Inno3D GTX 470 Hawk Reviewccokeman -
» Discuss this article (6)
Inno3D's GTX 470 Hawk is based on the 40nm GT 100 Fermi architecture and comes equipped with 448 CUDA cores,14 Streaming Multi Processors, 56 Texture units and 40 ROP's. Like the GTX 480, GDDR5 memory is used for the GTX 470 you just get a lower capacity (1280 vs. 1536), with lower clock speeds of 837Mhz running through a narrower bus width of 320bits. One look at the Hawk's cooling solution tells you that this is not your standard GTX 470. This large cooling solution looks like it came straight out of the Arctic Cooling catalog with its three fans, heatpipe construction and massive fin array. This large cooling solution is built to tame the immense heat generated by the large die size. It will also take up in excess of two expansion slots, making it a 3 slot solution. Hope you have room!
Connectivity options are standard fare for the GTX 470 with two Dual Link DVI ports and a Mini HDMI port. The back end of the card is wide open for good airflow over the board and voltage regulation components.
Along the top spine of the card, you have the SLI bridge connections that allow up to four cards to be run in a quad SLI configuration, as long as you have a board that supports it. The power connections used on the GTX 470 Hawk are dual PCIe 6 pin connections. The GTX 470 has a maximum TDP of 215 watts, which is 35 less than its big brother, the GTX 480. The recommended power supply for the GTX 470 is at least 550 watts.
Pulling the Hawk apart is the only way to see exactly how the cooling solution is built and how it is implemented on the PCB. The heatsink is held on with a series of of ten screws that also hold on the aluminum heatsink for the memory, MOSFETS, and the VRM circuit. To get to the base of the assembly, you need first remove the main heat sink, the memory and finally the VRM heat sink. You are then left with the base board, with all of the components open for inspection.
The cooling solution used on the Hawk resembles something straight from Arctic Cooling, but I have been told this is an in-house design. This heatpipe equipped cooling solution uses a total of five copper heatpipes that feed into two separate aluminum fin arrays. The three fans used on this heat sink are PWM controlled to offer both silence and airflow. The maximum airflow through the heatsink is 35CFM at 29dbA. The fans run between 900 and 2000 RPM. As I mentioned before, the heat sink is large, but the actual measurements come in at a hefty 260x114x53mm at 561g. The base is shows some machining marks, but nothing that you can feel with a fingernail run across the surface. Inno3D states that this cooling solution will show a 24% reduction in load temperatures. But for the enthusiast, that just means more room to overclock! So will it allow higher clock speeds? We'll have to see.
The clock speeds employed by this card are 607Mhz on the Fixed function units and and 1281 Mhz on the CUDA cores. We are used to seeing amounts of memory that comes in sizes we are used to, such as 512MB,1GB, 2GB and more, but Nvidia has added a little more to the equation with a bump to 1280MB. Samsung GDDR5 memory is used for this implementation and is clocked at a mild 837MHz or 3348Mhz effective.
So we now know whether or not the Hawk is anything special. But to really prove that, it has to be tested.