The most complex processor in the semiconductor history will reportedly have 7.1 billion transistors and will be complemented by up to an amazing 24 GB of GDDR5 video memory.
Nvidia’s Co-Founder and CEO, Jen-Hsun Huang recently confirmed that the GK110 will actually have just a “little” bit more than 7 billion transistors.
That’s 100 million more to be exact.
Compared with the previous complexity record holder, a FPGA processor from Xillinix that has more than 6 billion transistors, Nvidia’s GK110 is considerably more complex.
When compared with AMD’s Tahiti GPU with its 4.31 billion transistors, the GK110 is almost double the size.
It was obvious that such a complex chip must not be kept hungry for data so the logical move was towards a wider memory BUS.
Therefore the GK110 comes with a 384-bit memory BUS just like AMD’s Tahiti and the possible memory configurations range from 6 GB to 24 GB or VRAM.
A TESLA K20 card holding 12 GB of memory is the most likely configuration although 24 GB are possible and are also considered.
The rumored power consumption is around the 300 watts mark and that’s something Nvidia was able to deal with before so it’s likely that they’ll do it again.
As GPU compute becomes more and more popular, the market shifts the view and the TESLA cards are regarded as professional GPU solutions versus the old professional “hero”, the Quadro.
While there are many things discussed about the K20 TESLA card and there are even server contracts already planned with the K20, the GK110 Quadro is a strange silence.