Nvidia GTX Titan Z

[Bokishi]

$3,000 GPU Holy Hell! 12gb vram is cool though

 

http://www.pcgamer.com/2014/03/25/nvidia-announce-geforce-gtx-titan-z-brings-12gb-vram-for-3000/

[Humanoid]

Two Titans for the price of three, that's genius.

 

(Hell, for the purposes of most users, it's two 780Tis for the price of five....)

[Keyrock]

I don't see any situation where this card makes any sense to buy.  I mean, I suppose if someone only had one PCI-E lane for gfx and needed that amount of power, but, realistically, anyone that needs that kind of power has a motherboard with at least 2 lanes.  The only advantage this card gives is the 12 GB of vRAM over 6 GB you'd get from a SLI setup, but not even a 5k setup would effectively use 12 GB of vRAM.  In a 5k setup the extra 6 GB of vRAM might give a slight benefit, but certainly not $1000 worth.  The other scenario is if you want to SLI two of these monstrosities together, but there is nothing out there that requires that kind of power, nor will there be anytime soon.  This card is for people with money to burn that want bigger numbers on benchmarks because pen1s size, or whatever.

[AwesomeOcelot]

Wait for the benchmarks. They say it works better than SLI, and that it's quieter and more compact. This is a card for people with money to burn, that don't much care whether it's $1000 or $3000, they'll pay a premium for small benefits. People are running Titans in SLI so Nvidia know there's people willing to buy it, I can't imagine many people though. This is probably akin to a showpiece sports car.

 

I never drop much money on a GPU because they advance so fast it's better to keep upgrading. SSD, HDD, and RAM change in price, CPUs are more steady in price, performance is increased but not nearly to as much benefit as the different generations of GPU. I don't see how the Titan Z is better than getting 3 generations of Titan at $1000 each.

Edited March 31, 2014 by AwesomeOcelot

[Keyrock]

I don't see how the Titan Z is better than getting 3 generations of Titan at $1000 each.

The only way it could be is in a scenario where the extra 6 GB of vRAM is more beneficial than the extra two thousand whatever cores.  I don't think such a scenario exists, even taking into account that SLI doesn't scale quite as well from 2 to 3 as it does from 1 to 2.

[Humanoid]

I don't see any situation where this card makes any sense to buy.  I mean, I suppose if someone only had one PCI-E lane for gfx and needed that amount of power, but, realistically, anyone that needs that kind of power has a motherboard with at least 2 lanes.  The only advantage this card gives is the 12 GB of vRAM over 6 GB you'd get from a SLI setup, but not even a 5k setup would effectively use 12 GB of vRAM.  In a 5k setup the extra 6 GB of vRAM might give a slight benefit, but certainly not $1000 worth.  The other scenario is if you want to SLI two of these monstrosities together, but there is nothing out there that requires that kind of power, nor will there be anytime soon.  This card is for people with money to burn that want bigger numbers on benchmarks because pen1s size, or whatever.

Couple things: It having 12GB RAM listed means 6GB per GPU, which is the same as regular Titan (Black), so no advantage there. Second thing is that a PCI-E lane is a specific thing, different from a PCI-E slot. A full PCI-E slot has 16 lanes, hence the 16X terminology, though sometimes these slots only have 8 or even 4 lanes dedicated to them. The small PCI-E slots - the ones about an inch long - are single lane slots.

 

Anyway, the point is that these cards are for people with an insufficient number of slots, yes, either wanting two Titans on an ITX board, or four on a mATX board. Thing is though, the large cooler means it probably won't fit in most ITX cases. And mATX, eh, you could upgrade your board and case instead for a fraction of the 2x Titan Z cost.

[LadyCrimson]

Even if I was rich, I'd have no desire to pay $3000 for a personal-computer GPU. I'd rather build a 2nd puter for $2000 with a $1000 GPU in it, instead.

[ManifestedISO]

Wait, so it's basically pointless on standard ATX ... and will barely upon barely fit inside micro ATX ... even less can I see the point of this card. Maybe a living room solution, but bragging rights are hard to come by with no one else but the dog around. I would be embarrassed 3000% to show off this card to any tech-minded friends.      

[Keyrock]

 

I don't see any situation where this card makes any sense to buy.  I mean, I suppose if someone only had one PCI-E lane for gfx and needed that amount of power, but, realistically, anyone that needs that kind of power has a motherboard with at least 2 lanes.  The only advantage this card gives is the 12 GB of vRAM over 6 GB you'd get from a SLI setup, but not even a 5k setup would effectively use 12 GB of vRAM.  In a 5k setup the extra 6 GB of vRAM might give a slight benefit, but certainly not $1000 worth.  The other scenario is if you want to SLI two of these monstrosities together, but there is nothing out there that requires that kind of power, nor will there be anytime soon.  This card is for people with money to burn that want bigger numbers on benchmarks because pen1s size, or whatever.

Couple things: It having 12GB RAM listed means 6GB per GPU, which is the same as regular Titan (Black), so no advantage there. Second thing is that a PCI-E lane is a specific thing, different from a PCI-E slot. A full PCI-E slot has 16 lanes, hence the 16X terminology, though sometimes these slots only have 8 or even 4 lanes dedicated to them. The small PCI-E slots - the ones about an inch long - are single lane slots.

 

 

Yes, I know the difference between lanes and slots, it was a brain fart on my part.   

 

The advantage of vRAM is still there.  When you SLI (or crossfire, for that matter) a pair (or more) of video cards, you only get access to the memory of one card, you don't get the memory of both cards (or all 3 cards in a 3-way configuration).  So if you SLI 2 Titan Black cards together you still wind up with 6 GB of usable vRAM.  Whatever black magic they're working with this TItan Z will make all 12 GB of vRAM accessible, at least that's how I read it.  That said, 6 GB of accessible vRAM is still not worth the extra $1000.

Edited April 2, 2014 by Keyrock

[Humanoid]

It's not possible. Each GPU has its own frame buffer, it's the only way multi-GPU can be handled with today's technology. It is literally two 6GB Titans, including the memory restrictions: no black magic going on here, and certainly no memory advantage.

 

I read a claim that it may be possible for compute tasks to utilise the memory in a different way, but I know nothing about that field so I can't really comment on the truth of it - but given it's no longer operating under SLI as such under those conditions, it makes sense. Perhaps that is the source of the confusion (or it may be just some people reading too much into the 12GB headline figure).

 

 

EDIT: Confirmation on that last thing by an EVGA rep. Which makes total sense of course, 2x regular Titans with 6GB each is also effectively 12GB RAM for compute purposes. Having the RAM chips on the same PCB makes absolutely no difference. This is two Titans on one PCB for all intents and purposes, whether for graphics or compute.

Edited April 2, 2014 by Humanoid

[mkreku]

Behold true power:

 

http://www.eurogamer.net/articles/digitalfoundry-2014-radeon-r9-295x2-review

 

The Radeon R9 295X, 11.5 teraflops of awesome.

 

(half the price of the Titan Z too)

[Janmanden]

I guess you need to have your natural eyes replaced with optical neural enhancers to catch all the frames per second, but if you are even thinking about paying $3000 for a gfx, then you probably got that fixed already..  

[mkreku]

Maybe I'm turning into a raging AMD fanboi, or maybe I have something against a card that's priced at 200% of the competing product, yet delivers around the same performance.

 

Anyhow:

 

http://www.tomshardware.com/news/nvidia-gtx-titan-z-delay,26652.html

 

..it's been delayed and I felt a slight tinge of amusement.

[Keyrock]

Hopefully the 295X coming out at half the price of the Titan Z slapped some sense into Nvidia as to just how absurdly priced that card was going to be.

 

As for Fanboyism, I've generally bought Nvidia cards, but that's mostly down to me spending much of my time running Linux and Nvidia's Linux drivers being superior to AMD's.  I want the companies to be competitive, that way the prices are kept reasonable.  It's the same way I look at Intel and AMD.  It makes me roll my eyes when Intel fanboys revel in the fact that AMD is so uncompetitive in the mid-high end desktop segment, because that's the very reason that Intel can set whatever prices they want in that segment and sit on their hands in terms of increasing IPC, opting instead to concentrate on reducing power consumption to try to push into ARM's mobile space.  If AMD was more competitive with Intel, then Intel couldn't afford to sit on their hands, instead, we all lose.  So the fanboys' prize (and, unfortunately, the fate for the rest of us as well) for Intel being so insanely superior to AMD on desktops at the moment is higher priced, lower performing desktop CPUs for all.  Congratulations.  Thankfully AMD and Nvidia are still neck and neck with GPUs.  It would truly suck if we got the same situation with GPUs as we do with CPUs right now.

Edited May 1, 2014 by Keyrock