As a GPU enthusiast, you are probably waiting for the release of the NVIDIA GeForce GTX 680, which is based on a new architecture codenamed Kepler. Although it has already received its fair share of leaks, a real-life photo of it has appeared on a Chinese website, along with a Powerpoint slide, reportedly from NVIDIA, that confirms or details much coveted specifications which are fairly impressive on paper: the number of CUDA cores jumps from 512 (GTX 580) to 1536 (GTX 680). That’s quite a jump, but we will need to see how the if the “performance per core” has evolved since the GeForce GTX 580. My guess is that we’re not going to see a 3X performance at all, but having more cores, could be beneficial to power-efficiency, if each core has to “sweat” less. Apparently, NVIDIA has also added an automatic overclocking feature similar to what is found in CPUs today.
And the good news is that the power efficiency seems to have been improved, although the information out there doesn’t seem very reliable yet. For instance, some media are talking about a drop from 244W (the GTX 580′s max power draw) down to 195W. However, the 195W number is not the “maximum graphics card power” of the GeForce GTX 680, it’s the TDP, or Thermal Design Power, which is the power that the cooling system has been designed to dissipate. It is really comparing two different metrics. Of course, TDP tends to be an interesting indicator, but take it with a grain of salt for now.
From the leaked benchmarks, it seems that NVIDIA positions this new high-end graphics card as having between 10%-40% performance above AMD’s best, but we don’t know in which test conditions, so I would recommend waiting for independent benchmarks. In the meantime, NVIDIA has reportedly cut the GeForce 580 GTX pricing, so it’s safe to assume that the GeForce GTX 680 should show up soon. A GeForce 670Ti is also rumored to be in the works, and “Ti” typically means that NVIDIA thinks that it is a “perfect product” in terms of performance/price, so stay tuned.