Nvidia Titan P at Gamescom

[Deli]

1 hour ago, patrickjp93 said:

Unprecedented demand and likely wafer supply problems because of Apple and Qualcomm having priority. Nvidia already stated more 1080s have sold than the total sales of 980TIs.

Nvidia sells more 1080 in less than two months than 980ti in a year? I have some doubt. Especially most 1080 available initially were F.E. variances.

[SLAYR]

13 minutes ago, Deli said:

The spec for the 6/8 pin cable and connectors are very conservative. The actual power they can safely handle is much higher. 250w easy per 6/8 pin connector, so 500w for two.

They are only rated for 150w each, and especially on lower end power supplys, they have to stay within that limit, otherwise you will have power supplies exploding.

41 minutes ago, Briggsy said:

 

Even the Rufy X with its 2x 8-pin config and high potential power draw was still barely touching the slot.

 

The fury x was 275w tdp, so it didn't need power from the slot, but if the new titan had a 375w tdp it will be pulling 75w from the slot, since the eight pin connectors are only rated for 150w each.

[Deli]

1 hour ago, SLAYR said:

They are only rated for 150w each, and especially on lower end power supplys, they have to stay within that limit, otherwise you will have power supplies exploding.

But we're talking about Titan P power consumption. If someone decides to pair their Titan P with a cheapy PSU, then good luck for them.

 

Someone's willing to spend more than 1K for a GPU, but not $100 on PSU. They deserve to have their PC blow up.

[Briggsy]

8 minutes ago, SLAYR said:

if the new titan had a 375w tdp it will be pulling 75w from the slot, since the eight pin connectors are only rated for 150w each.

That's not really how it works. Where the power comes from is deliberately assigned.

 

How about an example that can't be so easily dismissed (they cut and pasted the 6990 peak total in by accident, that number is wrong).

 

 

[Briggsy]

double post ftw

[patrickjp93]

24 minutes ago, Deli said:

Nvidia sells more 1080 in less than two months than 980ti in a year? I have some doubt. Especially most 1080 available initially were F.E. variances.

Well, according to Nvidia and AIB partners EVGA and MSI, that seems to be the case.

[Guest Kloaked]

6 minutes ago, patrickjp93 said:

Well, according to Nvidia and AIB partners EVGA and MSI, that seems to be the case.

I bet they're lying, especially EVGA and Nvidia. They're team green, so they obviously inflate sales... somehow...

 

/s

[Deli]

1 hour ago, patrickjp93 said:

Well, according to Nvidia and AIB partners EVGA and MSI, that seems to be the case.

I think Nvidia means the number of 1080 sold compares to 980Ti for the same initial period(first month). Not the total number sold for the 980TI since launched(more than a year).

[Coaxialgamer]

I firmly believe this is bs.  At least the part about titan being gp100-based.  Gp 100 measures a massive 610mm². Even titan x wasn't as big,  and that was late in 28nm's lifespan.  16 nm is new,  so they could never achieve good yields on a chip that large.   

The chip is simply not manufacturable in large quantities.  There is a reason the p100 costs 10000 dollars.

Not to mention there are plenty of features consumers won't use on gp100 : nearly 2000 fp64 cores,  an nvlink controller...  All this takes up space and costs money,  and won't be used.   It makes MUCH more sense for nvidia to release a titan /1080ti based on gp102. 

[MageTank]

This rumor stinks to the high heavens. First of all, 12 inches long? Really? Why would they need a reference card that long? I thought with HBM being 3d-stack based, It wouldn't need that large of a physical PCB surface area? Not only that, but putting the PCIE connectors on the front will add another half an inch or more. 

 

Next, we have them claiming 50% faster than a 1080, but its TDP is 110% higher? It's been a very long time since Nvidia last sacrificed efficiency for power, and I doubt they chose to do it this time around. Two variants of the card (8+8 and 8+6) is another doubt in my mind, as Nvidia already prevented non-reference models of the card. Why would they go out of their way to make two different Titan cards? One would assume they would just lift that silly rule and let AiB's go crazy. 

 

With HBM supposedly using less power than traditional GDDR5 (at least, this was the original claims), I can't think of any reason why the performance:watt would get out of control on these Titan cards. This seems like pure speculation, with very little evidence to go on. Will definitely not trust this source until I see more proof of their claims. 

[Lays]

3 minutes ago, MageTank said:

This rumor stinks to the high heavens. First of all, 12 inches long? Really? Why would they need a reference card that long? I thought with HBM being 3d-stack based, It wouldn't need that large of a physical PCB surface area? Not only that, but putting the PCIE connectors on the front will add another half an inch or more. 

 

Next, we have them claiming 50% faster than a 1080, but its TDP is 110% higher? It's been a very long time since Nvidia last sacrificed efficiency for power, and I doubt they chose to do it this time around. Two variants of the card (8+8 and 8+6) is another doubt in my mind, as Nvidia already prevented non-reference models of the card. Why would they go out of their way to make two different Titan cards? One would assume they would just lift that silly rule and let AiB's go crazy. 

 

With HBM supposedly using less power than traditional GDDR5 (at least, this was the original claims), I can't think of any reason why the performance:watt would get out of control on these Titan cards. This seems like pure speculation, with very little evidence to go on. Will definitely not trust this source until I see more proof of their claims. 

 

Are they saying the PCB will be 12 inches long? Or just the whole card?

 

It's possible they have a cooler longer than the PCB, "Fury" style like with what Sapphire and some other people were doing.

[MageTank]

2 minutes ago, Lays said:

 

Are they saying the PCB will be 12 inches long? Or just the whole card?

 

It's possible they have a cooler longer than the PCB, "Fury" style like with what Sapphire and some other people were doing.

They just said the card itself is 12 inches long. Still, I can't remember a reference Nvidia card being longer than 10.5 inches. If it is, then my case design will need to be changed. RIP 4L ITX machine. 

[Guest Kloaked]

8 minutes ago, MageTank said:

This rumor stinks to the high heavens. First of all, 12 inches long? Really? Why would they need a reference card that long? I thought with HBM being 3d-stack based, It wouldn't need that large of a physical PCB surface area? Not only that, but putting the PCIE connectors on the front will add another half an inch or more. 

 

Next, we have them claiming 50% faster than a 1080, but its TDP is 110% higher? It's been a very long time since Nvidia last sacrificed efficiency for power, and I doubt they chose to do it this time around. Two variants of the card (8+8 and 8+6) is another doubt in my mind, as Nvidia already prevented non-reference models of the card. Why would they go out of their way to make two different Titan cards? One would assume they would just lift that silly rule and let AiB's go crazy. 

 

With HBM supposedly using less power than traditional GDDR5 (at least, this was the original claims), I can't think of any reason why the performance:watt would get out of control on these Titan cards. This seems like pure speculation, with very little evidence to go on. Will definitely not trust this source until I see more proof of their claims. 

Well we know a Titan card is coming. The rumors are almost always far from reality just to get them clicks. Though, Titan cards do use a lot more power than the cards below it - I don't know the math to get the exact numbers. 

 

If the speculation is even close to the real thing, they may be doing something different with this Titan card than previous ones (well obviously but you know what I mean).

[SurvivorNVL]

If this is true, I guess it is time to upgrade.

[patrickjp93]

43 minutes ago, Deli said:

I think Nvidia means the number of 1080 sold compares to 980Ti for the same initial period(first month). Not the total number sold for the 980TI since launched(more than a year).

No, it's overall total sold (note this includes some orders that haven't been filled yet too).

[Suika]

38 minutes ago, Coaxialgamer said:

I firmly believe this is bs.  At least the part about titan being gp100-based.  Gp 100 measures a massive 610mm². Even titan x wasn't as big,  and that was late in 28nm's lifespan.  16 nm is new,  so they could never achieve good yields on a chip that large.   

There was a rumor awhile back stating the Titan P / 1080 Ti (if such a thing comes to fruition) would rely on GP102 rather than GP100, due to issues with yields on GP100. If I'm not mistaken the currently known Tesla P100 actually has several SMs disabled, possibly hinting that yields really aren't good enough for a full fledged GP100 core, at least not for release, which may add up with the rumor.

[biggiebody]

Titan ePeen

[Lethal Seraph]

That naming sounds pissy.

[Maslofski]

so whoever gets SLI titan Ps has a pretty big "PP"

[Serin]

Only 16gigaborts of vram. Disappointed but also relieved.

[Lays]

6 minutes ago, Serin said:

Only 16gigaborts of vram. Disappointed but also relieved.

wat

[Serin]

5 minutes ago, Lays said:

wat

Isn't enough to justify going up from 12gb. If it'd been over 20 I'd have been seriously considering it.

[Lays]

3 minutes ago, Serin said:

Isn't enough to justify going up from 12gb. If it'd been over 20 I'd have been seriously considering it.

I don't really see the point in having some crazy amount like 16gb anyways, by the time we have titles that require that much VRAM, we're probably going to have WAY faster GPU's.  

I think it's cool to be overkill, but for the most part 20+ just doesn't seem needed at all, or even in the near future atleast. 

[Serin]

1 minute ago, Lays said:

I don't really see the point in having some crazy amount like 16gb anyways, by the time we have titles that require that much VRAM, we're probably going to have WAY faster GPU's.

Then you mustn't be aware of how important vram is to 3d modelling and rendering.
I'd agree that getting any of the titans is completely unnecessary for gaming in anything but multi-4k setups, but for those of us in the 3d industry the more vram we can get the better. 

[Lays]

4 minutes ago, Serin said:

Then you mustn't be aware of how important vram is to 3d modelling and rendering.
I'd agree that getting any of the titans is completely unnecessary for gaming in anything but multi-4k setups, but for those of us in the 3d industry the more vram we can get the better. 

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?