GeForce GTX 580 might sacrifice HPC functionality

The colleagues over at Nordic Hardware heard NVIDIA sacrificed the high-performance computing (HPC) functionality of Fermi to make the GeForce GTX 580 a smaller, more efficient card that's more suitable for the retail market. If the rumors are right, the card uses a 40nm GF110 GPU with 512 CUDA cores but has 300 million transistors less than the GeForce GTX 480 due to the canning of the HPC part. GPU clockspeed is said to be 775MHz, with 1544MHz shaders and 1536MB GDDR5 clocked at 4008MHz.

"According to sources it can be as many as 300 million transistors that NVIDIA has been able to cut in this way. The effect is that GF110 will be a GPU targetting only retail and will not be as efficient for GPGPU applications as the older siblings of the Fermi Tesla family. Something few users will care about.

The latest information that has reached us are said to be the final specifications for NVIDIA GeForce GTX 580. These were published at it168.com and strengthen the information we received earlier. There we can clearly see how NVIDIA beside processing units also raised the clock frequencies for both memory and GPU. This was made possible through a more mature circuit manufacturing where NVIDIA has managed to lower power consumption of GF100 with 10-15% without any physical alterations. This gave the comany more margin to work with."

The GeForce GTX 580 is expected to be announced somewhere next week, which is before the launch of AMD's Radeon HD 6900 series.