Nvidia lagged behind rival AMD in the introduction of a Direct X 11 GPU but it seems that the company is ready to make an impressive come back with the upcoming GT300 chip, which bloggers already describe as a "computational beast".
Nvidia's upcoming 40nm GPU codenamed "Fermi" or simpler "GT300" is slated for a Q4 2009 launch, meaning that it will most probably show up around the end of November.
Rumors about the new chip indicate that it will pack a lot of architectural changes compared to Nvidia's GT200 generation of GPUs.
Manufactured by TSMC, the 40nm chip is expected to pack three billion transistors and 16 Streaming Multiprocessor (former Shader Cluster). Each multiprocessor has 32 cores, with each core to be
able to execute an integer or a floating point instruction per clock per thread.
The chip also packs six 64-bit GDDR5 memory controllers for a grand total of 384-bit, matching AMD's current offering with the Radeon HD 5800 series.
Possible configurations of the new boards should include
1.5, 3.0 GB and 6GB of GDDR5 memory.
Of course, the GT300 gives direct hardware access for CUDA 3.0, DirectX 11, OpenGL 3.1 and OpenCL.
Nvidia should have been very busy with the design of the new architecture fir the GT300 chip. According to early reports, the chip will be more than three times more powerful than the GT200 GPU.
Nvidia has made an official statement on the state of the GT300 yields last week, after some stories online reported that the GT300 yields were bad, with only nine chips to work per wafer. An Nvidia' product manager said that the company's 40nm yields were "fine" and described the information as "baseless."
It seems that the complicated design of the new GT300 chip was the reason behind the "late" introduction of the chip in the market. AMD has already made a strong statement with the release of its Radeon HD5800 series but Nvidia believes that they can make its DirectX 11 GPU even faster.
All we have to do is to wait to see if the GT300 can outperform the HD 5870, HD 5850X2 and the HD 5870X2 chips, in a very interesting race in the GPU market.