On March 22nd, Nvidia introduced its first "Kepler" -based graphics card. The card, known as the GeForce GTX 680, is powered by a 28nm GPU, codenamed GK104, which crams 3540 million transistors into a 294mm2 chip. This is an improvement over the 40nm GTX 580, which contains 540 million fewer transistors yet is almost three times the size, and in part underscores Kepler's overall goal: improved efficiency.
This refinement is visible in all aspects of the map, not least the raw performance. Based on the second generation of Fermi's Streaming Multiprocessor (SM) architecture, the GTX 580 has 512 CUDA cores, 48 ROP units (Raster Operations) and 64 TAU units (Texture Addressing Units). The GTX 680 increases the performance to 1536 CUDA cores, 128 TAUs and 32 ROPs and brings a lot of power into the race.
Made some clock speed changes while removing the shader clock entirely. Instead, everything now runs from the core clock, which now also contains a so-called "Boost Clock", a technology that is similar to Intel Turbo Boost. By default, the GTX 680 is clocked at 1006 MHz – already 30% higher than the reference GTX 580 – with a dynamically changing boost clock of 1058 MHz. In the meantime, the GDDR5 memory frequency has increased by 50% from 4008 MHz to 6008 MHz.
Since the Radeon HD 7970 is only 9% faster on average than the GTX 580, AMD has thrown itself into a corner with its flagship, which was originally set at $ 549. However, the company was quick to respond and cut prices on much of its HD 7000 series. The HD 7970 was priced at $ 479, just below the GTX 680's suggested price of $ 499.
We were surprised that AMD acted so quickly. Although the GTX 680 is faster than the HD 7970, the card from Nvidia is only available in limited quantities, if at all. At the time of writing, Newegg doesn't have a single card, and Amazon has low stock warnings on most GTX 680 models. While Nvidia is struggling to meet demand, the board partners have been busy making custom cards. Since we didn't receive a reference sample last month, our GTX 680 review will instead feature one of the Special Edition products, namely the Gainward GTX 680 Phantom …
GTX 680 Phantom in detail
A week after the GTX 680 was released, Gainward unveiled its Elite Phantom Edition card, which touts a redesigned circuit board with improved performance phase, factory overclocking, and a massive three-slot cooler – the last of which is the most notable change. Although Gainward showcased its phantom cooler on some GTX 500 series cards, the GTX 680 is the first to launch the company's second generation solution.
The Phantom II has the same elegant design as its predecessor, but offers better thermals with lower noise levels and a more stable construction. It's unlike any three-slot cooler we've seen so far in that its heat sink has five 6mm heat pipes that draw heat from the base and distribute it evenly across the heat sink. The most unusual part of this design is the fans, or rather their position.
While fans are usually attached to the top of the heat sink, Gainward has instead embedded two extremely quiet 80mm PWM fans directly into the heat sink. The company claims that its factory overclocked GTX 680 is six degrees cooler and 11.5 dB quieter than a reference GeForce GTX 680. This is particularly impressive when you consider the efficiency gains Nvidia has made with the standard design.
The heat sink is 225 mm long, 70 mm wide and 40 mm high (8.85 "x 2.75" x 1.57 "). It has a black fan cover that forces the 80mm fans to suck in air through the fins above and at the same time push it over the fins below. A black aluminum heat spreader moves past the heat sink, which surrounds the top of the card and cools the eight 256 MB GDDR5 memory chips together with the 6-phase PWM.
By using a 6-phase design, Gainward includes two additional phases to power the GPU, which is designed to improve performance under heavy loads and improve the cards' overclocking ability. Not only does this card improve its overclocking potential, it also improves its efficiency while reducing choke and EMI noise as well.
Speaking of overclocking, Gainward did a little of the heavy lifting by increasing the core clock from 1006 MHz to 1084 MHz, which is a slight 8% increase, while increasing the boost clock from 1058 MHz to 1150 MHz which is corresponds to an increase of 9%. The GDDR5 operating frequency has also increased by 5% from 6008 MHz to 6300 MHz, which means that the memory bandwidth has increased by 5% from 192.2 GB / s to 201.6 GB / s.
The rest of the Gainward card remains standard, including a pair of SLI connectors, 6-pin and 8-pin PCIe power connectors, and an I / O panel configuration made up of HDMI, DisplayPort, and two DVI connectors .