Presentation is loading. Please wait.

Presentation is loading. Please wait.

Performance modeling in GPGPU computing Wenjing xu Professor: Dr.Box.

Similar presentations


Presentation on theme: "Performance modeling in GPGPU computing Wenjing xu Professor: Dr.Box."— Presentation transcript:

1 Performance modeling in GPGPU computing Wenjing xu Professor: Dr.Box

2  GPU-accelerated computing is the use of a graphics processing unit (GPU) together with a CPU to accelerate scientific, engineering, and enterprise applications. What’s GPGPU?

3  a simplified representation of a system or phenomenon  it is the most explicit way in which to describe a system or phenomenon  use the parameter we set to build formula to Analysis system What’s modeling

4  Hong and Kim [3] introduce two metrics, Memory Warp Parallelism (MWP) and Computation Warp Parallelism (CWP) in order to describe the GPU parallel architecture.  Zhang and Owens [4] develop a performance model based on their microbenchmarks so that they can identify bottlenecks in the program.  Supada [5] performance model consider memory latencies are varied depending on the data type and the type of memory Relate work

5  Different application and device cannot use same setting  Find the relationship between each parameters in this model, and choose the best block size for each application on different device to get peak performance. 1Introduction and background

6 varies data size with varies size of block have different performance

7 How GPU working

8 Memory latency hiding

9 The structure of threads

10 Specification of GeForce GTX 650

11 Parameters

12  N MB >= N TB = N* N TW (N is integer) >= N RT / N RB Block size setting under threads limitation

13 Memory resource

14  M R / M TR >= N* N TB (N is integer)  N* N TB (N is integer) <= N RT N<= M SM / M SB Block size setting under stream multiprocessor resource

15  Though more threads can hide memory access latency, but the more thread use the more resource needed. Find the balance point between resource limitation and memory latency is a shortcut to touch the peak performance. By different application and device this performance model shows it advantage, adaptable and without any rework and redesign let application running on the best tuning. Conclusion

16


Download ppt "Performance modeling in GPGPU computing Wenjing xu Professor: Dr.Box."

Similar presentations


Ads by Google