Tesla Kepler Features Comparison
This training note explains the features of 4 Tesla cards based on the Kepler generation of Nvidia GPU technology.
- Kepler is the 3rd generation of Tesla cards for General Purpose parallel computing. Tesla is 1st, Fermi is 2nd, Kepler is 3rd, and Maxwell is 4th.
- Nvidia released the first Maxwell card GTX750Ti in February 2014 which is obviously in the GTX series for the gaming market. There is no Quadro or Tesla cards yet based on Maxwell in August 2014.
- Tesla cards are for computation purpose and their visual display capabilities are de-emphasized and ports are removed on many models. Computation capabilities are expressed in FLOPS for floating point operations per second in SP or DP. The precision of floating point is defined in IEEE754. Briefly SP allows up to 8 significant digits.
- Tesla K10 is made up of 2 GPU chips on one card and without 2 advanced features as shown in the table. It is suitable for applications on SP and without involving the advanced features. Also each chip has 4GB memory only. Nvidia suggested seismic, image, signal processing, and video analytics.
- The table missed one important data- the PCIe version. K10 and K40 are based on PCIe V3 and the other two are based on PCIe V2.
- Tesla K40 has the largest memory of 12GB and as such it is best for big scientific data modelling.