Nvidia Quadro K2000M

That doc actually says Fermi on p 2, doesn't it?

 

I think it's based on Kepler. 384 fermi cores and 55W makes no sense.

 

yes it says Fermi in the document.
Kepler makes no sense as the 104 core is crap for GPGPU and im thinking we need to wait fir the next chip

dont think so the nvidia document linked above says its fermi on page 2 and not using kepler makes sense as fermi is faster in OpenCL, and FP64. kepler got neutered to compete in gaming and sacrificed GPGPU functions

 

KCETech1 - Yeah I pointed out earlier in the thread that it said Fermi on that doc too. I was wondering how Syllogistic deduced that the other document confirmed that the gpu was Kepler.

Regardless, I'm pretty sure Nvidia said Quadro Kepler wasn't coming until next spring or something...

 

As others have said, it just seemed to make more sense for it to be Kepler given the new generation, "K", TDP vs CUDA cores etc, but it seems like it would be a good thing if I was proven wrong so I hope that's the case!

 

So here (Kepler GPU hard launch now expected in April

"The company originally scheduled GeForce Kepler for the launch in the first quarter (i.e. fiscal fourth quarter 2012), Quadro Kepler was scheduled for the second quarter (i.e. fiscal first quarter 2013) and Tesla Kepler was scheduled for the third quarter (second fiscal quarter 2013)."

 

I want to come and help this discussion since I have one last hope for Nvidia on doing something proper. They have disappointed me with the 660M with the minimal increase over 560M while keeping the same TDP.

From the 2nd document posted here, I believe we are dealing with a Kepler based Quadro, coming from the big, GK110 Tesla Die. A 384 shader Fermi, to fit in 55W, would mean really really low clock speeds. Somewhere in the realm of 300-400 Mhz. Performance would be downright bad.

So, after fiddling with the big Tesla Die, I believe this is the most probable outcome of the K2000M:


Although it says Pci-Ex 2 in the Nvidia Paper, I didn't manage to find a proper header like this one with the Pci-Ex 3. I did my best with the limited functionality of Paint. The small yellow blocks would be the Double Precision Units. It appears Nvidia went on the route of making one architecture for gaming, and one for computing. If what I assume is true, they would keep those units in the mobile Quadros and not give us a gaming GK107 with some optimized drivers.

On a more pessimistic closing note, Why did they not cram 768 shaders into 75W TDP? They managed to squeeze 1344 in the realm of 100+W TDP, the 680M.

 

There is no FP64 precission support in K2000M. Look at this table:
http://www.nvidia.com/content/PDF/product-comparison/Quadro_Mobile_Product_Comparison.pdf

 

exactly, ONLY the older 5010

 

Well, if that holds true and there are no double precision units, we shall receive the gaming GK107 and GK104 as Quadro's. Now that future is really grim for mobile computing