NVIDIA released their next-gen GPU tech today, codenamed Kepler. Both notebook and desktop graphics received a refresh, and by all accounts, the chips are more than delivering. Boutique suppliers, like AVADirect & ORIGIN PC, are already offering pre-built systems with the new cards.
NVIDIA is making bold claims with the GTX 680, calling it “the fastest, most efficient GPU [they’ve] ever built.”
The green machine promises that the new revamped GPU architecture, building upon the 4XX series Fermi chips, can deliver anywhere between 10 and 40% increased performance over the competition (competition, in this case, comes in the form of AMD’s Radeon HD 7970). What’s just as interesting, however, is that NVIDIA isn’t only touting the raw numbers, like GPU manufacturers did in the past – they’re also crowing over the performance per watt.
Performance per watt is some measure of computational performance achieved while drawing one watt of electricity. The TDP of the GTX 680 sits at 195W, while the HD7970 is over 25% higher, at 250W. As a result, while the latter requires one 6-pin and one 8-pin power connector, the GTX 680 can get by with a pair of 6-pin cables.
Raw gaming ability isn’t the only improvement to the new 680; NVIDIA engineered the card to be quieter, thanks to in-house engineered heatsink and fan/fan blade improvements. The company swears an average 10dB reduction in noise over the HD7970.
There’s also a new hardware H.264 video encoding engine, known as NVENC. This is a brilliant move on behalf on NVIDIA, since H.264 has become the defacto standard for video storage, both on and off the web. NVIDIA first revolutionized video encoding improvements several years ago with their GPGPU-accelerated CUDA standards; since H.264 is so pervasive, adding dedicated circuitry for that task alone can improve performance even more.
In fact, the new GeForce GTX 680 can encode H.264 video at a rate of almost 4x the previous CUDA times, all while using less power. Video enthusiasts, take note.
The other interesting advance in the new GPU is something called “GPU Boost.” Remember Intel’s Turbo Boost? Or AMD’s clone, Turbo Core? It’s the same thing. When the GPU has underutilized power at its disposal, it can selectively overclock the card from its base 1006MHz clock speed. On average, NVIDIA reports that when GPU Boost is invoked by the card, it can boost performance by 5%.