Just hours before Nvidia decided to share the Fermi, GT300 specification with the world, we managed to get most of the spec and share it with you.

The chip has three billion transistors, a 384-bit memory interface and 512 shader cores, something that Nvidia plans to rename to Cuda cores.

The chip is made of clusters so the slower iterations with less shaders should not be that hard to spin off. Each cluster of the chip has 32 Cuda cores meaning that Fermi has 16 clusters.

Another innovation is 1MB of L1 cache memory divided into 16KB Cache - Shared Memory as well as 768KB L2 of unified cache memory. This is something that many CPUs have today and as of this day Nvidia GPUs will go down that road.

The chip supports GDDR5 memory and up to 6 GB of it, depending on the configuration as well as Half Speed IEEE 754 Double Precision. We also heard that the chip can execute C++ code directly and, of course, it's full DirectX 11 capable.

Nvidia’s next gen GPU is surely starting to look like a CPU. Looks like Nvidia is doing reverse Fusion and every generation they add some CPU parts on their GPU designs.