Kepler performance

[JAE]

Originally Posted by sdlvx

Things are seriously messed up. There shouldn't be discrepancies of nearly 400% between reviews.

Some sites follow the Nvidia Reviewer's Guide to the letter while others do their own thing.

[sdlvx]

Originally Posted by JAE

Some sites follow the Nvidia Reviewer's Guide to the letter while others do their own thing.

That makes sense. The ones that have GTX 680 with the biggest leads are the ones that don't talk about clocks.

[Junior]

Originally Posted by JAE

I don't think it sucks but I do think it's a bit of a disappointment. From my point of view if you are going to show up to the party 3 months late then you should have a card that outperforms the competition by more than the paltry 10% exhibited by the GTX 680. It's a good card and does outperform the competition in many of the important metrics but I don't think it has lived up to the hype.

AMD is actually quite fortunate that the only thing NV are able to launch at this point is the GK104, which is obviously lacking bandwidth. Things would likey look pretty bad for them if the 500mm+ BigK showed up with a 384 bit bus and 50% more alu's. I guess we will see on that later in the year...

Look, the thing is 294mm, 256bit and is going toe to toe with AMD's finest chip and surpassing it some tests. Alot of you claimed it would get its rear handed to it in 3 monitor scenarios and as [H] pointed out it holds its own.

Basically if this card from NV doesn't impress you even moderately for gaming purposes then no card they will ever produce will. If you want to look at perf/mm GK104 is ~38% larger than pitcairn and usually is 30-40% faster. The true test comes in a couple months when GK106 duels with pitcairn at similar silicon real estate. We shall see....

[Jimbob]

check out evga's site. looks like they'll be launching a 4GB version at some point

[JAE]

Originally Posted by Jimbob

check out evga's site. looks like they'll be launching a 4GB version at some point

Not surprising but expect it to be much like the GTX 580 3gb; limited supply and at a substantial price premium.

[sdlvx]

Originally Posted by Junior

AMD is actually quite fortunate that the only thing NV are able to launch at this point is the GK104, which is obviously lacking bandwidth. Things would likey look pretty bad for them if the 500mm+ BigK showed up with a 384 bit bus and 50% more alu's. I guess we will see on that later in the year...

There's a pretty good reason why a 300mm chip costs $499 right now. I don't think it's even possible to make a 500mm+ chip right now on TSMC without it being outrageously expensive.

In the fall, 28nm will get cheaper, GK104 and Tahiti will probably get slight tweaks and have their marketing names incremented by a generation with massive price drops.

Here's how I see fall playing out:

NV rereleases GK104 as GTX 760 and/or GTX 770 for around $299 or $399. Big Kepler comes out for about $499 or $599 and NV gets painted as having a huge generational jump in performance, when they just made a big chip.

AMD is caught with its pants down, releasing a Tahiti 8970 running at 1.1or 1.2ghz, which doesn't compare with Big Kepler. AMD goes back to the mid segment with 8970 priced at about $299 or $399 with a dualie to compete with Big K.

I don't see why AMD has no interest in making big chips. If they did, they'd have something competitive.

The only saving grace I see for AMD is the fact that they've landed all the next gen consoles, which means they're probably going to sell more GPUs than NV even if NV had the entire PC market.

[Jimbob]

sorry, HERE is the link

[DCO]

Originally Posted by Medallish

According to OBR it obliterates 7970.. With it's 5% advantage, funny how an average of 24+% is a dissapointment when it goes the other way around.. Sorry just had to ridicule that moron, I can't take fanboys like that, especially now when people think he has anything legitimate to say.

Hehehe, fanboys are funny with their charade:
- Perf/mm^2 is not important ---> Perf/mm^2 is important.
- X% perf difference sucks --> (x+10)% perf is incredible.
- 500$ card, greedy morons --> 500$ card awesome, give me two.
- It has lower minimums, boo --> It has lower minimums, bah, now I prefer the average.
...

[JAE]

Originally Posted by sdlvx

I don't see why AMD has no interest in making big chips. If they did, they'd have something competitive.

Lots of development costs for relatively few sales. I doubt AMD will make another very large GPU until they gain more market share in the workstation market where you could actually make enough profit on the big chips to make their development worthwhile.

[Copper]

Originally Posted by JAE

I don't think it sucks but I do think it's a bit of a disappointment. From my point of view if you are going to show up to the party 3 months late then you should have a card that outperforms the competition by more than the paltry 10% exhibited by the GTX 680. It's a good card and does outperform the competition in many of the important metrics but I don't think it has lived up to the hype.

Remember back when Charlie wrote up that NV taped out the smaller chip first and not the big one? This was supposed to be their mid-range and lower (for cutting and binning) chip.