Introduction to the first 55nm consumer GT200

If it seems like it has been forever since NVIDIA has released an extremely high-end consumer graphics card, you'd be partially correct.  The last release that NVIDIA aimed for the flagship spot with was the GeForce GTX 280 based on the GT200 architecture back in mid-June.  Six months might seem like an eternity compared to some previous GPU release cycles and with the AMD Radeon HD 4870 X2 2GB firmly taking the performance crown back in mid-July, NVIDIA has actually been off the top spot for enthusiast graphics for quite a while. 

NVIDIA is hoping that all of that will change with the New Year as a new flagship graphics card makes it way to the scene in the form of the GeForce GTX 295.  In the same lineage as the GeForce 9800 GX2 and 7950 GX2, this new card is in fact utilizing a pair of GT200-based GPUs on two PCBs all contained on a single dual-slot card.  The core of these GPUs is based on the GT200 architecture of course, though these new parts are first consumer cards to see the new 55nm cores on them.  (The underlying technology is the same though - see my first GT200 review for the full break down on this GPU's inner-workings.) This was no doubt a requirement as getting a pair of the very hot and power hungry 65nm GT200s was a difficult task. 

The specs on the new GTX 295 are impressive:

• 480 stream processors (240 per core)
• 576 MHz core clock
• 1242 MHz shader clock
• 1000 / 2000 MHz GDDR3 memory clock
• 1792MB total frame buffer (896MB per core)
• 448-bit memory interface per core
• 56 total ROPs (28 per core)
• 160 texture units (80 per core)

What is interesting about these specs is the mix between GTX 260 and GTX 280 heritage.  The 480 stream processors indicate that the GTX 295 is basically a pair of GTX 280 GPUs (since the GTX 260s have either 192 SPs or 216 SPs depending on your place in time) on a single board while the clock rates are exactly like those reference clocks on the GTX 260 GPU.  Also, the 448-bit memory bus and 896MB of memory per core are also indicative of the GTX 260 product.  NVIDIA has obviously created a hybrid part here that properly balances GPU performance and power efficiency.

Let's take a look at the physical product and see what else we can learn about it.

The NVIDA GeForce GTX 295 1792MB

The new GeForce GTX 295 is a heavy, beefy graphics card; it even outweighs the previous champ the Radeon HD 4870 X2 which is quite a feat.  Because NVIDIA's dual-GPU graphics boards require a unique design, with two physical PCBs and a cooler sandwiched between them, the design of the GTX 295 is completely new yet very familiar.

The card has that very familiar NVIDIA "look" that has been around since the introduction of the GeForce 9800 GX2 though this time the texture of the card is almost rubbery thanks to a "soft touch" matte paint NVIDIA chose.

The cooling fan is exposed on both the front and rear of the card to allow air on both sides to come inside the PCBs and to be pushed across both GPUs held inside.

The heatsink that rests between the two PCBs is responsible for cooling a pair of 55nm GT200 GPUs and thus is HAS to be efficient at what it does.  These aluminum fins along the top of the card are partly responsible for that.

The GTX 295 requires both an 8-pin and a 6-pin PCIe power connection to operate - this is actually a pretty good feat in terms of power efficiency considering the 65nm GT200 GTX 280 cards EACH required an 8-pin and 6-pin connector.  My only gripe here is that with the two power connectors facing each other removing the PCIe plugs after installing them is difficult unless you have tiny hands or want to get needle nose pliers involved. 

You can also see the digital audio input next to the 6-pin plug for passing audio through the included HDMI output.

Here on the back of the card we can see the single SLI connector on the GeForce GTX 295 that allows the GPU to support Quad SLI by adding in a second card.  Yum!

Looking the at the connectors on the card we see that NVIDIA has included two dual-link DVI outputs (both of which support 2560x1600 resolutions) and a single HDMI output for those interested in creating a truly kick-ass home theater / gaming rig.  There are also two lights to the left of the HDMI and DVI ports; one indicates that proper power is plugged into the card and the other indicates that at least one display is correctly attached.

In terms of size, the GeForce GTXC 295 1792MB card is the slightly shorter than the Radeon HD 4870 X2 but is the same length as the GTX 280 and 9800 GX2. 

 

You can see here how similar the GTX 295 (top) and 9800 GX2 designs really are.

In this explosion diagram provided by NVIDIA (taking this apart involves a bit more work than normal)
you can see that each GPU has its own separate PCB that is connected via a ribbon cable along the bottom of the card that acts as the physical SLI bridge.  The double sided heatsink between the two GPUs is pretty weighty as you would expect and allows for the air from the fan to be passed from the back of the card out the exhaust.  There is definitely a LOT of technology crammed in a small space with the GeForce GTX 295.