On February 18th, Nvidia introduced the GeForce GTX 750 and GTX 750 Ti, two affordable new video cards. What makes this introduction particularly interesting is that both GTX 750 variants are the first to feature a chip from the new Maxwell generation. This made us especially curious about their performance.
Nvidia promises that Maxwell will deliver an enormous increase in efficiency and claims that the GM107 chip, on which the new cards are based, is a particularly good fit for modern PCs, such as laptops and small Mini-ITX systems. In this review, we will discuss the new Maxwell architecture and extensively study the performance of these two new video cards.
That a new generation called Maxwell was going to succeed the Fermi generation (GTX 400/500 series) and the Kepler generation (GTX 600/700) will not have been news to those who have been following Nvidia's roadmaps for the last few years. However, the fact that Maxwell makes its first appearance in two affordable cards such as the GTX 750 and GTX 750 Ti is surprising, since Nvidia typically releases each new generation as a high-end GPU, initially. Only later would the new technology also become available to smaller, less complex chips for more affordable cards.
Nvidia's change of plans is almost certainly caused by TSMC, the Taiwanese company which produces Nvidia's chips. TSMC has been producing Nvidia's chips for years, and new generations of chips are typically introduced together with a new process node from TSMC: Fermi was 40nm, Kepler was 28nm, and it's hardly a secret that Maxwell chips were meant to be produced on a 20nm node. However, TSMC's 20nm process node has seen substantial delays. According to the original planning, it should have been ready a long time ago, but the opposite is true. Nvidia, being one of TSMC's largest customers, probably knows more about what's going on at TSMC than we do, but to us it isn't clear when the new 20nm process node will be ready for production.
Still wanting to put a competitive new product on the market, Nvidia must have decided to release at least one Maxwell design on the existing 28nm process node. They decided to use the GM107 chip, which is now at the core of the GTX 750 and GTX 750 Ti. Nvidia stresses that the new cards are based on the 1st generation Maxwell architecture, which implies that we can also expect a 2nd generation, although the documentation currently does not mention anything about 20nm at all.
Despite the fact that the GM107 chip of the GTX 750 and GTX 750 Ti is being produced with exactly the same type of transistors as the GPUs belonging to the GTX 600 and 700 series, Nvidia still promises extreme improvements in efficiency, that is to say, the power to performance ratio. For example, the GTX 750 should offer twice the performance of the GT 640, despite having an almost identical TDP. For an even more striking promise: the GTX 750 Ti's performance should be roughly equivalent to the performance of 2010's flagship model, the GTX 480, which was far more expensive and used way more electricity.
In this review, we will discuss the properties of the new architecture, followed by the specifications of the GTX 750 and 750 Ti cards. The new GeForce GTX 750 and GTX 750 Ti have an MSRP $119 and $149, respectively. Naturally, we tested both of them extensively. Nvidia sent us a reference design of the GTX 750 Ti, while MSI sent us their special Gaming edition of both the GTX 750 and 750 Ti. Similarly, ASUS sent us the OC edition of both the GTX 750 and 750 Ti.