Review: No reason to shy away from a reference board
The new GTX 780 is a pretty impressive card even in its plain vanilla reference edition and only a handful of Nvidia partners thought they could do better. Small wander than that Gainward stuck with the reference design and has yet to introduce another version, although it still wants to offer a few perks.
We found out that it is already working on a new cooler design, but let’s keep it hush-hush. For now all we can do is test Gainward’s reference GTX 780, but like we said this is not necessarily a bad thing. The card features an all-reference PCB, with reference clocks and it comes bundled with Gainward’s ExperTool software.
Although the GT110 is based on the Kepler architecture of GTX 680 fame, the GTX 780 ends up significantly faster than Nvidia’s top GPU of 2012.
Like we said in previous reviews, the new GTX 780 is basically a watered down Titan. The GTX 780 has 12 active streaming multi-processors (SMX) out of a total of 15 SMX blocks in the GK110. It has 2304 CUDA cores and even the Titan isn’t the full GK110, as it has 14 active SMX blocks or 2688 CUDA cores. The GK104 chip used in the GTX 680 has “just” 1536 CUDA cores.
There’s more memory bandwidth, too. The GTX 780 has a 384-bus and the frame buffer was upped from 2048MB to 3072MB. Nvidia also made it possible for partners to come up with custom cards with 6GB of GDDR5 memory. Most users don’t even have that much RAM in their systems, but high-end graphics cards aren’t intended for the average consumer, they go after connoisseurs who don’t mind spending a bit more for a proper gaming experience.
The GTX 780 can draw as much as 250W of power, whereas the GTX 680 is happy with just 195W. The card requires both an 8-pin and a 6-pin PCIe power connector for proper operation. NVIDIA suggests a 600W or greater power supply for the GTX 780, although the recommendation sounds rather pointless - we doubt anyone getting a GTX 780 will have a weaker PSU in their rig.