[bryanlarsen]

The big problem with that strategy is that it's a really bad time to buy an expensive video card.

Jen-Hsun Huang, CEO of nVidia has stated that the next generation of video cards will be 10x as fast as current cards.[1] That's probably an extremely cherry picked example, but it is obvious that the next generation will be a lot faster. All of the following are lining up:

- 28nm -> 14/16nm node jump allows twice the transistors in the same space. For video cards, that basically means double the performance.

- 28nm -> 14/16nm promises a 65% increase in transistor switching speeds

- 28nm -> 14/16nm promises a 70% decrease in power consumption. The very high end cards are currently power-limited, and many are limited by cooling.

- HBM2 memory promises a 3x increase in memory throughput

- Vulkan / DX12 promise to reduce CPU bottlenecks, this is the first generation to be explicitly optimized for Vulkan & DX12.

- This generation of video cards is the first to receive explicit optimizations for VR

- Both AMD & nVidia are promising significant architectural improvements.

Both AMD & nVidia are aiming for summer releases so that they solidly nail the August/September buying season.

Waiting a few months will also give a chance for the reviews to roll in. Early indications are that the Vive is the one to buy, but that's early. Things can change quickly as games come out, and nobody's had a really good look at the PSVR, AFAICT.

If I had a proper video card, I'd be jumping in now on the Vive. But I don't, so I'm going to wait, as painful as that is.

1: http://wccftech.com/nvidia-pascal-gpu-gtc-2015/

[saiya-jin]

i heard such claims before, you can safely ignore them. nobody is jumping even 2x in performance, not straight away, it's a stupid move from financial point of view and they never did it before so why start now.

extremely happy to be proven wrong in near future :)

[bryanlarsen]

Lots of historical precedence. The first few generations of 3D cards used to more than double in performance every generation. They'd get a process node bump, an architectural bump, a DRAM bandwidth bump and they'd often be bigger than previous generations.

[saiya-jin]

You are not correct, there was NEVER 10x jump (apart from software render of same scene vs 3dfx one, which is not apples to apples).

True, initial first few generations were sometimes progressing fast, but only because architecture was clunky and everybody was just exploring new realm of chip & API design.

So, any outrageous 10x performance jumps in this industry IN LAST 15 YEARS? :)

[yread]

8800gtx could do more than 2x the fps compared to 7900gtx so in theoretical perfomance the difference was likely even bigger

[bryanlarsen]

I explicitly said that the 10x claim was nonsense. I expect 2x-3x.

[saiya-jin]

you are right, my bad for reading too fast. still unconvinced we'll see a big performance breakthrough, although list you posted seems impressive

[bryanlarsen]

No problem, I do the same too, sometimes.

Your point about financials is spot on, though. We don't have the competition we had during those early generations; I expect nVidia to bump prices. I think that the Titan and 980 ti cards have sold surprisingly well, so I expect nVidia to try even harder to find enthusiast's price ceiling, as well as try to bump the $200-$300 sweet spot up in price a bit.

[newman314]

Yup. I want a new machine but am holding off because of Pascal. Not sure what cheap card I can get as throwawayish to play The Division. Guess I'll have to go something else instead :)