Kepler lineup

[glimpse]

But it haz thousands moar of CUDA cores!! I believe 3-thousand?... That's like six 580's

If it would render like having this much cores..or at least half that, which translates into twice as good as GTX580 - i would take two please..but it seems that at the moment it's no going to happen.

dual-GPU GTX 690 is out.... and apparently, its cost is 999 USD

Well, when You look to materials they've used, their way of branding it leaves very natural question..how much a version from EVGA is going to cost? (1,2-1,3k$ =) and wheather at all we are going to see one..think Nvidia stepped a bit further in order cut some possibilities from those vendors and put in it's own bag =p

[bb3d]

Anandtech reviewed the new GTX 670:
http://www.anandtech.com/show/5818/nvid ... at-evga/15
They extended the compute benchmarks. LuxGPU still looks bad, but everything else looks pretty nice. Maybe we shouldn't give up hope...

[Daniel]

It's my understanding that all the 600-series cards have been optimised for gaming and are not intended for computational power. That's what the Tesla cards will achieve, that I think are supposed to be announced either today or tomorrow at GTC.

Edit: Just found this, 4 days old. http://www.brightsideofnews.com/news/20 ... -rate.aspx

8GB RAM?

[PolderAnimation]

Yeah but it is a dual gpu card so 4gb for each gpu, so octane will only see it as 4gb.

[glimpse]

Well, day or two and we'll have a good idea on how this beast is going to perform..anyway..these are rumors =) interesting what the actual product will be =) coulnd't sit still..GPU Technology Conference starts today.. =) let's wait and see what they are going to show! Pretty sure they figured out how to boost KEPLERS performance, compared to GTX line, 'cos without that their willingness to outsell TESLAs all previous quantities combined would be very naive.

[Daniel]

Yeah but it is a dual gpu card so 4gb for each gpu, so octane will only see it as 4gb.

Yeah, but at the moment the max we're getting is 2x2GB. Even 4GB is double what we can get at the moment!

[PolderAnimation]

not true, you can buy a gtx680 4gb version .

- Gainward GeForce GTX 680 Phantom 4GB

- Palit GTX680 Jetstream 4 GB.

[Daniel]

Oh, I was under the impression those were 2x2GB. I should have read up on them a bit more. Doesn't matter though. Those cards are useless for Octane as far as cost efficiency is concerned. But still, if you had one, the speed of a GTX 460 with 4GB RAM isn't too shabby.

[bb3d]

Now it's official:
http://www.legitreviews.com/news/13138/
http://forums.anandtech.com/showthread.php?t=2246084

There are 2 Tesla models - the K10 with 2x GK104 (available soon) and the K20 with 1x GK110 (available Q4 2012).
The interesting thing is, that Nvidia uses the "gaming chip" GK104 also for the Teslas and is claiming 3x SP performance compared to Fermi. This would mean, that the hardware of GK104 is capable of high performance GPU computing, but probably software developers have to optimize code for CUDA 5.0 to make use of it and unleash the power of the new chip.
Concerning GK110 Nvidia is talking only about the DP performance which is expected to be 3x of Fermi. But the SP performance seems to be comparable to 2x GK104.
So, the good news is, because SP performance is the important thing for us Octane users, the only thing we might have to wait for is a CUDA 5.x optimized Octane, but there is no need to wait for GK110 which will not be available before Q4 2012.

[mbetke]

1500-2500 USD per card. Big number... I would need two of them. :-7

German article: http://www.heise.de/newsticker/meldung/ ... 75385.html