Max power and low costs

Last week, Nvidia launched its first graphics card based on the new Maxwell architecture, the GeForce GTX 750 Ti. Later this month, it will be followed by a slower sibling, the GTX 750.

Maxwell succeeds Kepler, which was used in the GTX 700 series and most of the GTX 600 series graphics cards we’ve been seeing for the last two years from Nvidia.

So, Nvidia’s Maxwell has become their Haswell in a superficial way. Maxwell was built with mobile in mind. Efficiency is what Nvidia is shouting about, and yes, it’s a very efficient architecture. Nvidia’s basically tweaked Kepler by reorganising the same basic structure, making sure everything’s utilised better.

The basic building block of a GPU is a streaming multiprocessor, or SM. The Kepler SMs were called SMXes, and the Maxwell ones are called SMMs. I suppose they’ve done this to denote a change in layout. Anyway, with Fermi, each SM consisted of 32 ALUs (GF114 got 48, though) that would process one execution thread. Nvidia calls these ALUs “CUDA cores” but that’s a marketing term, if anything. The closest analogue to a CPU core in a GPU is probably the SM itself. Kepler basically made an SM bigger, with 192 ALUs and more of the other units, and they called it an SMX. Now, you can’t make a direct performance comparison with an SM, because of the way Nvidia handled clocks for both architectures. Maxwell reorganises this SMX, and divides it into four sub groupings of four units, each with 32 ALUs. So, a Maxwell SMM contains 128 ALUs, other units remaining roughly the same.

However, instead of launching Maxwell with a high-end GPU like the GK110 or GK104 (GTX Titan/780/780 Ti and GTX 770/680/690/670/660 Ti respectively), it’s decided to go with a bottom up approach, with the lightest two versions of Maxwell launching first: the GTX 750 Ti and the GTX 750.

But the GM107 is no slouch, and competes readily with the more mid-range GK106, found in the GTX 660 and 650 Ti. The 750 Ti and 750 are even priced similar to the 650 and 650 Ti (naming would suggest that, I guess, however, the 650 and its Ti aren’t direct siblings).

Performance wise, the 750 Ti comes in a bit lower than AMD’s R7-265, while the 750 comes just under the R7-260X. Both cards soundly beat the GTX 650 and are easily beaten by the GTX 650 Ti boost.

However, the main attraction of these GM107 based cards is their low power consumption numbers. Rated with a TDP of 60w, the 750 Ti offers at least 88 per cent of the performance of the R7-265 (150w TDP) with around 82 per cent of the power consumption, Or 111 per cent of the R7-260X’s performance with 84 per cent of the power consumption.

The 750 Ti and 750 are great upgrades if you have a low power PSU, since they draw all their power from the motherboard’s PCIe connector. The cards are small, and their low power footprint makes them great upgrades for small form factor configurations, not to mention that Dell desktop you bought that could use a dedicated GPU.