GEFORCE GT 640
GEFORCE GT 640 is the first model based on the Kepler architecture, which instead of GPU The GK104 used the GK107. The GK107 chip, with 1,3 billion transistors, has a surface area of 118 mm², which is significantly smaller than the GK104 with 294 mm2. In essence, the GK107 chip of the GeForce GT 640 uses a quarter of the shader resources of a full-featured GPU GK104. 384 CUDA cores are divided into two SMX blocks, each containing 16 texture units (32 in total). The core frequency of the new chip is 900 MHz. GeForce GT 640 cannot dynamically increase the frequency when the thermal reserve allows it because the card does not have the function GPU Boost inherent to cards with GK104.
Filling GPU consists of two groups of rasterization units, each capable of outputting eight 32-bit pixels per clock cycle, for a total of 16. Two 64-bit memory buses together make up 128-bit. 2 GB of DDR3 memory operates at 891 MHz, resulting in a memory bandwidth of 28,5 GB/s.
Characteristics of GeForce GT 640
| Name | GeForce GT 640 |
| Core | GK107 |
| Process technology (µm) | 0.028 |
| Transistors (millions) | 1300 |
| Core frequency | 900 |
| Memory operating frequency (DDR) | 891 (1782) |
| Bus and memory type | 128-bit GDDR3 |
| Bandwidth (Gb/s) | 28.5 |
| Unified shader units | 384 |
| Unified shader unit frequency | 900 |
| TMU on conveyor | 32 |
| ROP | 16 |
| DirectX | 11.1 |
| Memory Capacity | 2048 |
| Interface | PCI-E 3.0 |
The power consumption of the card, according to Nvidia, is 65 W, which does not exceed the limit of the 16-lane PCI Express slot, which can supply power to a load of up to 75 W.
