• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Titan V announced £2700 15TF 12nm Volta

I'm sure people could find a gaming use for them, but if you ran a game dev would you assign people and time to something that only works on PCs with a £2700 graphics card?

Of course. Today's premium product is tomorrow's commodity.

For a 73% increase in die area,and a 76% increase in transistors,FP32 performance has gone up by a whole 15% in FP32 performance and that is on a "new node".

Is double precision useful or relevant when dealing with 10 bit colour and high dynamic range?
 
Of course. Today's premium product is tomorrow's commodity.



Is double precision useful or relevant when dealing with 10 bit colour and high dynamic range?

Its more important for commercial and scientific computing. Almost all consumer software like games uses FP32 not FP64. FP16(AMD calls it Rapid Packed Math) might be useful in the future as the PS4 PRO and Vega also have it as does the GV100,but most cards don't so I suspect adoption will take time,and TBH the consumer cards will have probably better FP32 anyway relative to its core size.

I mean on a technical level though,Nvidia and TSMC making an 815MM2 chip is really pushing things,so its quite amazing to see something that size see the light of day!
 
Last edited:
So, nVidia went balls deep and all in!! I like the way these ****ers roll :D
Yep, damn those who speculate about NV and "holding back". There's no competition at the high end and look at what they do - take it to another level.
Looks like a gaming card to me, if one has the funds.
Doubt I'll be buying though as the original TXP is plenty fast for my needs for a while. Looks gorgeous though. The removal of GeForce I think is to create it's own premium Titan branding.

I did say we'd get volta before year end :p.
Actually saw the Nvidia release info on linked in and thought WTF :D
 
Last edited:
Since the big GP100 Pascal GPU had the same amount of shaders as the GP102 Pascal GPU we found in the Titan Xp,does this mean that big consumer Volta or consumer Ampere,will have 5120 shaders,but with the additional stuff removed like the Tensor Cores and reduced FP64 performance??

I also wonder whether it will have GDDR6 too??
 
Of course. Today's premium product is tomorrow's commodity. [..]

Games are usually developed with console hardware in mind. Most PC gamers have cards around the 1060 level. This forum is skewed well over average in terms of money. The latest technology that's now on a £2700 graphics card that's not even designed for gaming is not going to be on consoles or low midrange PCs tomorrow or any time soon. Games are not developed for hardware that might become mainstream in maybe 5 years time. They're barely developed for hardware that's mainstream today, judging by how normal buggy releases are. There's more time spent on ways of getting more people to pay extra after having already paid full price for a game.

Since the big GP100 Pascal GPU had the same amount of shaders as the GP102 Pascal GPU we found in the Titan Xp,does this mean that big consumer Volta or consumer Ampere,will have 5120 shaders,but with the additional stuff removed like the Tensor Cores and reduced FP64 performance??

It doesn't mean that it will or that it won't. It will depend on what nvidia can make cheaply enough for a gaming card. It's a plausible suggestion that nvidia will reduce their manufacturing costs by removing parts that are of little or no relevance to gaming, but it's not proven by what they did with Pascal. I'd take a bet that they will do the same this time (including cut-down cheaper versions to create a similar range of cards) because it makes sense, but it's not proven yet.
 
Last edited:
So whats the point of buying a watch for 100 000 bucks?? Shows same time as 1 pound one :D
i would choose a 100k rolex over a gpu, but i reckon if they made this with 24k gold plated, with diamond studs and a titanium heatsink and Lois Vuitton packaging, there would still be a market for it, hell theyd probably stock it in OCUK
 
It doesn't mean that it will or that it won't. It will depend on what nvidia can make cheaply enough for a gaming card. It's a plausible suggestion that nvidia will reduce their manufacturing costs by removing parts that are of little or no relevance to gaming, but it's not proven by what they did with Pascal. I'd take a bet that they will do the same this time (including cut-down cheaper versions to create a similar range of cards) because it makes sense, but it's not proven yet.

There is room for Nvidia to pack in more shaders on their next consumer part. TSMC 12NM is basically a lower leakage variant of their 16NM process as on their website its called 16NM/12NM:

http://www.tsmc.com/english/dedicatedFoundry/technology/16nm.htm

The GP102 is around 471MM2,and previous Nvidia consumer parts have been upto around 600MM2(GM200 was that size IIRC).
 
Games are usually developed with console hardware in mind. Most PC gamers have cards around the 1060 level.

And that was the premium level 2 generations ago - the GTX 780.

This forum is skewed well over average in terms of money. The latest technology that's now on a £2700 graphics card that's not even designed for gaming is not going to be on consoles or low midrange PCs tomorrow or any time soon. Games are not developed for hardware that might become mainstream in maybe 5 years time.

The development lifecycle of an AAA game is measured in years. The breakthrough, though, is more likely to come from an independent who will produce something cool and that technology will be licensed to the majors. And they will say to Nvidia, "We want this."
 
Anandtech are usually pretty credible but i think there is something odd with the specs they've published and i see a lot of cross linking to their article.

If you look at the numbers theyve given for the Tesla V100 then the calucalation works out perfectly, 5120 cores at 1370mhz is exactly 14TF sp (2 sp flops per core per clock), the Titan V with supposedly 5120 Cuda cores running at 1455mhz should deliver a theoretical 14.9TF sp but they've listed it at 13.8TF sp.

NVIDIA are advertising the full 5120 cores on their website so unless the NVIDIA marketing department are pulling another GTX970 and we are actually looking at more like 4800 cuda cores I think the Titan V is actually a fair bit faster than has been appreciated. It also means that an overclock to ~2000Mhz should hit nearly 20.5TF sp (probably looking at >400W at that point)

https://www.anandtech.com/show/1213...dia-titan-v-video-card-gv100-for-3000-dollars

Edit: The op for this thread has actually called out the correct (i think) performance
 
Last edited:
If NV sells enough Titan Vs, expect the Pascal refresh (aka Ambere) next years, starting with a nice 40% higher price tag.
 
Back
Top Bottom