Nvidia Titan V

razor1

[H]F Junkie
Joined
Jul 14, 2005
Messages
10,155
Because its a V100 card targeted for something else than gamers and not sold for 500$. Its the poor AI researchers card with 110Tflops for 3000$.

The hilarious part is that it got one broken HBM2 stack/controller while the rest of the chip is intact. Perhaps why HBM3 seems to be ditching the stupid parallel bus and try to copy HMC with SerDes whenever it comes in a few years.

This is the internal marketing line for AMD,

HBM for the masses, *whisper how the hell are we going to make money on this thing

This is the internal marketing line for nV

HBM, *whisper we will rake in big bucks
 

spine

2[H]4U
Joined
Feb 4, 2003
Messages
2,626
Because its a V100 card targeted for something else than gamers and not sold for 500$. Its the poor AI researchers card with 110Tflops for 3000$.

The hilarious part is that it got one broken HBM2 stack/controller while the rest of the chip is intact. Perhaps why HBM3 seems to be ditching the stupid parallel bus and try to copy HMC with SerDes whenever it comes in a few years.
I think it's also the first Titan card to actually have "Titan" written on the side. o_O
 

NukeDukem

2[H]4U
Joined
Feb 15, 2011
Messages
2,317
You do understand that this is not for gaming...right? Seriously this is not hard to understand....
You do understand you are speaking to VEGA, right? Seriously, his middle name is Overkill, it's not [H]ard to understand :punch:

Seriously though I'd be slightly concerned if he WASN'T in on this day one. I would have liked to see him go SLI, but alas it doesn't appear to be an option.
 
Last edited:

Araxie

Supreme [H]ardness
Joined
Feb 11, 2013
Messages
6,439
Any reason why nVidia is using HBM? Seems that AMD got a lot of crap here for doing the same.
because the advantages of HBM are under Scientifics usage and professional market, this card it's no more than a cheap Tesla V100, yes, it's at 3000$ it's a god damn bargain as a Prosumer card, getting Full GV100 Chip, when a Tesla V100 Goes easily for over 10.000$. you should expect a GV102 later without tensor cores, with crippled DP performance and GDRR6 for the average gamer and consumer which will probably clock and perform better. This card it's a paradise for small studios who can't afford tesla V100
 

Sprayingmango

[H]ard|Gawd
Joined
Jan 20, 2012
Messages
1,259
You do understand you are speaking to VEGA, right? Seriously, his middle name is Overkill, it's not [H]ard to understand :punch:

Seriously though I'd be slightly concerned if he WASN'T in on this day one. I would have liked to see him go SLI, but alas it doesn't appear to be an option.

It’s not overkill though...a 1080Ti or Titan Xp will be faster than this for gaming. It’s just an awful waste of money for no reason. It’s a downgrade.
 

Araxie

Supreme [H]ardness
Joined
Feb 11, 2013
Messages
6,439
It’s not overkill though...a 1080Ti or Titan Xp will be faster than this for gaming. It’s just an awful waste of money for no reason. It’s a downgrade.
what are you smoking?.. this thing should be a in the range of at minimum 40% to 50% better performer than 1080Ti or Titan Xp, and it can go easily beyond that factoring architectural improvements more than Raw specs and clocks,

Titan Xp 3840 Shaders, 240 TMUs, 96 ROPS, boost clock of 1582Mhz, 547GB/s bandwidth in a 384 bit bus, 4mb cache L2

Titan V 5120 Shaders, 320 TMUs, 128 God damn ROPS boost clock of 1455Mhz, 652GB/s bandwidth in a 3072 bit bus, 2 extra mb L2 cache for total of 6mb cache L2

and you think it will be a downgrade.. lol.. what a joke..
 

Armenius

Fully [H]
Joined
Jan 28, 2014
Messages
22,822
It’s not overkill though...a 1080Ti or Titan Xp will be faster than this for gaming. It’s just an awful waste of money for no reason. It’s a downgrade.
I don't know how you can say that for certain. If this was a Pascal vs. Pascal card I'd agree with. But this is a new architecture.
 

Sprayingmango

[H]ard|Gawd
Joined
Jan 20, 2012
Messages
1,259
what are you smoking?.. this thing should be a in the range of at minimum 40% to 50% better performer than 1080Ti or Titan Xp, and it can go easily beyond that factoring architectural improvements more than Raw specs and clocks,

Titan Xp 3840 Shaders, 240 TMUs, 96 ROPS, boost clock of 1582Mhz, 547GB/s bandwidth in a 384 bit bus, 4mb cache L2

Titan V 5120 Shaders, 320 TMUs, 128 God damn ROPS boost clock of 1455Mhz, 652GB/s bandwidth in a 3072 bit bus, 2 extra mb L2 cache for total of 6mb cache L2

and you think it will be a downgrade.. lol.. what a joke..

I really thought H members were smarter than this....*sigh*

It's gear towards deep learning / AI algorithms and highly optimized as such. Nothing in the gaming world can even utilize the Tensor cores and the architecture that this card uses. Hey if you want to blow money and make your GAMING PC slower then who am I to stop you.
 

razor1

[H]F Junkie
Joined
Jul 14, 2005
Messages
10,155
I really thought H members were smarter than this....*sigh*

It's gear towards deep learning / AI algorithms and highly optimized as such. Nothing in the gaming world can even utilize the Tensor cores and the architecture that this card uses. Hey if you want to blow money and make your GAMING PC slower then who am I to stop you.
depends on the drivers, Quadro's have no problem hanging around gaming cards in games, yeah little bit slower in some cases, but not by much.

Titan V's driver is bundled with the Geforce driver, they just released a new Geforce driver today with Titan V in it.
 

defaultluser

[H]F Junkie
Joined
Jan 14, 2006
Messages
13,547
Any reason why nVidia is using HBM? Seems that AMD got a lot of crap here for doing the same.
It's a cut version of the card with a single disabled HBM2 stack, so this i s actually making Nvidia money (they couldn't sell it as a Quadro with such low bandwidth, since the performance is so much higher than p100).

HBM2 stacks have higher failure rates compared to GDDR5X on a PCB.

We only complain about Vega's HBM2 because AMD was selling cards with 8GB for under $500
. This thing is priced at $3000, which is way higher profit margins than even the 16GB Vega Frontier Edition.

Nvidia was able to sell P100 with a disabled block of HBM2 because it had more than enough bandwidth for compute. But I expect this much higher performance is pushing the limits of HBM2, so the cut v100 cards are more suited for gaming?

https://exxactcorp.com/index.php/product/prod_detail/1662

Could also be a cover for the possibility that GDDR6 is late?
 
Last edited:

DejaWiz

Oracle of Unfortunate Truths
Joined
Apr 15, 2005
Messages
19,799
Keep buying up nVidia's $1-3K halo video cards, folks...that kind of profit helps their R&D provide me with having a killer mid-upper to upper range GPU available to purchase that will give me at least double the performance as a reward for waiting a couple generations between upgrades. :D
 

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
2,410
I really thought H members were smarter than this....*sigh*

It's gear towards deep learning / AI algorithms and highly optimized as such. Nothing in the gaming world can even utilize the Tensor cores and the architecture that this card uses. Hey if you want to blow money and make your GAMING PC slower then who am I to stop you.

I don't know about Tensor, but the amd 7900 had alot of fp64 capability that was not needed for gaming. Didn't make it a bad gaming card for the time.
 

Krisium

Weaksauce
Joined
Feb 16, 2016
Messages
75
Source for 128 ROPs? Not believing until I see it.

Also if the memory and L2 is cut by 25%, then ROPs are too. At least that is how every Nvidia card has been this decade.
 

Comixbooks

Ignore Me
Joined
Jun 7, 2008
Messages
15,336
I wonder if you can trade in a Jedi order card for this that someone bought last week.
 

Comixbooks

Ignore Me
Joined
Jun 7, 2008
Messages
15,336
Someone will get one of these by the weekend and post some benchmarks not even sure what game would be best suited.
 

razor1

[H]F Junkie
Joined
Jul 14, 2005
Messages
10,155
Keep buying up nVidia's $1-3K halo video cards, folks...that kind of profit helps their R&D provide me with having a killer mid-upper to upper range GPU available to purchase that will give me at least double the performance as a reward for waiting a couple generations between upgrades. :D

Its so tempting, but that 12gb HBM is a deal breaker for the stuff I do, need to see it in action to see if it will work with high def textures, 16k + rendering, otherwise its will be bottlenecked to what why Titan X is doing right now.
 

Aluminum

Gawd
Joined
Sep 18, 2015
Messages
687
#yolo

Also:

I really thought H members were smarter than this....*sigh*

It's gear towards deep learning / AI algorithms and highly optimized as such. Nothing in the gaming world can even utilize the Tensor cores and the architecture that this card uses. Hey if you want to blow money and make your GAMING PC slower then who am I to stop you.
You're going to want some heavy seasoning for that crow, these will be showing up on doorsteps in a few days to prove you wrong. Yes, it is compute focused with additional hardware and functions but all of the units he listed that is has (quite above a full fat pascal) are completely usable in games today.
 

DejaWiz

Oracle of Unfortunate Truths
Joined
Apr 15, 2005
Messages
19,799
Its so tempting, but that 12gb HBM is a deal breaker for the stuff I do, need to see it in action to see if it will work with high def textures, 16k + rendering, otherwise its will be bottlenecked to what why Titan X is doing right now.
12288 MB HBM2 Total Video Memory
3072-bit Memory Interface
652.8 GB/s Total Memory Bandwidth
384 GigaTexels/sec Texture Rate (Bilinear)

That's not enough?
 

defaultluser

[H]F Junkie
Joined
Jan 14, 2006
Messages
13,547
12288 MB HBM2 Total Video Memory
3072-bit Memory Interface
652.8 GB/s Total Memory Bandwidth
384 GigaTexels/sec Texture Rate (Bilinear)

That's not enough?
He's actually using it for professional applications, where memory size and bandwidth are often key.
 

Dayaks

Supreme [H]ardness
Joined
Feb 22, 2012
Messages
7,879
I really thought H members were smarter than this....*sigh*

It's gear towards deep learning / AI algorithms and highly optimized as such. Nothing in the gaming world can even utilize the Tensor cores and the architecture that this card uses. Hey if you want to blow money and make your GAMING PC slower then who am I to stop you.
Tensor is separate from the 5120 cores. Shaders alone with absolutely no other improvements will net you 33% more performance in games.

$3k is a bit steep for 33% personally.... will wait for reviews.
 

5150Joker

2[H]4U
Joined
Aug 1, 2005
Messages
4,016
Tensor is separate from the 5120 cores. Shaders alone with absolutely no other improvements will net you 33% more performance in games.

$3k is a bit steep for 33% personally.... will wait for reviews.
Since this isn’t a GeForce card I doubt review samples will be given out. Users who buy it will be our primary source of information.
 

Mega6

2[H]4U
Joined
Aug 13, 2017
Messages
2,735
Since this isn’t a GeForce card I doubt review samples will be given out. Users who buy it will be our primary source of information.
Don't worry - there is someone out there now gaming the crap out of this compute card with a big $3K grin on their face.
 

DejaWiz

Oracle of Unfortunate Truths
Joined
Apr 15, 2005
Messages
19,799
He's actually using it for professional applications, where memory size and bandwidth are often key.
Titan X...
12GB GDDR5X
384bit VRAM Interface
480GB/sec memory bandwidth
379.7 Gtexels/s fill rate


Titan V seems to have an amazing advantage to bandwidth, but the RAM size and texel fill rate are essentially identical. Maybe it won't be worthwhile, unless the bandwidth plays a much greater factor here.
 

Hagrid

[H]F Junkie
Joined
Nov 23, 2006
Messages
8,914
So I guess somebody will report back on how it does in games. I read it wasn't made for gaming, but dam 3K is way too high even for a gaming card.
 

NukeDukem

2[H]4U
Joined
Feb 15, 2011
Messages
2,317

Interesting what Steve has to say from 2:29-3:13... maybe not such a gaming beast after all?
 

razor1

[H]F Junkie
Joined
Jul 14, 2005
Messages
10,155
well right now I'm doing fine with my Titan X, I hope they increase the memory sizes this time around on their Prosumer geforce cards.
 
Top