Small Clip Transcoding Speed: APP Versus CUDA Versus Quick Sync
We downloaded three 1080p clips (H.264/MOV container) from Apple's trailer service to show more storage space-friendly bitrates (~9.5 Mb/s) that you would likely encounter on a day-to-day basis. Since we are more interested in the GPU side of things, we are only enabling the hardware encoding when we can.
| AMD Radeon HD 6970 | ||||
|---|---|---|---|---|
| AppSetting | MediaEspressoSW Decode / APP Encode | MediaEspressoSW Decode / Encode | MediaConverterHW Decode / APP Encode | MediaConverterSW Decode / Encode |
| Up! | 0:00:37 | 0:01:05 | 0:00:38 | 0:00:30 |
| Fast & Furious | 0:00:33 | 0:00:53 | 0:00:27 | 0:00:25 |
| Letters from Iwo Jima | 0:00:36 | 0:00:59 | 0:00:31 | 0:00:26 |
| Nvidia GeForce GTX 580 | ||||
|---|---|---|---|---|
| AppSetting | MediaEspressoSW Decode / APP Encode | MediaEspressoSW Decode/Encode | MediaConverterHW Decode / APP Encode | MediaConverterSW Decode / Encode |
| Up! | 0:00:33 | 0:01:05 | 0:00:26 | 0:00:30 |
| Fast & Furious | 0:00:27 | 0:00:53 | 0:00:20 | 0:00:25 |
| Letters from Iwo Jima | 0:00:28 | 0:00:59 | 0:00:21 | 0:00:26 |
| Intel HD Graphics 3000 (Core i5-2500K) | ||||||
|---|---|---|---|---|---|---|
| AppSetting | Media EspressoPerformanceSW Decode / Quick Sync Encode | Media Espresso PerformanceSW Decode / Encode | Media EspressoQualitySW Decode / Quick Sync Encode | Media EspressoQualitySW Decode / Encode | Media ConverterHW Decode / Quick Sync Encode | Media ConverterSW Decode / Encode |
| Up! | 0:00:37 | 0:00:46 | 0:00:38 | 0:01:20 | 0:00:24 | 0:00:31 |
| Fast & Furious | 0:00:29 | 0:00:37 | 0:00:31 | 0:01:03 | 0:00:15 | 0:00:20 |
| Letters from Iwo Jima | 0:00:31 | 0:00:41 | 0:00:34 | 0:01:11 | 0:00:20 | 0:00:25 |
When it comes to smaller files and lower bitrates, the CPU route is always slower than employing GPGPU or fixed-function encoding, in the case of Quick Sync.
The only outlier here is AMD in MediaConverter. For whatever reason, we are still getting longer transcode times when we select UVD 3-based decoding/APP-based encoding. Unfortunately, MediaConverter doesn't allow us to separate the two settings, so we have no way to know which part of the pipeline is causing the slowdown. We let both AMD and Arcsoft know about the issue, and we are told they are working on a resolution to this problem. This anomaly isn't exhibited in CyberLink's MediaEspresso.
