Rather than trim back the GK104 found on the GTX680/670, the GTX650 features the same GK106 core found in the GTX660, a totally new processing unit.
The GK106 core still retains many of the features found on the flagship GK104 processor. SMX performance features are still there offering more shader, texture and processing performance while consuming less power than previous DX11 GPU's. Indeed, NVIDIA claim that the new Kepler core is capable of delivering twice the performance/watt ratio than the 1st gen DX11 graphics cards!
The new GTX650 contains four of these SMX units which gives us a total of 768 CUDA cores and 64 texture units. The GTX650 ships with a 128-bit memory interface spread over two 64-bit memory controllers, which allows NVIDIA to support mixed density memory modules (2GB instead of 1.5GB). The reference design ships with just 1GB of GDDR5 however, our GIGABYTE edition has twice this amount.
Missing Feature Sadly, the GPU Boost feature found in higher end GeForce offerings is disabled so overclocking the card, at least at the time of this review was not available. This is discussed more in the 'Overclocking' section of the review.
GIGABYTE GTX650 Ti Specification
Graphics Processing Clusters 3 SMXs 4 CUDA Cores 768 Texture Units 64 ROP Units 16 Base Clock 1033 MHz Boost Clock NA Memory Clock (Data rate)6200 MHz L2 Cache Size 256KB Video Memory 2048MB GDDR5 Memory Interface 128-bit Total Memory Bandwidth 99.2 GB/s Fabrication Process 28 nm Transistor Count 2.54 Billion Connectors 2 x Dual-Link 2 x HDMI Form Factor Dual Slot Power Connectors 1 x 6-pin Recommended Power Supply 400 Watts Thermal Design Power 110 Watts Thermal Threshold 98° C