To allow workload benchmarking, all of the Intel® Media SDK tutorial source code samples contain the macro “ENABLE_BENCHMARK”. If set, the macro activates measurement of the complete workload run time.
For benchmarking accuracy, it’s important to understand the performance impact of uncompressed frame read/write file access. Without disabling such file access, the benchmarks become inaccurate and do not convey the real performance of the HW. To address this, the macros “ENABLE_INPUT” and “ENABLE_OUTPUT” can be disabled to prevent uncompressed frame file access. However, note that when disabling input of uncompressed data, the input surface will be initialized as a static “one color” image. This naturally does not represent a typical encoder or VPP input. But the purpose of the captured benchmark is not to convey real life performance but to illustrate relative performance improvements using various implementation techniques.
The table below summarizes a performance benchmark snapshot of all the workloads analyzed using Intel® Graphics Performance Analyzers (Intel® GPA). As can be seen, for each use case, the run-time performance and GPU utilization improves gradually as optimizations are applied. For details about each workload go here.
Workload Name |
GPU load (%) |
CPU load (%) |
Run time (s) |
||
Overall | EU | MFX | |||
simple_2_decode | 72 | 0 | 72 | 36 | 3.30 |
simple_2_decode - d3d | 95 | 0 | 95 | 15 | 1.84 |
simple_3_encode | 70 | 52 | 18 | 5 | 12.50 |
simple_3_encode - d3d | 84 | 63 | 21 | 4 | 8.10 |
simple_3_encode - d3d – async | 97 | 84 | 32 | 6 | 5.00 |
simple_4_vpp_resize_procamp | 20 | 20 | 0 | 30 | 2.25 |
simple_4_vpp_resize_procamp - d3d | 38 | 38 | 0 | 7 | 1.10 |
simple_5_transcode | 70 | 43 | 26 | 11 | 22.50 |
simple_5_transcode_opaque | 82 | 55 | 33 | 6 | 17.55 |
simple_5_transcode_opaque - async | 97 | 82 | 54 | 9 | 7.70 |
Configuration: 3rd generation Intel® Core™ i5-3317U, 1.7 GHz., 4 GB RAM. 32-bit Microsoft Windows* 8. ~50s 1080 p content. Microsoft DirectX* 9 workloads. For more information go to http://www.intel.com/performance
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations, and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products.
"