Caporegime
I think you should get used to gains not always being as big as before going forward. We'll get the odd time when major architectural changes coincidence with a new node shrink or something, but otherwise, things are just getting much harder and more expensive to find the improvements that people have gotten used to.
In terms of the 1070 not being 20% more than the 980Ti like the 670 was over the 580:
1) The 1070 is more 'cut down' than the 670 was.
2) Pascal is not the same leap that Fermi->Kepler was. I think Pascal is more than just 'shrunk Maxwell'(you dont get these sort of clock gains for free), but it's also certainly not a major architectural change, either. Pascal was only inserted into the lineup a couple years ago, where Maxwell and Volta were the main architectural 'revolutions' that they were working on.
In terms of Nvidia/AMD 'milking' anything, do you think Nvidia just has Volta sitting on the shelf somewhere? Pascal was brought in *because* Volta is not ready and wont be for a little while. The alternative was that they stay on 28nm Maxwell until Volta was ready. You'd have been ok with that?
And neither Nvidia/AMD can do anything about the length of time it takes for the next process shrink to come about. I'm sure Nvidia were originally planning to release Maxwell alongside 20nm wayback when they first started working on Maxwell, but obviously those plans were scuppered. Releasing Maxwell on 28nm wasn't 'milking' anything, it was them doing what they could to release an improved product, which is what customers expect.
As I said, improvements are becoming harder and more expensive to come by. Moore's Low is slowly coming to an end and that will continue until some serious breakthroughs get made.
The thing is Nvidia mostly makes money from GPUs and their margins have gone up by nearly double over the last 4 years or so,so the strategy is very sound financially and kudos on Nvidia managing people's expectations. From their perspective its done very well for them,and hats off that they managed to get it to work for them but it is not so good for consumers.
The thing is though,if you looked at it in a basic view,the 314mm2 GP104 is not doing that badly,but the pricing structure and product segmentation is just out of whack.
But the problem it is screwing over the sub £250 market in the process.
Look at the last JPR report:
https://jonpeddie.com/images/uploads/news/graph-pr-2rev2.png
The Enthusiast market is above $300(or around £200 in our money),and according to reports a while back the market above $449(£310) is much smaller. This means by its very pricing the GTX1070 is more niche.
Most cards sold are under $300 still,but more people are spending $300+ on cards.
This is because the sub $300 market has had such rubbish improvements people are forced to spend more and more on cards,by moving to the tier above.
I could even understand if these were the straight 20NM node cards we might have had last year,but Nvidia is using second generation TSMC 16NM and AMD second generation GF/Samsung 14NM. These are bascially third generation 20NM based processes.
I am now getting a bit worried what Polaris 10 and the GTX1060 will bring to the table. I expect they will be £250 and be around R9 390X/GTX980 level which will make them the same price/performance as the GTX1070.
Sadly it also means only a 10% to 20% improvement over their predecessors. This means instead of spending the same to get a performance improvement,people will now have to go to the segment ABOVE,AGAIN.
Realistically only the launch of BOTH the GTX1080TI and Vega 10 will make things more compeitive,ie, a full product stack from top to bottom.
Maybe,we will need to agree to disagree and leave it at that,but I am just getting more and more dissapointed by what we are seeing so far about Pascal and Polaris.
I am sure others will find the GTX1070 more exciting.
Last edited: