Eric Bowen
Well-known member
Hello all. These are the first benchmarks so far. I will have more coming for the 6950X, 6900K, and other hardware platforms. I still have to create 5K R3D media benchmark project file though it's much faster now with a template to use. Should have that done today or soon. Still could use some higher frame 5K and 6K media as well as some 8K media to benchmark and compare with the initial media. However this should give everyone a good ideal of the data the benchmarks will present over time and compare different hardware as well as media and workflows. I will also include some observations so far in regards to the current state of Adobe CC with regards to Premiere, AME, and AE. Those will be updated and expounded on as the data presents itself. Below those I will start posting the benchmarks.
Observations:
1. The 6950X @4GHz can push a 980TI card to 99% ie it's limit with the current GPU acceleration with Adobe
2. Adobe's GPU acceleration averaged 1700MB to 2000MB Vram for all single layer tests and 3000MB to 4000MB Vram for 4 layer tests. Adobe did use up to 5400MB vram which is more than any previous testing on older versions tested
3. Premiere threaded on all 20 threads flawlessly without issue regardless of export codecs.
4. Realtime playback in Adobe has changed to start buffering into ram as the timeline is played back the first time until max ram available is reached or ram required for timeline. Playback of 6K R3 at Full res stutters some during that buffering as the ram buffers are filled and ram usage skyrockets. CPU load stays at 100% during that time. After buffering is finished CPU load drops to less than 10% on average and 6K 1 layer or 4 Layers Playback seamlessly at full res preview. GPU load increases close to double after playback buffer is built into ram and CPU load drops to under 10%.
5. R3D 6K also pushed 980Ti to 99% on render. Average GPU load during playback was 40% to 65%.
6. H265 did not use GPU acceleration for some reason when exporting to 6K and not scaling. When using 4 layer H265 or scaling then GPU load was normal. Something Adobe needs to look at.
7. DPX render maxes Intel 540 series SSD's even in raid zero on export. None of the other codecs pushed discs above 15%. CPU load averaged 50% which reflects far greater potential for improvement with a much faster SSD raid array. The 540 series averaged 800MB/s to 980MB/s.
8. Premiere used up to 40GB of ram playing back or exporting the 30 minutes 1080AVCHD and 20 to 30GB of ram playback 30 seconds of 6K R3D. Render often reached 40GB of ram for the 6K R3D
8. AE render queue utilized 30 to 40% cpu load during render regardless of codec rendered to. Also used 40GB of ram
9. AE Ram preview generated within 3 to 5 seconds for single layer 6 K and played seamlessly while generating. The 4 Layer took about twice as long but maintained playback and the ram preview build rate so playback was uninterrupted as ram preview generated.
10. DNxHR OP1A MXF so far performs the best for rendering based on CPU load and GPU load.
11. Raw DNG very low 10 to 20% CPU load on the system while GPU load can reach 40% with 980Ti. Render was 100% CPU and 99% GPU load.
12. VLC does not like Adobe DNxHR MXF. But all clips encoded without issues or anomalies.
More to come as benchmarks continue. Onto the benchmarks posting below.
Observations:
1. The 6950X @4GHz can push a 980TI card to 99% ie it's limit with the current GPU acceleration with Adobe
2. Adobe's GPU acceleration averaged 1700MB to 2000MB Vram for all single layer tests and 3000MB to 4000MB Vram for 4 layer tests. Adobe did use up to 5400MB vram which is more than any previous testing on older versions tested
3. Premiere threaded on all 20 threads flawlessly without issue regardless of export codecs.
4. Realtime playback in Adobe has changed to start buffering into ram as the timeline is played back the first time until max ram available is reached or ram required for timeline. Playback of 6K R3 at Full res stutters some during that buffering as the ram buffers are filled and ram usage skyrockets. CPU load stays at 100% during that time. After buffering is finished CPU load drops to less than 10% on average and 6K 1 layer or 4 Layers Playback seamlessly at full res preview. GPU load increases close to double after playback buffer is built into ram and CPU load drops to under 10%.
5. R3D 6K also pushed 980Ti to 99% on render. Average GPU load during playback was 40% to 65%.
6. H265 did not use GPU acceleration for some reason when exporting to 6K and not scaling. When using 4 layer H265 or scaling then GPU load was normal. Something Adobe needs to look at.
7. DPX render maxes Intel 540 series SSD's even in raid zero on export. None of the other codecs pushed discs above 15%. CPU load averaged 50% which reflects far greater potential for improvement with a much faster SSD raid array. The 540 series averaged 800MB/s to 980MB/s.
8. Premiere used up to 40GB of ram playing back or exporting the 30 minutes 1080AVCHD and 20 to 30GB of ram playback 30 seconds of 6K R3D. Render often reached 40GB of ram for the 6K R3D
8. AE render queue utilized 30 to 40% cpu load during render regardless of codec rendered to. Also used 40GB of ram
9. AE Ram preview generated within 3 to 5 seconds for single layer 6 K and played seamlessly while generating. The 4 Layer took about twice as long but maintained playback and the ram preview build rate so playback was uninterrupted as ram preview generated.
10. DNxHR OP1A MXF so far performs the best for rendering based on CPU load and GPU load.
11. Raw DNG very low 10 to 20% CPU load on the system while GPU load can reach 40% with 980Ti. Render was 100% CPU and 99% GPU load.
12. VLC does not like Adobe DNxHR MXF. But all clips encoded without issues or anomalies.
More to come as benchmarks continue. Onto the benchmarks posting below.
Last edited: