized months in advance, and the release window is decided almost one year in advance. There is only room for minor adjustments near the release (like tweaking it a few MHz). Anyone who know anything about GPU development knows this, and it defeats your whole argument.

Also note that in DX12 games the Fury X destroys the TItan X

Only in your fantasy land. AMD doesn't crush Nvidia in unbiased benchmarks, they scale pretty much the same, exactly as any competent person expected.

Lastly, what will they call the GP100 gaming card? Easy, the 1180! Let AMD have some sales in January - March and then launch the HBM2/3840/1800 MHz 1180! Who cares if it screws over Titan buyers? That's what always happens to them anyways....

Once again, what would be the purpose of releasing GP100 as a consumer card? It would be a waste of a compute/datacenter GPU with no real benefit for end users. We would be better off if Pascal-refresh will offer a full GP102 instead.

 

[Primalz]

This new titan X will cost $2400 - $2500 here in Australia if it's twice the price of the gtx 1080. The US price for 1080's is what about $600US. Nvidia says this new card is $1,199US. I could build a kick ass gaming pc for $2400, that could have sli'd 1070's & i5 6600K for about that price. I don't think people here in Australia would fork out $2.5K on a gpu.

 

[the54thvoid]

The price is silly, no doubts but people will but it. There are plenty of folk with comfortable incomes that don't have a problem spending that sort of cash.
Others choose not to or cannot afford to. But business is business and Nvidia wouldn't release it if it wouldn't sell.
Many people want the fastest, many will buy it. In a fantasy land, if this was sold at cost, it would probably take AMD out in one generation.
Who doesn't want the fastest? GTX1080 already beats AMD in all titles, even DX12 AMD sponsored. This will be more so.
Vega can't come soon enough but even then, Pascal has proven worthy.

 

[newconroer]

Confused why they would even go with a Titan that isn't utilizing Pascal's full capabilities - especially since the TI (or so we think) will launch with HBM2.

So we are paying a few hundred more for a larger bus with this?

 

[BiggieShady]

... especially since the TI (or so we think) will launch with HBM2.

Wishful thinking ... .... ... unless ... unless there will be also full enabled chip Titan H version with HBM2 skewing all prices back after this initial cashgrab

 

[GC_PaNzerFIN]

Confused why they would even go with a Titan that isn't utilizing Pascal's full capabilities - especially since the TI (or so we think) will launch with HBM2.

So we are paying a few hundred more for a larger bus with this?

No it does not have HBM2 unless Ti is released more than a year from now. All the very limited HBM2 stock goes to GP100 and GP102 has G5X memory controllers...

 

[dont whant to set it"']

So it boils down to a crippled chip, way overpriced? allmoast applelike. It kinda looks like the green camp is doing another "reach-around".

 

[TheHunter]

I already have one. Bitch please











Jk.

 

[Slizzo]

No it does not have HBM2 unless Ti is released more than a year from now. All the very limited HBM2 stock goes to GP100 and GP102 has G5X memory controllers...

Sigh...

Only the very limited stock from HYNIX goes to AMD. nVidia has been sourcing their HBM2 from Samsung.

 

[Tsukiyomi91]

Titan X Pascal is touted as the most powerful single GPU? I will wait for the upcoming wallet-friendly GTX1080Ti & see whether the refreshed Titan X can hold on to it's crown or not. Titan X Pascal is not to be confused with the current Titan X (GM200) FYI folks. For HBM v2, it's most likely that 2nd Gen Pascal will be utilizing it, which won't be unveiled until mid-Spring 2017 or Summer 2017, which ever comes first. BTW, the differences between Hynix & Samsung is very small since both companies are living under one roof IMO.

 

[Prima.Vera]

Anyways, I don't think nVidia is going to release the 1080Ti this year, or that it would ever do that. Not unless AMD will bring something to challenge the 1080...

 

[RejZoR]

You obviously have a hard time reading numbers, R9 290X didn't defeat the Titan(the first one), they performed the same. GTX 780 Ti and Titan Black performed significantly better than those, without any competition from AMD at all!


Graphics cards' details are finalized months in advance, and the release window is decided almost one year in advance. There is only room for minor adjustments near the release (like tweaking it a few MHz). Anyone who know anything about GPU development knows this, and it defeats your whole argument.


Only in your fantasy land. AMD doesn't crush Nvidia in unbiased benchmarks, they scale pretty much the same, exactly as any competent person expected.


Once again, what would be the purpose of releasing GP100 as a consumer card? It would be a waste of a compute/datacenter GPU with no real benefit for end users. We would be better off if Pascal-refresh will offer a full GP102 instead.

Love it how everything is "unbiased" when NVIDIA is winning, but when AMD humiliates NVIDIA in equally unbiased DX12 and Vulkan titles with 1 generation old flagship, it's somehow "biased". Hilarious. Nothing is biased, AMD is just better in DX12 and Vulkan and that is a fact as solid and real as gravity itself. And it has been known for as long as Mantle existed and for as long as AMD was pushing for this closer to the metal API system. And it's GeForce owner telling you all this. So, do your math.

 

[the54thvoid]

AMD is just better in DX12 and Vulkan and that is a fact as solid and real as gravity itself

That's not strictly true. AMD is generally better in DX12 than it is in DX11. If you compare the core and transistor counts of GP104 and the full Fiji chip (node size made irrelevant due to comparison of hardware components) the GTX1080 is faster in DX12 and even Doom Vulcan.
The 'AMD is better at DX12' chant is only as solid as gravity when comparing it to itself in DX11.

 

[RejZoR]

That's not strictly true. AMD is generally better in DX12 than it is in DX11. If you compare the core and transistor counts of GP104 and the full Fiji chip (node size made irrelevant due to comparison of hardware components) the GTX1080 is faster in DX12 and even Doom Vulcan.
The 'AMD is better at DX12' chant is only as solid as gravity when comparing it to itself in DX11.

You can't compare NVIDIA's very latest generation with AMD's last generation (because they haven't actually released the new one yet). But if you compare R9 Fury X with GTX 980 or even GTX 980Ti, the story is different.

 

[Ascalaphus]

I might get 2 of them.

 

[the54thvoid]

You can't compare NVIDIA's very latest generation with AMD's last generation (because they haven't actually released the new one yet). But if you compare R9 Fury X with GTX 980 or even GTX 980Ti, the story is different.

To refute your point I can. DX12 has simply given AMD the performance it should provide for the hardware on the chip. AMD should provide performance very close to Nvidia when it has more transistor counts with higher shader counts. It's just that in DX11, that hardware isn't all being used very effectively but in DX12 it is. Nvidia's hardware is used optimally in DX11 and DX12 - it wont produce a much higher performance.

As for generational stuff, okay - the RX480 has 29.5% more transistors and 80% more shaders. The GTX1060 has 50% more ROPS. In Doom Vulkan we know the RX480 is far better but in Ashes they are very equal. So hardware versus hardware, RX480 should be all over the 1060 - but it's not. So no, DX12 doesn't make AMD better, necessarily, than Nvidia. And again, it very much depends on who is developing the game.

 

[jihadjoe]

True, but who's AMD going to compete with first?
Because this is where AMD stands right now: View attachment 77172
And if you think about the mobile market, Qualcomm is worth almost $100Bn... I don't know who would want to be in AMD's place right now.

The old handset division they sold to Qualcomm a couple of years ago could probably buy them out now.

 

[Vayra86]

Lastly, what will they call the GP100 gaming card? Easy, the 1180! Let AMD have some sales in January - March and then launch the HBM2/3840/1800 MHz 1180! Who cares if it screws over Titan buyers? That's what always happens to them anyways....

If you really even think that the '1180' will be out with HBM2 on it, you've lost one customer here. Not gonna happen. Nvidia's surely not saying 'HBM2 is expensive, so let's drop it in the card for the larger market at lower price/margins where it doesn't even have any point - go quickly bois!' Besides, it is more than clear that GDDR5X is here to stay for a while and has plenty of headroom to eat into. I think we can even question the existence of a non-top-end HBM2 card in 2017. Maybe AMD will do a Fury V2 but I strongly doubt it, and Nvidia wants to sell the GP100 with HBM2, not the measly 102.

All Nvidia did with Pascal was a huge clock bump and Pascal is worse for it clock-for-clock compared to Maxwell of Kepler. All AMD did with GCN was optimize and update a little, while taking full advantage of the smaller node. GCN essentially hasn't changed much, it was already optimised for the API the way it always was, it's just the API that is now changing towards that architecture. None of this makes either camp have a bad GPU at the moment or a bad architecture. In fact I would argue they both have the architecture that fits the business strategy and it differs in the same way too. Evidenced by Nvidia's 'async workarounds' that perform just as well as AMDs native and compliant Async. Even the perf/watt of Polaris is on par or better than Pascal's in some API's.

 

[R-T-B]

I'll probably buy one... albeit second hand, and when it falls below $400.00, just like my present Titan X.

That's a upgrade scheme I can get behind. I like big chips.

 

[thesmokingman]

Whatever the price is, anyone who have the money and passion for fastest graphics should buy this... buy it fast and stay on the top... until the next 1080Ti comes and then buy that to stay on top.

It's rough being at the top.

 

[jigar2speed]

And you know for a fact HBM2 is needed here? Or do you just like shiny new things?

$1200 and no shiny ?? WTF is going on here.

 

[Prima.Vera]

For this price, the card should come with 1 year free pass to the local whore house or something....