Originally Posted by sherlock
OG Titan(built on a still new 28nm process, like Titan P is on a still new 16nm process) is a better example than Titan X(which is built on a mature 28nm process), and Titan X(6 month after 980) is where all the optimistic People get their "Titan P in 2016" idea from.
Nvidia like Intel is competing with their old lineup, what does/did Intel do when competing with their old lineup:
- Drag the release cycle longer(12 month cycle enlongate into 16-18 month cycle)
- Make Generational Product Improvement smaller(Tick-Tock goes to Tick-Tock-Tock and s)
- Increase Price for the same SKU for new generations
True, the OG Titan is a better data point than the Titan X, but it is only one data point and it is the first use of the "Titan" branding in a situation where the next planned node failed and Nvidia was reinventing their roadmaps. I seriously doubt we can use the OG Titan as a reference for the future given the 780Ti and Titan Black, those will never happen again.
The entire 28nm node was new and different, GK104 was released as the flagship for the first time, GK100 failed and needed to be reworked into the GK110, the 20nm node failed, and the invention of a new high margin price point all confused the release schedule and make basing future predictions on what happened then a bad idea, in my opinion.
A big difference between Nvidia and Intel is that Nvidia's new products are still significantly faster than their old ones. Intel is also chasing very different markets right now, the server market is well planned, mature, and predictable. The GPU market is still weird and evolving quickly, Nvidia holding back on GPU R&D right now is not safe in the way holding back on CPU R&D is for Intel.
Originally Posted by sherlock
All of this works against the optimistic projection of August 2016 Titan P at $1000 with 50% improvement over 1080
. When competing with your old lineup only, it makes more sense to slow down pace of development and save on R&D instead of keep pushing out new product.
Given How much more silicon a Titan P needs(610 vs GP104 which is around 314) and HBM2 price vs GDDR5X, not as much as you think if they keep the price at $1000. 1080 at $650-700 might be more profitable per unit than a $1K Titan P at this stage of yield(and all the good GP100 chip have to go to Compute because that's more profitable) and HBM2 price/avaliablity. Also Titans don't sell much to begin with, There are already more 1080 than Titan X
in people's rigs according to the June Steam Survey.
I agree that a Titan P in late August seems unexpectedly soon but I think we might have another new situation with the HBM2 cards. At a $1000 I will agree that the Titan might not have great margins, buy is $1000 really the market limit for a Titan P with 16GB of HBM2, I don't think so. This is the first time the Titan would actually be significantly different compared to the X80Ti. What does Nvidia do with all the leaky, high power use, Tesla chips? Selling them to me now, for well above $1000, might strike Nvidia as a good idea. I am not going to buy a 1080 but I am willing to drop a lot of money on something that would be significantly faster than my 980Ti, and, like a lot of people, I will do this every year if something better is available. Nvidia really wants people like me buying a Titan instead of Ti so they need something to differentiate the Titan enough that I am willing to pay that extra. Releasing something like the 780 Ti after the OG Titan hurts a lot, and the Titan X is much too similar to a 980 Ti to do it. However, a version with HBM2 that gets released early? Their average price per card sold could go through the roof.
The Titan P 12 and 16GB cards could be very expensive (as if $1000 wasn't) and be out a long time before the 1080 Ti. We might not see the GP102 on a 1080 Ti until spring, Nvidia might even call it the 1180 like @guttheslayer keeps claiming. This would allow the Titan line to sell more units, without the Ti undercutting it, it gives flush 1080 owners something to upgrade to quickly*, and it even reinforces the Titan brand as the "super high margin
I think Nvidia is still figuring out how the Titan, x80Ti, and standard x80/x70/x60 lineup fit together to generate the most money.
*six months might be enough time that people want get something new and many just aren't the type to drop stupid amounts of money on the Titan line.Edited by Asmodian - 7/9/16 at 3:01pm