First Images of Nvidia GeForce “Fermi” Show Up

Nvidia G300/NV60 Graphics Card Pictured, Examined

by Anton Shilov
09/30/2009 | 10:37 PM

Nvidia Corp. has demonstrated one of the first products based on its newly-announced Fermi architecture. The first card, which pictures appeared on Japanese PC Watch web-site, based on the graphics processing unit (GPU) known as G300, GT300 or NV60 is apparently not a GeForce, but Tesla, which is designer for high-performance computing or simulation market.

 

Nvidia Tesla Fermi-G300 card, which is intended for massively parallel computing and not video games, does feature DVI-I output, unlike today’s Tesla cards. An interesting thing to point out is that the shown-off Tesla card does not seem to feature Nvidia IO chip, which contains display controllers, which means that either the new flagship GPU features integrated TMDS transmitters, or the DVI-I output does not work.


Nvidia graphics card powered by Fermi-G300 graphics processor. Image by PC Watch

The demonstrated unit has surprisingly short print-circuit board (PCB), which is a good news for those, who use smaller computer cases and cannot fit longer add-in cards. The novelty has rather small cooling fan, which is natural for Tesla products, but a six-pin (which has neither pins nor soldering points) and an eight-pin (which soldering points are not placed correctly, which may imply that the connector also does not work) PCI Express power connectors, which means that the board should be a rather power-hungry one. The cooling system itself looks rather flashy with a carbon-fiber like inset, which should probably imply on fast sport cars.


Nvidia graphics card powered by Fermi-G300 graphics processor. Image by PC Watch

The Tesla add-in-card based on a Fermi graphics processor demonstrated by Nvidia is a rather strange one, in fact. Even though the card has two mounting places for DVI-I connectors, it sports only one connector and the bracket does not allow to install the second one. Besides, the PCB seems to be rather simplistic, which is a surprise, considering the fact that the GT300 graphics processor has 384-bit memory bus. Even though the board features two MIO connectors for SLI multi-GPU setups, the second connector cannot be utilized due to presence of a large heat-spreader on the backside of the card. In general, the shown-off add-in-card may be a very early sample of a Tesla/GeForce product that is intended for testing internally, but not for sale.


Nvidia G300/NV60 graphics processing unit based on Fermi architecture. Image by PC Watch

Nvidia Fermi-G300 graphics chip resembles all top-of-the-range graphics processors from the company in the last three years: it is covered by an aluminum heat-spreader. The A1 revision chip was made on the 35th week of 2009, in late August. Typically, Nvidia ships A2 revisions of the chip on commercial products.


Die shot of Nvidia G300 graphics processing unit based on Fermi architecture. Image by PC Watch