Nvidia BigK GK110 Kepler Speculation Thread

[boxleitnerb]

(When) do we expect any leaks?

 

[Alexko]

Official reviews are tomorrow. So anywhere between then and now.

 

[trinibwoy]

With scalar code and repeating patterns, I am seeing up to 69,x% efficiency in GK104, GF110 was at 99, GF114 at 68 (max.).

So, dual issue seems to be the same as seen with GK104.

Hmmm thanks. Best I've seen is 66%. I'm sure we'll never get a straight answer.

 

[CarstenS]

I'll keep trying.

 

[UniversalTruth]

You mean Audi and Fiat?

No, I mean Lamborghini and Ferrari.

 

[Grall]

Italian cars are quite commonly designed by 3rd party design houses like Pininfarina etc. So it's not Audi (Volkswagen, really) or Fiat who design those vehicles anyway...

Btw. Nvidia Titan's cooler remind me of the dials of 1980s american TV sets. It has that tacky chrome look so common from a few decades past.

 

[CarstenS]

I am sorry to the utmost extent to have mentioned a car in this thread.

 

[Alexko]

I am sorry to the utmost extent to have mentioned a car in this thread.

In matters of computer hardware, beware of car analogies!

 

[DSC]

http://blogs.nvidia.com/2013/02/nvidia-to-build-a-new-home-20-years-after-our-founding/

 

[silent_guy]

http://blogs.nvidia.com/2013/02/nvidia-to-build-a-new-home-20-years-after-our-founding/

Using the Kepler 3D architecture?

 

[trinibwoy]

Nice building. I guess they found something to spend all that cash on.

 

[Homeles]

While Nvidia is off buying yachts and fancy buildings, AMD will pull a fast one and release a groundbreaking architecture that demolishes all Nvidia cards. Then Nvidia will go out of business, all because they spent money on a new office building instead of on R&D.

/AMDZone

 

[jimbo75]

Double the number of employees that they had 6 years ago (8k vs 4k).

 

[keldor314]

I'm pretty convinced that nVidia's SP flop numbers for Kepler are pure bullshit. A Fermi SM had sufficient register bandwidth for 64 MADs/clk. nVidia claims that a Kepler SMX has 2x the bandwidth so 128 MADs/clk makes sense. I have no idea where they're getting 192 from.

A Kepler SMX is drastically different from a Fermi SM. It runs instructions from 4 warps every cycle instead of two, and is two-way superscalar. This means it can in theory issue 8 instructions per cycle (=8x32=256 MADs), but only has 6 execution ports, giving you the 192 (=6x32) figure.

The bandwidth number is misleading in this regard since you also have to take into account the clock rate being lowered.

 

[keldor314]

I am sorry to the utmost extent to have mentioned a car in this thread.

At least you didn't mention a 'cuda.

 

[Man from Atlantis]

mobile.pcgameshardware.de/Crysis-3-PC-235317/Tests/Crysis-3-Test-CPU-Benchmark-1056578/

 

[DavidGraham]

Damn , that game is heavily CPU bound !

 

[jimbo75]

Damn , that game is heavily CPU bound !

I would say that it appears to be utilising all cpu resources.

 

[trinibwoy]

A Kepler SMX is drastically different from a Fermi SM. It runs instructions from 4 warps every cycle instead of two, and is two-way superscalar. This means it can in theory issue 8 instructions per cycle (=8x32=256 MADs), but only has 6 execution ports, giving you the 192 (=6x32) figure.

The bandwidth number is misleading in this regard since you also have to take into account the clock rate being lowered.

Eh? I know the marketing numbers. That's not what I'm asking about.

 

[DavidGraham]

I would say that it appears to be utilising all cpu resources.

Sadly this is a sign of a bad port , I can't think of a recent game that should such behavior than GTA IV , you simply had to run the game with a Quad core CPU to be decently playable, and worst of all the Graphics was mediocre at best.