Nvidia GTX TITAN P to debut at Gamescom, Cologne in August

[LSG501]

Personally I can't see there being two versions of the titan p, I'm going to go on a limb and say the 12GB version is the gtx1080ti or whatever it's called.

Also the titans are aimed at people like me, people who want the memory of the top end quadros but aren't stupid/rich enough to pay the extra money needed for it

[Xlucine]

Titan Z wasn't a flop because of pricing - it was never meant to make sense from a value perspective. Titan Z was a flop because the 295X2 turned up and blew every other graphics card now and forever into the weeds.

The lower memory part sounds a lot like a 1080ti, if the specs are true - that drop in bus width looks like a salvaged part with only 3 stacks of HBM2. I thought AMD was certain to beat Nvidia to market with HBM2? Vega might be closer than previously thought

[Clubman]

It looks like GTX 10XX series become a training field an a needed step towards the Volta architecture.First,the Green ($) Team has speculated with phrases like 4K gaming,then with HBM2..and finally we recieved some middle result between both of them (the GTX 1080). Probably GP100 (Titan) and GP102 (hopefully Ti) will achieve spot on 60fps 4k gaming,but having in mind the price,the extortion will be completed, and that's why they have "circumcised" the GTX 1080.

[SRPlus]

The reason the Titan Z was a complete flop was not because of the 295x2, and definitely because of pricing.

The Titan Z was launched at an extortionate £2400+ in the UK, and I think at the time Nvidia’s reasoning for this decision was because they were worried Titan sales would kill off their Quadro and Tesla lines but it was a ridiculous decision, as you could purchase two Titan Blacks and have slightly better performance for £600+ less than one Titan Z.

The Titan Z would not have been popular with gamers even if it would have been priced right because of the 295x2, but it still would have been hugely popular amongst the professional crowd, people like me that use the Cuda technology for GPU rendering with Iray, scientific computation and development etc.

For instance, I have 4x Titans in my machine that burn 1000watts when rendering with Iray, but I could have opted and would have opted for the Titan Z at the time, as it would have meant 700+ watts with two of them to replace my 4x Titans, and better cooling with a nice gap between the cards instead of having 4 dual slot cards bunched together, but because of the extortionate price tag at launch, Nvidia made sure that the Titan Z was unappealing for gamers and professionals.

When they did eventually drop Titan Z pricing to around £1200 which it should have been in the first place, it was too late and Maxwell was already on the horizon. Titan Z still sells very quickly on places like Ebay as people buy them up for render nodes etc, so it was always about pricing..

So even though the 295x2 was a better option for gamers at the time, Titan Z still would have been successful with professionals if someone at Nvidia had done the maths and realised people will just by 2x Titan Back instead of one Titan Z if the pricing is too high.

My first post stands, if Nvidia decide to price the Titan P too high with fears it will eat into Quadro / Tesla sales too much, it could end up dead in the water like Titan Z, but I doubt Nvidia will make the same mistake….


I didn’t upgrade to 4x Titan X because I didn’t think a 46% performance increase in Iray rendering was worth spending £4K on, but I am very tempted with Pascal, and hoping Titan P wont be too far over the £1K mark, hopefully below £1K based on GP102, GP100, not so sure….

[Xlucine]

Fair points, I'd forgotten that professionals used titans - I'd assumed it was just for oil sheiks and the like with too much money to spend

[mapesdhs]

Titan's aren't meant to be gaming cards though - they're meant to be cheap computing cards - basically a consumer-grade Tesla. Look at the FP64 compute on them - it's generally much higher than on the X80 Ti cards. ...

Wow, is that ever wrong. NV stopped doing that kind of FP64 split with the Titan/Black/Z, ie. the 700-series based tech. Ever since then, the FP64 has been equivalently weak just like all the other gaming cards in the same series. Check the specs:

https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_900_Series

Indeed, even a GTX 580 is faster than a Titan X for FP64.

The other problem with using gamer cards for pro tasks is they don't have the same speed PCIe return path, they lack caching features present in Tesla cards and they lack ECC (critical for many pro tasks). NV gave up on the whole idea of pseudo-pro cards with the Titan X, it was nowhere near as attractive as a budget card for solo professionals as the 700-based Titans (only those who don't need FP64 or ECC would be interested in such products, which does include numerous video/3D tasks).

[mapesdhs]

Forgot to mention, SRPlus is spot on about the pricing issues. NV appeared paranoid about affecting Tesla/Quadro sales without seeming to understand the nature of those who would be interested in that kind of card. Where the differences between pro cards and gamer cards actually matter, users with such tasks wouldn't choose a gamer card anyway, except for a budget testing rig at home or someting (definitely not for any kind of final deployment). Others would opt for gamer cards for budget CUDA (iRay and apps like AE are typical examples), but pricing is critical. It's the same reason why those who can't afford the Titan level of GPU for CUDA can get so much value out of older 3GB 580s (my system with 4x 900MHz 580s is faster than two Titan Blacks).

[canopus72]

TitanP is actually an abbreviation for TitanPee-Pee