NVIDIA GeForce GTX TITAN II Detailed

[HisDivineOrder]

I think I'm looking forward to when they make the 6 GB version of this and call it the 990 Ti or whatever they devise as their new numbering scheme.

Since last time, they went from four digit numbers to three digit numbers, I guess this will be the Geforce 18?

Eventually bringing us back to the Geforce 1 after a few more generations.

 

[Ralfies]

Has the 3072 cores actually been confirmed as the full complement available to the GPU?

No, these leaked specs haven't even been confirmed. But another gimped Titan followed by a fully enabled one down the road is equally disappointing. I would assume there wouldn't be yield issues on the very mature 28nm process either, meaning this would be gimped for the sole purpose of releasing a fully enabled version later. Although that is just an assumption, at the risk of making an ass out off myself.

Edit: Die size is also around the same size as GK110, while GM204 is quite a bit larger than GK104. This seems to indicate this will be at least close to a fully enabled chip.

 

[GhostRyder]

What video applications out there are confirmed to use excessive amount of VRAM?

Well Octane is very hungry as far as things go off the top of my head but that's a Cuda render program. Using the H.264 video codec in Sony Vegas I heard can be intensive on the VRAM a bit but nothing crazy like a lot of the Cuda render programs which basically eat as much VRAM as you serve it (I do not use Sonay Vegas so that is only based off what I heard so I am not sure that's true) .

Honestly I always view leaked specs as being something that is subject to change or even false because you never know where the information comes from. I think we should all take this as just a possibility of what to expect and wait to see what happens.

 

[Hilux SSRG]

Feel like we've all been through this before.

Titan 2 will be on 28nm and the fully enabled chip 990Ti ? will hopefully be on 16nm if the fabs aren't lying through their teeth.

 

[MxPhenom 216]

Feel like we've all been through this before.

Titan 2 will be on 28nm and the fully enabled chip 990Ti ? will hopefully be on 16nm if the fabs aren't lying through their teeth.

990Ti, when did that ever come about? All ive heard rumored is 980Ti on GM200 and 28nm. With TSMC track record, im not so sure that 20nm will be skipped either, unless we want more delays.

EDIT: Maybe not. http://www.techpowerup.com/forums/t...ss-achieves-risk-production-milestone.207127/

What video applications out there are confirmed to use excessive amount of VRAM?

A lot of professional rendering apps will use whatever you can give it. Gaming won't really but there are apps out there that will.

 

[Hilux SSRG]

990Ti, when did that ever come about? All ive heard rumored is 980Ti on GM200 and 28nm. With TSMC track record, im not so sure that 20nm will be skipped either, unless we want more delays.

EDIT: Maybe not. http://www.techpowerup.com/forums/t...ss-achieves-risk-production-milestone.207127/

Not sure what Nvidia may call the 780/780ti successors, whether they are on 20nm or 16nm. The more I see how preferred customers in phones and tablets get first dibs the more I think AMD/NVidia will be at the mercy of TMSC and may only get 20nm in 2015 and 16nm in 2016/2017.

Thanks for the article post, TMSC is all over the place with what's ready. Truly they must be using another dictionary than the rest of the world.

 

[Serpent of Darkness]

So in a perfect world, the following is probably true:

BF4:
GTX 980 reference at 1080p with 4xAA = 88 FPS.
GTX Titan Maxwell will probably have an FPS at around 129 and 162 FPS.

AMD R9-290x at 1080p with 4xAA = 67.8 FPS.
AMD R9-390x at 1080p with 4xAA will be around 98.6 FPS.

Gap between GTX 980 and R9-290x @ 1080p = 29.79%.
Gap between Titan-M and R9-390x @ 1080p = 31.0%.

Assuming of course, Memory Frequencies are not the bottleneck...

GTX 980 reference at 1600p with 4xAA = 53.0 FPS.
GTX Titan-M will probably have an FPS of around 77.7 and 98.1 FPS.

AMD R9-290x at 1600p with 4xAA = 42.1 FPS.
AMD R9-390x at 1600p with 4xAA = 61.2 FPS.

Gap between GTX 980 and R9-290x @ 1600p = 25.98%.
Gap between Titan-M and R9-390x @ 1600p = 26.96%.

GTX 980 reference at 4K with 0xAA = 29.3 FPS.
GTX Titan-M at 4K with 0xAA = 43.0 FPS.

AMD R9-290x at 4K with 0xAA = 24.9 FPS.
AMD R9-390x at 4K with 0xAA = 36.2 FPS.

Gap between GTX 980 and R9-290x @ 4K 0xAA = 17.7%.
Gap between Titan-M and R9-390x @ 4K 0xAA = 18.8%.

Assumptions:
1. GTX Titan Maxwell has a spec of 3,072 Cuda Cores at 1.1 to 1.39 Ghz.
2. R9-390x has a spec of 4096 SP at 1.0 Ghz.
3. This is a perfect situation.
4. Bottleneck occurs only at the GPU's end.

In theory, R9-390x is a 45.5% increase over R9-290x.
In theory, Titan-M is a 46.5% to 85.2% increase over GTX 980.

Titan-M > R9-390x > GTX 980 > R9-290x.

 

[alwayssts]

I didn't see this until today.

Feels nice that this is exactly what I thought it would be...I've been saying 24SMM 1200-1400 for months, 11xx-1390 is pretty close.

Now I just wait for amd to put out their (probably) 3840sp part....one would think they would be relatively similar per clock (granted no way in hell amd's clock speed is going to be close to 1400mhz).

I simply wonder if amd will go 28 or 20nm. On 28nm it should be roughly the same size to slightly smaller...but they would be at really low clocks (900-1000, maybe overclock to <1100?), on 20nm they might fair better (1200mhz?).

As I have said before, 4xHBM gives amd 512gbps, or enough to do roughly 3840 @ 1200 or so (1220?) with their current architecture.

What this does is give nvidia a guarantee it will be faster than FijiXT, and if I were a betting man (which I'm not, but they have positioned things as such before) their base clock is probably ever just so slightly above where they think their competition will overclock. IOW, nvidia probably thinks FijiXT is 28nm.

Speculation, yes...but makes sense don'it?