...I currently have a max-spec 2017 iMac and it is used almost exclusively for editing H.264 footage (1080p/59.94) and exporting back out to H.264 at the same resolution and frame rate (with not many effects, mostly edited for content)...I'm wondering if either the i9-9900k in the iMac 2019 or even the T2 chip in a 10-core iMac Pro (I assume the T2 chip is what does the encoding on iMac Pro since Xeon chips don't have Quick Sync) would be able to get me even faster exports?...Have there been any improvements to QS from 7700k-9900k and/or would the additional 4-cores + Vega 48 make any difference whatsoever in export speed?
And/or would the 10-core Xeon + Vega64+ be able to beat out the 7700k Quick Sync?...
I've edited many large 4k H264 documentaries and done extensive FCPX performance testing on both top-spec 2017 iMac 27 and 10-core Vega64 iMac Pro. I haven't tested the 2019 i9 iMac. In general I think the 2019 i9 iMac is a great machine and for H264 I might get that over the iMP. Whether it's faster than the 2017 on H264, I don't know. The latest tests by Max Yuryev seem to indicate H264 decode/encode performance is limited by Quick Sync not the CPUs
As originally released the iMP was sometimes slower than the 2017 iMac on several 4k H264 codecs. This includes decode performance which affects scrubbing, JKL lag time and proxy creation, also encode performance to H264 output. I assume 1080p would be similar, although with 1/4 the pixels it would be faster on both machines.
The 2017 iMac 27 is much faster than the 2015 on the same FCPX H264 workload. This is probably due to the Kaby Lake Quick Sync improvements in the i7-7700K. Unfortunately this improvement sets a high bar to beat, and initially the iMP failed to equal this on some encoding cases and was definitely slower on decoding, evidenced by more sluggish scrubbing.
Be advised that different H264 codecs can vary greatly in editing smoothness on the same Mac hardware. The decode/encode performance on 4k H264 4:2:0 8-bit material from a GoPro may differ from similar material from a Panasonic GH5 using the same NLE and hardware.
I initially assumed the iMP was using AMD's UVD/VCE hardware acceleration in the Vega64, but apparently it was using similar logic in the T2 chip. The iMP was obviously hardware accelerated, just not as smooth as Quick Sync on the 2017 iMac.
With recent macOS and FCPX updates, the iMP has been improved and is generally faster in most cases, however it still doesn't feel as smooth when scrubbing 4k H264.
I nonetheless prefer the iMP. It is very quiet and seems more reliable. I don't know if that's the ECC RAM or the Vega64 GPU or what. For 4k H264 even the 2017 iMac wasn't fast enough for multicam without proxies, and if proxies are created almost any machine is fast enough.
We currently only shoot 4k ProRes and the iMP is a real speed demon on that. I wish it were faster on 4k H264.
Below are some tests I did in December 2018. Sorry about the primitive formatting, it's just a copy/paste from my notes. If you have specific camera codecs you want tested, let me know and maybe I have material to test on my iMac Pro vs 2017 iMac. However we've only shot 4k for the past four years so only our older material is 1080p.
December 2018 H264 tests:
- FCPX 10.4.3 vs 10.4.4
- Pro Video Formats 2.06 vs 2.07
- 2017 iMac 27 vs iMac Pro
- Pre-rendered vs non-rendered timeline
FCPX 10.4.3, macOS Mojave 10.14.1, 1:03 4k XAVC-S test file on iMP
1080p Fast Encode: 00:52.2, 00:50.8, re-run with PVF 2.06 = 00:29.3
1080p HQ: 1:34.2, pass 2 = 1:34.9
4k Fast Encode: 00:48.6
4k HQ: 01:30.1
Same test conditions as above, but on 2017 iMac 27:
1080p Fast Encode: 00:41.3
1080p HQ: 01:12.9
4k Fast Encode: 00:40.0
4k HQ: 01:13.4
HEVC 8-bit: 01:23.7
HEVC 10-bit: 40:15.0
H264 Src, ProRes422 NoFX: 00:18 (anomaly?)
ProRes422 Src, ProRes422 Out, Color+Aged Film, No PreRen: 01:17
ProRes422 Src, ProRes422 Out, Color+Aged Film , With PreRen: 01:18
Same test conditions as above, but 2017 iMac on FCPX 10.4.4, Without and With Pro Video Formats 2.07
1080p Fast Encode: 00:42.1, PVF 2.07 = 00:40.5
1080p HQ: 01:12.9, PVF 2.07 = 01:13.0
4k Fast Encode: 00:39.5, PVF 2.07 = 00:39.4
4k HQ: 01:14.6, PVF 2.07 = 01:14.4
HEVC 8-bit: 01:35, PVF 2.07 = 01:35
HEVC 10-bit: (long, cancelled, presume same as 10.4.3)
H264 Src, ProRes422 NoFX: 00:43, PVF 2.07 = 00:43
ProRes422 Src, ProRes422 Out, Color+Aged Film+Water Pane, No PreRen: 01:09, PVF 2.07 = 01:09
ProRes422 Src, ProRes422 Out, Color+Aged Film+Water Pane, With PreRen: 01:10, PVF 2.07 = 01:11
FCPX 10.4.4, same test conditions as above, but on iMac Pro:
1080p Fast Encode: 00:50.6, PVF 2.07 = 00:32.9, re-run with PVF 2.06 = 00:31.8, re-run after re-update to 10.4.4 and PVF 2.07 = 00:34.5
1080p HQ: 1:36.7, PVF 2.07 = 00:56.8, re-run with PVF 2.06 = 00:56.1, re-run after re-update to 10.4.4 and PVF 2.07 = 00:57.0
4k Fast Encode: 00:47.5, PVF 2.07 = 00:38.7, re-run with PVF 2.06 = 00:38.5, re-run after re-update to 10.4.4 and PVF 2.07 = 00:38.5
4k HQ: 01:29.5, PVF 2.07 = 01:13.6, re-run with PVF 2.06 = 01:12.89, re-run after re-update to 10.4.4 and PVF 2.07 = 01:14.6
4k HEVC 8-bit: 00:48.7, PVF 2.07 = 00:47.9
4k HEVC 10-bit: 17:01.9, PVF 2.0 = 17:41.4
H264 Src, ProRes422 NoFX: 00:43, PVF 2.07 = 00:26
ProRes422 Src, ProRes422 Out, Color+Aged Film+Water Pane, No PreRen: 01:09, PVF 2.07 = 00:12.8 (!), re-run with PVF 2.06 = 00:48.0, re-run with FCPX 10.4.3 = 00:44.9, re-run after re-update to 10.4.4 and PVF 2.07 = 00:49.3
ProRes422 Src, ProRes422 Out, Color+Aged Film+Water Pane, With PreRen: 01:09, PVF 2.07 = 00:12.8 (!), re-run with PVF 2.06 = 00:49.4, re-run with FCPX 10.4.3 = 00:45.4, re-run after re-update to 10.4.4 and PVF 2.07 = 00:48.8
