The 7900 GX2
NVIDIA is has made significant strides in their press materials to ensure that we all know that this is a single card solution. NVIDIA are advertising that this is the fastest single card solution money can buy, despite the fact that this the GX2 is simply two PCBs and two GPU's connected together. Fortunately however, the GX2 comes in weighing significantly less than two 7900 GTX's, featuring but a single PCI-E power connector and two small coolers.
The two PCB's are both very simplistic and are comparitively, quite plain in relation to that of a 7900 GT or GTX. While it could be expected that there would be four DVI-D out ports, there are in fact only two, and as SLI supports only two monitors anyhow, this isn't such a large issue.
The 7950 features specs that are almost identical to the 7900 GX2 dual board that has been available for some time from system integrators. The 7950 GX2 is basically the production version of that board, and NVIDIA seem to have knocked 5 MHz off of the memory with the 7950 - hardly consquential. The main difference between a 7900 and a 7950 is in board length. While we don't have a 7900 GX2 to compare against, the shots of the boards in systems that have floated around the net recently is comparison enough. The shorter board length of the 7950 is a very welcome move.
In the press materials that NVIDIA sent out, they make sure we know that the GX2 is based not on two 7900 GTX chips, but uses instead two GX2 chips. The differences however between a GX2 chip and a 7900 GT chip are very minor, and are limited to clock rates.
|
7900 GTX |
7900 GT |
7900 GX2 |
Core Clock Rate |
700 MHz |
470 MHz |
500 MHz |
Memory Clock Speed |
800 MHz |
600 MHz |
600 MHz |
Memory Size |
512 MB |
256 MB |
512 MB x 2 |
Pixel Pipes |
24 |
24 |
24 x 2 |
Memory Interface |
256 bit |
256 bit |
256 bit x 2 |
Vertex Shaders |
8 |
8 |
8 x 2 |
Transistors |
278M | 278M | 278M x 2 |
ROP's |
16 |
16 |
16 x 2 |
Process |
90nm |
90nm |
90nm ...x2? :) |
Standard Full HDCP |
No |
No |
Yes |
While saying that this card has a 512 bit memory bus or 1 GB of memory would not be a lie, it's hardly the full extent either. The GX2 works almost identically to two 7900 series cards in SLI. As a result, saying that this card has a 512 bit memory bus or 1 GB of memory is like claiming the same of an SLI setup. Each graphics core only has access to its own 512 MB of memory via a single 256 bit bus. The same rules that apply to SLI apply here as well.
Because of the switch NVIDIA have incorporated into the GX2 design, not all motherboards will be able to interface properly with the two cards. NVIDIA are keeping an updated list of motherboards that are know to work with the GX2 without any issues.
BFG 7900 GX2
When we received our BFG branded 7900 GX2, we noticed that this card is conspicuously missing the famous OC label we have become accustomed to with past BFG cards. BFG made this decision because the safe thermal operating envelope for these cards is very small due to physical limitations. If BFG were to ship this with boosted clocks, they would likely face similar problems as some manufacturers are with 7900 cards currently.
BFG have packaged the GX2 in their basic video card boxing, and the card comes with the usual bundle which includes an XL sized T-shirt, their teflon mouse feet, a DVI-VGA dongle, an HDTV dongle, power connector, and driver CDs.
External Views
The GX2 is definitely a thing to behold. The card is the same length as a 7900 GTX, but actually appears slightly smaller, simply because the coolers used on the GX2 are quite small relative to the behemoth on the GTX.
Length-wise, the GX2 is exactly the same length as a 7900 GTX, while the GT comes in roughly an inch shorter. The copper reference cooler on the GT has been criticised for being too small and allowing the card to get too hot. Unfortunately, while the aluminum cooler on the 7900 GX2 is larger, the cooler on the bottom PCB gets very warm as it has very little breathing room. We definitely prefer the large, quiet and efficient cooler of the 7900 GTX but with very limited work space, NVIDIA would have been hard pressed to come up with a larger cooling solution.