GeForce GTX 750 Ti

At the heart of the GTX 750 Ti is the GM107 GPU. The number of GM107 compute units is significantly larger than that of the GK107: 640 CUDA cores and 40 texture units instead of 384 and 32, respectively. However, the GM107 still has 16 ROPs and a 128-bit memory bus. By themselves, these numbers are not surprising, if not for the fact that the GM107 remained within the same power consumption range as the GK107 (the declared TDP is even slightly less - 60 vs. 64 watts). In fact, this is the purpose of the Maxwell architecture - to increase performance per watt.

Judging by the presented diagrams, the main innovation of Maxwell is a seemingly insignificant, but potentially very effective reorganization of GPU blocks. At the global level, GM107 follows the principles laid down by Kepler. All computational logic is concentrated in a structure called the Graphics Processing Cluster (GPC), which is only one in the GM107, although there will be more of them in larger chips.

The entire back-end of the processor is located outside the GPC in the form of ROP blocks and two 64-bit memory controllers, as well as the Giga Thread Engine, which performs the functions of context change, simultaneous execution of kernels and distribution of load flows between the GPC (the latter is not yet our case) . The quantitative difference from Kepler here is the L2 cache, increased from 256 to 2048 KB, which should compensate for the narrow 128-bit bus, as well as reduce the power consumption for transactions with a very voracious memory, which is GDDR5 SDRAM.

Specifications GeForce GTX 750 Ti

The base frequency is 1020 MHz. Since it uses GPU Boost 2.0 technology, there is also a Boost Clock equal to 1085 MHz. Video memory operates at an effective frequency of 5400 MHz. Thus, in terms of aggregate characteristics, the GTX 750 Ti is close to the GTX 650 Ti based on the GK106 chip, even without taking into account the more efficient Maxwell architecture.