NVIDIA GK110: The True 7-Billion Transistor Kepler GPU

NVIDIA GeForce GTX logo

Inside Kepler is a talk that will take place at the GTC 2012 (GPU Technology Conference, May 14-17, 2012, San Jose, California). On the presentation of this talk, NVIDIA refers to a new 7-billion transistor GPU:

In this talk, individuals from the GPU architecture and CUDA software groups will dive into the features of the compute architecture for Kepler – NVIDIA’s new 7-billion transistor GPU. From the reorganized processing cores with new instructions and processing capabilities, to an improved memory system with faster atomic processing and low-overhead ECC, we will explore how the Kepler GPU achieves world leading performance and efficiency, and how it enables wholly new types of parallel problems to be solved.

The new 7-billion transistor GPU is probably the full implementation of NVIDIA’s Kepler GK110, the true successor of the GF100 (GTX 480) / GF110 (GTX 580). The GeForce GTX 680 (based on GK104, with only 3.5 billion transistor) is actually the successor of GeForce GTX 460 / GTX 560 (GF104 / GF114). The GK110 will be a monster in GPU computing, which is the weakness of the current GK104 GPU.

What will be the name of the GK110 based graphics card? Maybe the GTX 685 or GTX 780. The GTX 690 is already reserved for the dual GK104 board that should be announced shortly.

According to this thread, the GK110 will have 3072 CUDA Cores (twice the number of cores of the GK104 GPU).


4 thoughts on “NVIDIA GK110: The True 7-Billion Transistor Kepler GPU”

  1. Promilus

    @JeGX – we’ll have to wait to see that in action. If it has the very same microarch. as GK104 just with more SMX it might easily dethrone Tahiti from GPGPU pedestal, however it’s still a step back in field of efficiency. No hot clock, much more ALUs per scheduler… that is exactly what AMD dumped (because it’s easy to design and implement, can work great with games but doesn’t with generic compute, also is highly dependent of compiler performance)

Comments are closed.