On the week starting March 20th, NVIDIA will unveil GeForce products based on the Kepler GPU architecture. We got our hands on a board and saw very interesting solutions…
In the first of a series of articles, we'll analyze the way how NVIDIA will supply power for the GK104, rumored to debut as the "GeForce GTX 670 Ti". The specs are wildly rumored "on the Internets", but it is clear that this might be another winner for NVIDIA. Or… not.
The final verdict will be given by the reviewers and more importantly – the consumers. What we are going to focus on is another thing, and that is how efficient is the power design on the GeForce 600 Series boards and more importantly, the Kepler GPU itself.
For starters, the thing that you need to know about Kepler is that NVIDIA is using Foxconn to manufacture the high-end board. According to various sources, NVIDIA has used the strategy of switching between Flextronics and Foxconn in the past, but they're now perfectly happy with Foxconn (that one with the anti-suicide nets).
The board in question was manufactured in the third quarter of 2011, meaning that NVIDIA has had Kepler for quite some time. Given that the original plan called for Kepler debut in 2011 and Maxwell in 2013, it's obvious that the yields were too low to initiate launch production. In the recent financial analyst call, Jen-Hsun commented that the 28nm supply will be "constrained through the year (2012)".
As you can see on the diagram above, the board requires a single six-pin and eight-pin power cables (Theoretical Max: 75W + 150W + 75W from the PCIe connector = 300W) even though it should operate with just two six-pin rails. Then again, if you purchase a performance product, it is expected that you'll have at least a 500W+ PSU where 8-pin connector is the norm.
Three phases are dedicated to feed the GPU (NVVDD-PH2, PH4, PH5), and dual phase is for driving the rest of the board – a couple of gigabytes of GDDR5 memory and 3rd party chips (NVVDD-PH1 and PH3).
We'll continue the examination in our future articles. Stay tuned.