Nvidia Titan P at Gamescom

[Serin]

11 minutes ago, Lays said:

Isn't that why we have quadros for tho? For 3d modelling/rendering, etc?

Didn't the last Titan have it's double floating stuff disabled?

Most of the popular renderers don't require DP so it doesn't matter so much.
For instance, cycles - the renderer I use for the most part is unbiased thus double precision is irrelevant.
As such geforce is generally a better price to performance outcome.

[NumLock21]

Found a leak spec chart. Added to OP.

[That Norwegian Guy]

Another "must have" for young adults with little experience in computers and life.

 

So many times I too wanted to buy one of these halo cards, stopped myself at the last minute and looked back 4 months later relieved to have dodged that bullet.

 

Titan Z and Titan 1 owners in particular can look back in shame and weep.

 

Surprisingly, the best halo card in the past 10 years was the AMD R9 295x2. These monsters tend to be more worthwhile purchases when the company making them aren't abandoning refinements to their software after 6 months because they pre-baked those refinements for launch to look better in early reviews - using up more of the future potential doing so.

[Briggsy]

12 hours ago, NumLock21 said:

That definitely looks more realistic for Pascal TDP and consumer die size, and 24GB on the Titan....?!?!11!

 

too bad these cards probably won't be out for awhile, I'd be very tempted to do EVGA step up for the 1080ti otherwise, that thing will be a monster with proper cooling, probably equal to 1070 SLI in gaming performance.

[Rune]

13 hours ago, Trixanity said:

Lower demand, yes, but still demand.

 

And what are the issues with GP104 then? It's still out of stock. It's been almost two months since it was unveiled (and over a month since it actually launched). 

My store has actually quite a few 1080's in stock. We're getting a lot more than expected, but they definitely are selling through.

[Kunter]

7 hours ago, Serin said:

Only 16gigaborts of vram. Disappointed but also relieved.

If this leak is bs and they use GDDR5X I would be more happy with 16GB rather than 24 because it means they will use a 512Bit bus. 

And 512bit bus is almost as fast as the 720GB/s of HBM2:

512bit bus with 10GBit/s (1080 speed) means 640GB/s 

512bit bus with 12GBit/s ("one generation further") means 768GB/s 

 

Same with 24GB VRAM and 384 lanes:

384bit bus with 10GBit/s (1080 speed) means 480GB/s 

384bit bus with 12GBit/s ("one generation further") means 576GB/s 

 

tl;dr: 512bit bus is almost on par with HBM2.

 

Anyway, I dont believe this rumour, if it has HBM2, why would it have the longest (single GPU) reference board in the near past? What are they gonna pack on the board the use not only the space won by not having VRAM chips but also make it longer? Two rows of VRMs on the board (At least a 12 or 14 phase design on a reference)?? I cant imagine a reason to make such a long board.

[FPSwithaWacomTablet]

21 hours ago, DarkBlade2117 said:

No offence, but people like you are why NVIDIA does so well.

"I'm getting it even if its a huge rip off, yayayayayyayayayay"

P usually stands for e-peen, usually corresponds to wallet size and arrogance. 

[asim1999]

I hope that they enable 4 way SLI for this as this is their core market (Titans) for greater than 2 way SLI

[Notional]

Looking at the VRAM amount and bandwidth, it seems like the smaller Titan will only have 3 HBM2 stacks. Which probably is due to defective memory controller blocks like we saw on the 970. So the small Titan is the Titan 970 

[othertomperson]

On 05/07/2016 at 2:56 PM, DarkBlade2117 said:

No offence, but people like you are why NVIDIA does so well.

"I'm getting it even if its a huge rip off, yayayayayyayayayay"

If you want this level of performance you have no other option. AMD's most high end product rivals a GTX 970. Blame AMD for Nvidia being able to charge what they like, not people who have literally nowhere else to go.

[TheRandomness]

8 minutes ago, othertomperson said:

If you want this level of performance you have no other option. AMD's most high end product rivals a GTX 970. Blame AMD for Nvidia being able to charge what they like, not people who have literally nowhere else to go.

At the moment* Why they didn't bother letting anyone know about a 490, I don't know.

[patrickjp93]

11 hours ago, Kunter said:

If this leak is bs and they use GDDR5X I would be more happy with 16GB rather than 24 because it means they will use a 512Bit bus. 

And 512bit bus is almost as fast as the 720GB/s of HBM2:

512bit bus with 10GBit/s (1080 speed) means 640GB/s 

512bit bus with 12GBit/s ("one generation further") means 768GB/s 

 

Same with 24GB VRAM and 384 lanes:

384bit bus with 10GBit/s (1080 speed) means 480GB/s 

384bit bus with 12GBit/s ("one generation further") means 576GB/s 

 

tl;dr: 512bit bus is almost on par with HBM2.

 

Anyway, I dont believe this rumour, if it has HBM2, why would it have the longest (single GPU) reference board in the near past? What are they gonna pack on the board the use not only the space won by not having VRAM chips but also make it longer? Two rows of VRMs on the board (At least a 12 or 14 phase design on a reference)?? I cant imagine a reason to make such a long board.

Sigh, no...

 

High-end GDDR5X can be overclocked to 8GHz for 16 Gbps effective rate

 

512/8* 2*(8*10^9) = 1.024*10^12 or 1TB/s

[patrickjp93]

7 hours ago, Notional said:

Looking at the VRAM amount and bandwidth, it seems like the smaller Titan will only have 3 HBM2 stacks. Which probably is due to defective memory controller blocks like we saw on the 970. So the small Titan is the Titan 970 

No, it's probably due to cost. HBM2 stacks are not cheap.

[Notional]

1 minute ago, patrickjp93 said:

No, it's probably due to cost. HBM2 stacks are not cheap.

Could be, but given the yield problems with 16nm FF+, making these huge chips probably results in many defaults on them. When even the P100 is cut down, there obviously are issues. Cutting off part of the MC and using one less HBM2 stack makes sense.

[grayperview]

21 minutes ago, othertomperson said:

If you want this level of performance you have no other option. AMD's most high end product rivals a GTX 970. Blame AMD for Nvidia being able to charge what they like, not people who have literally nowhere else to go.

er the fury's are the high end for AMD, significantly faster than a 970.

[patrickjp93]

1 minute ago, Notional said:

Could be, but given the yield problems with 16nm FF+, making these huge chips probably results in many defaults on them. When even the P100 is cut down, there obviously are issues. Cutting off part of the MC and using one less HBM2 stack makes sense.

There are no yield problems anymore. AIB partners have already said the 1080 is a bigger success than the 980TI. For GP100, the Teslas have already sold the bulk of the P100 orders, so there should be a solid supply of GP 100 chips being built up right now.

[patrickjp93]

3 minutes ago, super_skank said:

er the fury's are the high end for AMD, significantly faster than a 970.

Not really if you keep the custom 970s in your calculations.

[grayperview]

7 minutes ago, patrickjp93 said:

Not really if you keep the custom 970s in your calculations.

No a fury is still at least 20% faster than the fastest 970, and the furys overclock 10-12% too....Can a 970 do 15000 on firestrike, can it fuck.

[othertomperson]

9 minutes ago, super_skank said:

er the fury's are the high end for AMD, significantly faster than a 970.

 

7 minutes ago, patrickjp93 said:

Not really if you keep the custom 970s in your calculations.

This, and I don't know what it is about the members of this forum and needing every little thing pointed out to them, but the point I was very obviously making was that there are no AMD products that compete with anything that Nvidia have this generation, which means that they can charge whatever the hell they like and if you need the performance you have to pay.

[grayperview]

The point is there are. they are called the fury...they are only 1 year old, AMD only bring top tier cards out every 18-24 months,

[othertomperson]

4 minutes ago, super_skank said:

The point is there are. they are called the fury...they are only 1 year old, AMD only bring top tier cards out every 18-24 months,

Two of them in Crossfire might just compete with a 1080. And also cost about £200 more.

 

30 minutes ago, TheRandomness said:

At the moment*

At the moment... I really don't see why you'd even bother adding that when I'm talking about an industry which doubles performance every 18 months or so. No shit it's "at the moment". But when AMD release something new, so will Nvidia so it'll be "at the moment" again.

[grayperview]

They are not meant to compete with a 1080, what fucking company brings out a card to compete with a card that's not out for another 12 months?

 

 

 

 

 

[Wyzzy Moon]

Continuing with random letters like this, we will inevitably arrive at the glorious.
"Titan D"

And than everyone with a R9 Nano would claim to have one

[Kulegamez]

On 7/5/2016 at 8:33 AM, SamStrecker said:

If it isn't 40% faster then my 980ti OCed then I am not getting it. I will wait for Volta. Seeing it has HBM2 and a different core then the 1080ti it might help justify a steeper price. 

50% faster than 1080 which is 25% faster than titan x. Much more than 40%. This post is very brief and gives very little detail.

[SurvivorNVL]

Wish we could just skip HBM and go right for HMC.  If they deliver 30% above a 1080, I'll be happy, if they deliver well above it, and at around $1000 / $1200.  They'll be doing great.  A single Pascal Titan sounds great.  Hell, get rid of the P X Z.  If they just called the cards <Science name> Titan, it'd be cleaner and nicer.  Maxwell Titan.  Kepler Titan.  Pascal Titan.  Volta Titan.  Geforce Volta Titan.  Doesn't sound bad at all.