First thoroughly understand what is GPU:
GPU Computing is the use of GPU (graphics processor) with the CPU to accelerate general purpose scientific and engineering applications. GPU computing first proposed five years ago by the NVIDIA Corporation, quickly became an industry standard, with millions of users worldwide, almost all computing providers are using GPU computing. GPU computing applications by the computationally intensive part to the GPU, and the remaining part of the program is still running on the CPU, which allows for unprecedented application performance. From the user's perspective, the application just runs a lot faster speed than before.
The second is a clear difference in NVIDIA TESLA series of other products
The current high-end graphics card NVIDIA Geforce, Quadro, Tesla three series of products, and they support the NVIDIA CUDA parallel computing platform. However, NVIDIA GeForce and NVIDIA Quadro are for the consumer and professional graphics and visual design, only the Tesla product line is entirely designed for parallel computing, provides a unique computing features. Since Tesla series of professional products, so it is certainly destined to use in related areas, such as: seismic processing, signal and image processing, video analysis of graphics computing high demand industries.
Tesla K10 based on two GK104 core, it can be seen as a variant of GTX 690 graphics card has 3072 CUDA cores, 2 * 256bit bit wide, single-precision floating point capability 4.58TFLOPS, the bandwidth of 320GB / s. Thanks to good performance than the GK104 architecture, Tesla family for the first time there have been dual-core card model.
characteristic | Tesla K20X | Tesla K20 | Tesla K10 |
Number and type of GPU | 1 Kepler GK110 | 2 Kepler GK104s | |
GPU computing applications | Seismic processing, computational fluid dynamics, computer-aided engineering, financial calculations, computational chemistry and physics, data analysis, satellite imagery, weather modeling | Seismic processing, signal and image processing, video analysis | |
Peak double precision floating point performance | 1.31 Tflops | 1.17 Tflops |
Gigaflops 190 is (every single GPU 95 Gflops) |
Peak single-precision floating point performance | 3.95 Tflops | 3.52 Tflops |
Gigaflops 4577 (and every single GPU 2288 Gflops) |
Memory bandwidth (ECC closed) | 250 GB / sec | 208 GB / sec |
320 GB / sec (every single GPU 160 GB / sec) |
Memory capacity (GDDR5) | 6 GB | 5 GB |
GB. 8 (every single GPU 4GB) |
CUDA cores | 2688 | 2496 |
3072 (and every single GPU 1536 Ge) |
* Note: In the case of enabled ECC, 12.5% of the GPU memory is used for ECC data bits. For example, in the case of ECC enabled, if the total capacity of 3 GB, then the user available memory capacity of 2.625 GB.
This website uses cookies (Learn more) and has a privacy policy (Learn more).