NVIDIA Kepler speculation thread

[CarstenS]

http://www.bouweenpc.nl/wp-content/uploads/2012/03/MediaEspresso-339.png

What do you think of this?

I wouldn't want to convert my GTX 680 (if i owned one) to an iPad 2.

 

[Mize]

What do you think of this?

680 seems to have more CUDA cores as Mediawhateverespresso uses CUDA, right?
It's definitely fast.

 

[itaru]

Sandra gpgpu bench against tahiti AND pitcain:

http://www.theinquirer.net/inquirer...hed-amds-mid-range-radeon-hd-7870-gpu-compute
Wonder if the clock is really 700 or 1000 here..

8SP 8C 705MHz ???

 

[fellix]

What do you think of this?

Apparently, this test is compute throughput limited. The scaling to the GF110 follows the FLOPs doubling.

 

[HKS]

680 seems to have more CUDA cores as Mediawhateverespresso uses CUDA, right?
It's definitely fast.

Yes, MediaEspresso 6.5 uses CUDA for Nvidia and APP for AMD cards.

 

[Dave Baumann]

Might that be due to an encode engine...?

 

[jaredpace]

from 3dc

 

[CarstenS]

Might that be due to an encode engine...?

Does the one in Tahiti work with Media Espresso 6.5 yet?

 

[jlippo]

I'm not buying the "new architecture" bit just yet. So far it looks like an optimized GF114 Fermi on clock roids.

Neither do I.
Nvidia has always followed method similar to intels tick tock. (nv20 new architecture, nv25 speed upgrade.)
Fermi was new architecture, so Kepler is the upgrade part and Maxwell will be something else.

 

[Jawed]

I'm not buying the "new architecture" bit just yet. So far it looks like an optimized GF114 Fermi on clock roids.

I bet the scheduling nonsense of old has been deleted.

@Jawed, maybe 28nm wide+slow simply has better overall perf/watt/mm characteristics than narrow+fast on previous nodes.

It's worth noting that the entire chip is now running at much higher clocks. If rumors pan out the regfile, scheduler, tmus, rops, geometry etc will be running at speeds approaching G80's hot-clock.

This is a non-argument. Same applies to AMD's stuff.

 

[fellix]

Neither do I.
Nvidia has always followed method similar to intels tick tock. (nv20 new architecture, nv25 speed upgrade.)
Fermi was new architecture, so Kepler is the upgrade part and Maxwell will be something else.

Yep, just like NV40 to G70 and G80 to GT200.

 

[Creig]

So all we had to do to calculate the performance of the GTX 680 was take the numbers from a GTX 580 and multiply by three? Well, don't we look foolish now. We could have saved ourselves over 3200 posts of discussion if we had only known.

 

[Man from Atlantis]

 

[Kaotik]

So all we had to do to calculate the performance of the GTX 680 was take the numbers from a GTX 580 and multiply by three? Well, don't we look foolish now. We could have saved ourselves over 3200 posts of discussion if we had only known.

Err... multiply by 3? Having 3x ALUs doesn't increase performance by 3x, especially when the Fermi ALUs are hotclocked and these aren't, not to mention other possible differences

 

[SimBy]

Err... multiply by 3? Having 3x ALUs doesn't increase performance by 3x, especially when the Fermi ALUs are hotclocked and these aren't, not to mention other possible differences

Of course it does, ask RenderStream experts.

 

[Man from Atlantis]

norwegian etailer posts their kepler's 3dmark11 score.. P9398 graphics score

https://www.facebook.com/photo.php?...56145582074.250887.69170057074&type=1&theater

 

[elect]

That is a 6990 and the other a 680 (I suppose)

What we should get there? That a single-vga is shorter than a dual?

 

[boxleitnerb]

Thats about 7970@1100 territory. Very nice.

 

[tviceman]

This thread has gone downhill since the leaks started coming out. I think everyone needs to calm down until reviews by more popular sites (anandtech, hardocp, techpowerup, techreport, hardwareheaven, hardwarecanuks) get their official reviews online.

 

[Man from Atlantis]

http://www.facebook.com/photo.php?f...56145582074.250887.69170057074&type=3&theater

http://www.xtremesystems.org/forums/showthread.php?277763-%E2%80%9CKepler%E2%80%9D-Nvidia-GeForce-GTX-780&p=5071974&viewfull=1#post5071974

GT2 result is broken probably 2xSLi