Nvidia Kepler

It looks like the release of the Nvidia GTX 680 is imminent. All current reports point to it being based on GK104 and priced at $500 or $550, with about a 10% performance increase over AMD's 7970. It seems we will have to wait for a more powerful card based on GK110.

Also rumored is a new, much more effective anti-aliasing technique and support for three displays on one card.

Leaked photos show a smallish card given its supposed performance, and it uses just two 6-pin power connectors. It could be a boon for high-end HTPC gaming applications where significant power and low noise are wanted.

Anyway, the truth about the GTX 680 should soon be available.


EDIT: Some review links below.

http://hardocp.com/article/2012/03/2..._card_review/1

http://www.anandtech.com/show/5699/n...x-680-review/1

http://www.xbitlabs.com/articles/gra...e-gtx-680.html

http://hothardware.com/Reviews/NVIDI...Kepler-Debuts/

http://guru3d.com/article/geforce-gtx-680-review/1

http://www.tomshardware.com/reviews/...mark,3161.html

http://www.hardwareheaven.com/review...roduction.html

There's an Nvidia marketing video out for the GTX 680 that talks about its power efficiency, new anti-aliasing method (TXAA), PhysX improvements, and support for up to four displays on one card.

I was just reading that. Not bad at all, especially at 1920x1080.

I hope a site does PhysX benchmarks in games like Arkham City, and compares it to the 580.

Given how small the GK104 is, there should definitely be room for significantly better performance with the much larger GK110.

Of course, those of us sticking with 1920x1080 TVs or projectors should be served well by the GTX 680 for quite some time.

Those who bought a 7970 should not feel bad at all when it comes to performance.

I just wish competition dictated for this card to be priced as the mid-range card that it probably should have been with the more powerful GK110 looming. Now people will have to pay through the nose for the GK110.

Some review links have been added to the original post.

A $100 price drop for the 7970 would be a win for the consumer, but a rather small win in my opinion. The big problem I think is that AMD may have set the bar much too low.

While the 680 easily beats the 7970 in 1080p with most games, the gap closes considerably in highest resolution gaming where performance matters most to many high-end gamers.

It just seems that the price-to-performance ratio with 28nm should have been much better.

It will be interesting see where prices are at by the end of this year when yields are optimal.

The GK110 is a single-GPU solution. It is rumored to have a massive die size of about 550 mm², and is expected to be released later this year.

I would imagine that it's mostly owners of multi-monitor setups who are waiting to see what the GK110 is about.

As for adaptive vsync, I think (I could be wrong) the idea is for vsync to be engaged when the frame-rate gets higher than the display's refresh rate, and then disengaged when the frame-rate is lower than the display's refresh rate. The problem with the latter point is that tearing will still occur (albeit to a lesser degree) as the output will not be perfectly sync'd to the display's refresh cycles. It is a nice option to have, but I'll stick with regular vsync and enable triple buffering when lag (or more specifically, frame-rate halving) is an issue.

In my experience with Direct3D games, forcing triple buffering almost always eliminates noticeable mouse lag in games that have it with vsync on. This revelation first came about when I tried it with the game FEAR.

Skyrim does have all kinds of oddities going on.

Forcing triple buffering on top of vsync has made for perfect gameplay with a number of games that were at the time taxing my GPU. Some games use triple buffering by default, and some others give you the option to enable it via menu (such as some older Tomb Raider games). Other Direct3D games, like FEAR and Just Cause 2, require use of D3DOverrider for this function.

I am definitely interested in knowing more about adaptive vsync. The chart I saw made it look like a simple algorithm that enables vsync when the frame-rate would otherwise be above the display's refresh rate, and then disables vsync when the frame-rate goes below the display's refresh rate.