Microbenchmarks OO
For more informations about the extension, visit GitHub WebCL (Nokia Research).
To start the benchmarking, click the benchmark buttons.
If the microbenchmarks does not work, consider refreshing your browser or restarting it.
WebCL not found
We invite you to submit your results to our public database.
Just provide us a username and accept cookies.
At any time you can delete your data (see database section).
| Computational Latencies | ||||||
|---|---|---|---|---|---|---|
| 1 Ind. | 2 Ind. | 4 Ind. | ||||
| Λ | λ | Λ | λ | Λ | λ | |
| SP | ||||||
| DP | ||||||
| SF | ||||||
| Memory Access Latencies | ||||||
| Λ | λ | |||||
| Local | ||||||
| Global | ||||||
| Private | ||||||
| Constant | ||||||
- About the results:
- Peak Performance
- The maximal attainable computational instruction per second or bytes per second.
- Small Lambda
-
The issue latency of a warp/wavefront per compute unit, which is defined as the amount of cycles before the next instruction of another warp/wavefront can be scheduled.
For simple pipelines this is 1. - Big Lambda
-
The completion latency of a warp/wavefront per compute unit, which is defined as the amount of cycles before the next dependent instruction of the same warp/wavefront can be scheduled.
For simple pipelines this is the length of the pipeline. - Ridge Point
- The amount of concurrent warps/wavefronts needed within a compute unit to fill the pipeline completely and reach the peak performance.