scorsesefan
Veteran
I would def transcode.
Fuji X-T3 released
- Thread starter Ken1212
- Start date
G
Guest
Guest
Half fingers crossed that an eGPU will handle the 4k h.265. (Nervous laughter)
A recent GPU (e.g. GTX 1080 level, perhaps even less powerful) can easily handle H.265 10+ bit (even RGB 444 when it becomes available), including 4K 50, 60. Whether your NLE has a decent GPU implementation is another question. Here's PP CC playing X-T3 4K50p (60p would also work), captured at 4K 60 on a 4K60 display, as well as being compressed to 4K60 H.264 via OBS at the same time(!). Includes a simple cross fade transition which means 2 streams of 4K50 being played at once:
For more complex edits, turning on proxies solves workflow issues (at the expense of disk space): all NLEs now have proxy support. Depending on your proxy settings, in most cases you'll use less disk space using proxies than transcoding (since the proxies will be much smaller than 4K transcodes to high-bitrate ALL-I formats (to preserve same quality provided by H.265)).
G
Guest
Guest
Thanks, I’ll start with proxies, and then get an eGPU once i get into anything beyond quick assembly edits.
Your test are very reassuring! Really appreciate you sharing those!
Your test are very reassuring! Really appreciate you sharing those!
marvinhello
Well-known member
BIG no no.
Premiere Pro CC never uses GPU for H.264 and H.265 decoding/encoding. For RED materials and of course some effects, yes GPU is involved.
Sounds shocking? Yes it's true, had a chat with Adobe staff during NAB and they stressed that hardware accelerated decoding and encoding of H.264/H.265 leverage some particular Intel CPU instruction sets - AVX/AVX2/Quick Sync
To really use Nvidia GPU for decoding and encoding, you need Nvidia's SDK - NVDEC and NVENC. Adobe has not integrated them into Premiere but there are third party plugins for Premiere, however buggy and unstable.
Also, the hardware accelerated playback seems to be limited by bitrate as well, the above 4K 50p material was shot at 100Mbps. Generally we would use 200Mbps for that, and AII-I 400Mbps for other framerates, you will notice stutter with higher bitrate settings.
Premiere Pro CC never uses GPU for H.264 and H.265 decoding/encoding. For RED materials and of course some effects, yes GPU is involved.
Sounds shocking? Yes it's true, had a chat with Adobe staff during NAB and they stressed that hardware accelerated decoding and encoding of H.264/H.265 leverage some particular Intel CPU instruction sets - AVX/AVX2/Quick Sync
To really use Nvidia GPU for decoding and encoding, you need Nvidia's SDK - NVDEC and NVENC. Adobe has not integrated them into Premiere but there are third party plugins for Premiere, however buggy and unstable.
Also, the hardware accelerated playback seems to be limited by bitrate as well, the above 4K 50p material was shot at 100Mbps. Generally we would use 200Mbps for that, and AII-I 400Mbps for other framerates, you will notice stutter with higher bitrate settings.
G
Guest
Guest
I shot with shaprness set to 0, and then the same shot with -4, in HD 10bit. -4 is more flattering. 0 is sharpened, but tastefully sharp. It felt like -4 was reducing some detail. I may go with a -3 for general purpose. -4 for interviews. -1 for detailed scenes. But at the end of the day, i’d be happy running with -4 all the time.
EDIT: viewing HD footage at 100% there appears to be little or no softening. Was most likely the blowing up the image on a 5K screen or possible motion blur. I’d love it if a camera did some blurring at negative sertings, especially the extreme negatives.
Haven’t messed with 4K yet. Only got to play with it twice.
Last edited:
puredrifting
Major Contributor
Hmm...
I bought two ProGrade 128GB V60 cards when I bought the XT-3. They have played well for five months now, have shot lots of material as well as long interviews. Was shooting some tabletop yesterday and card 1 in slot 1 started shutting recording off, displaying a warning, "SD Card Error" but it appeared to actually save the file before shutting off.
Did testing yesterday and kept getting this error message with card 1 so I took it out of rotation for a shoot today. I switched to shooting with card 2 today, lo and behold, the same issue. These are premium V60 cards. In ordered to finish the shoot this morning, luckily I had my Sandisk Extreme Pros from my C200 with me. I used them to finish the
shoot today with no issues.
I am back home looking at the cards, it appears that I didn't lose anything off of the ProGrade cards, all of the files are there and playback, it's just that both were causing the camera to randomly stop the recording and save the file, then displaying an SD card error. I will contact ProGrade Monday, I think these have a year warranty, but weird that
they have worked great for 4-5 months, then all of a sudden two shut down the camera recording many times over two days? Anyone else use or have issues with the ProGrade V60 cards with the XT-3?
I bought two ProGrade 128GB V60 cards when I bought the XT-3. They have played well for five months now, have shot lots of material as well as long interviews. Was shooting some tabletop yesterday and card 1 in slot 1 started shutting recording off, displaying a warning, "SD Card Error" but it appeared to actually save the file before shutting off.
Did testing yesterday and kept getting this error message with card 1 so I took it out of rotation for a shoot today. I switched to shooting with card 2 today, lo and behold, the same issue. These are premium V60 cards. In ordered to finish the shoot this morning, luckily I had my Sandisk Extreme Pros from my C200 with me. I used them to finish the
shoot today with no issues.
I am back home looking at the cards, it appears that I didn't lose anything off of the ProGrade cards, all of the files are there and playback, it's just that both were causing the camera to randomly stop the recording and save the file, then displaying an SD card error. I will contact ProGrade Monday, I think these have a year warranty, but weird that
they have worked great for 4-5 months, then all of a sudden two shut down the camera recording many times over two days? Anyone else use or have issues with the ProGrade V60 cards with the XT-3?
Visitors from Planet Wrongulus might be visiting this forum lol (signs are downvoting of accurate information).
Bitrate generally doesn't affect decode CPU/GPU load if it's the same codec. In any case, we can do some real-world experiments plus a little math to figure out actual GPU load.
The GPU is being used for playback in PP CC, where GPU load scales with codec framerate. Baseline GPU frame copy/swap/GUI update can be measured, getting 8% GPU and 6% CPU playing Black Video in the same Sequence at 50p, and going up to 9% GPU and 8% CPU for Black Video in a 60p Sequence.
GPU: NVIDIA GTX 1080.
4K 50p load (GPU: 21%):
4K 60p load (GPU: 26%):
Subtracting baseline Black Video overhead, we get H.265 X-T3 GPU decode load:
21%-8% = 13% for H.265 X-T3 50p
26%-9% = 17% for H.265 X-T3 60p
Note for graphics+math operations, the GPU is much more powerful than the CPU, and still uses almost 1/5 the GPU for 60p. All things considered, more of the GPU could be utilized given the high CPU load.
For reference: 1DX II 4K 24 MJPEG 422 8-bit uses 8% GPU (5% GPU Blank Video) and 33% CPU (3% CPU Blank Video). Thus 3% GPU and 30% CPU for decode load.
Bitrate generally doesn't affect decode CPU/GPU load if it's the same codec. In any case, we can do some real-world experiments plus a little math to figure out actual GPU load.
The GPU is being used for playback in PP CC, where GPU load scales with codec framerate. Baseline GPU frame copy/swap/GUI update can be measured, getting 8% GPU and 6% CPU playing Black Video in the same Sequence at 50p, and going up to 9% GPU and 8% CPU for Black Video in a 60p Sequence.
GPU: NVIDIA GTX 1080.
4K 50p load (GPU: 21%):
4K 60p load (GPU: 26%):
Subtracting baseline Black Video overhead, we get H.265 X-T3 GPU decode load:
21%-8% = 13% for H.265 X-T3 50p
26%-9% = 17% for H.265 X-T3 60p
Note for graphics+math operations, the GPU is much more powerful than the CPU, and still uses almost 1/5 the GPU for 60p. All things considered, more of the GPU could be utilized given the high CPU load.
For reference: 1DX II 4K 24 MJPEG 422 8-bit uses 8% GPU (5% GPU Blank Video) and 33% CPU (3% CPU Blank Video). Thus 3% GPU and 30% CPU for decode load.
marvinhello
Well-known member
20% something GPU usage and 100% CPU usage is the perfect proof that there's no GPU acceleration involved. You'd NEVER see any CPU spike with GPU accelerated playback.
The difference with the GPU usage between HEVC and MJPEG is that GPU has to wait longer and process the more computationally intensive data from CPU for it to be drawn in the UI.
So James, getting an eGPU in the hope that XT3 HEVC files might play smoother is unrealistic.
The difference with the GPU usage between HEVC and MJPEG is that GPU has to wait longer and process the more computationally intensive data from CPU for it to be drawn in the UI.
So James, getting an eGPU in the hope that XT3 HEVC files might play smoother is unrealistic.
G
Guest
Guest
Well, it would be nice, but i am not using it professionally, so, i suppose playback will be what it be. Will shoot some stuff next week.
scorsesefan
Veteran
Did you upgrade to v3.00 firmware, Dan? If so I wonder if this had something to do with it? Not likely, but just a thought...
Grug
Veteran
Just had my first narrative shoot with the X-T3 yesterday.
It was for Instagram TV, so a very different style of storytelling to anything I've done before. We were replicating an influencer kind-of vlogging their way through the story, so had the camera on a Ronin S, and I rigged a secondary handle to the Ronin, so that our talent could 'hold' the camera while I operated.
The Fuji didn't miss a beat, but I had a hell of a time the night before, sorting out how to get a wireless feed out successfully to video village, when we were recording 4k @ 50p (instagram TV wants 30p at a minimum).
Because there was going to be so much gack on the rig, I lost the cage and grip and just went with the bare body of the camera, which was fine. But you really notice the battery life, running a single internal Fuji battery.
It was for Instagram TV, so a very different style of storytelling to anything I've done before. We were replicating an influencer kind-of vlogging their way through the story, so had the camera on a Ronin S, and I rigged a secondary handle to the Ronin, so that our talent could 'hold' the camera while I operated.
The Fuji didn't miss a beat, but I had a hell of a time the night before, sorting out how to get a wireless feed out successfully to video village, when we were recording 4k @ 50p (instagram TV wants 30p at a minimum).
Because there was going to be so much gack on the rig, I lost the cage and grip and just went with the bare body of the camera, which was fine. But you really notice the battery life, running a single internal Fuji battery.
G
Guest
Guest
Yeah, they put a lot of options in there. Kind of daunting digging through menus to get the right recipe of settings. Definitely nice to have a full day to mess with it.
Sounds like a fun shoot.
Tested playback on my late 2013 MBP (max spec) with X-T3 files- not enough juice so turned on Proxies for Ingest (PP CC). Imported (ingested) the files, proxies created in the background with AME (not as seamless as FCPX or Resolve, but works fine). Once the proxies were done, I added the toggle proxies button to the main control area, and it's trivial to switch between proxies. Exporting uses the originals; editing is very fast as expected. If using H.264 proxies, the proxy overhead is minimal (much less than transcoding to e.g. ProRes (I'd use Cineform wavelet in that case to save ~2x the space).
G
Guest
Guest
Well, dang! I may end up using an X-T3 for a 4K gig or two! That seems so much better than i was thinking. I have done my share of editing, but never had to proxy’s when editing. Been fortunate lately to leave editing to fulltime editors, but as a result, i’ve fallen behind the times. Thank you.
Back in 2011, a beefy MBP could barely get theough 1080 h.264 real time playback. The proxy workflow seems pretty easy now.
Another data point from our live stream capture box: same motherboard (ASUS X99), 6850K 6-core (vs. 6950X 10-core), GTX 980ti (vs. GTX 1080), 32GB RAM, Samsung SATA SSD (vs. NVMe), same X-T3 50p test footage:
With 1280x720 proxies:
76% CPU and 24% GPU show we've got even better performance on this much lower cost machine! It's clear the GPU is helping and also clear why it's so important to run benchmarks: higher spec machines don't always run faster. As a software developer I can explain one reason why: there can be design decisions that make the more common case, e.g. 6 cores run faster than 10 cores (note actual clock speeds in this test were similar: 3.53 vs. 3.82 (actually less HP x6 vs x10)). Similarly, the 980ti's '(ti) boosted' aspects may be helping it perform better than the 1080.
The tiny 10-12Mbps 1280x720 H.264 proxies show almost no load whatsoever lol. 1920x1080 and even 4K H.264 might also be fine as proxies for similarly spec'd machines. AME proxy transcodes were blazing fast on this box.
EDIT: after adding Lumetri effect, GPU load rose to 41%, still played fine until crossfade, then GPU dropped to 3% and performance stalled (2nd clip had no effects at all). I need to run a similar test on the higher spec'd box- don't think it would have any issue there after also handling 4K H.264 recording during playback. Switching to the tiny H.264 proxies playback was totally smooth. Thus proxies look like the best option until Adobe updates their systems to better utilize the hardware for H.265 (note some Intel chipsets provide H.265 encode/decode support too, which Adobe supports). H.264 proxies are tiny so disk space shouldn't be an issue (only issue is time to transcode the proxies on slow hardware).
EDIT2: haha awesome, the high-spec machine (6950X + GTX 1080 + 64GB RAM) easily handled the effects and transition- no slowdown or issues with the cross fade. This is now a ~2 year old box, not clear how much lower one can go to keep performance real-time (6850K + GTX 980ti + 32GB RAM is too slow). Current top spec AMD + Intel CPUs (TR/i9) and Radeon + NVIDIA GPUs can handle even more (benchmarks needed to see how much more). This looks like reasonable parts recommendations (from Dec 2018): http://www.logicalincrements.com/articles/videoediting
For PP CC, NVIDIA still leads by a fair margin: https://www.pugetsystems.com/labs/a...19-AMD-Radeon-VII-vs-NVIDIA-GeForce-RTX-1395/ . Interesting that the R VII rocks in Resolve and drops the ball in PP CC: card specific optimizations are helping/hurting. Overall NVIDIA is the safer bet for now (has been for a long time- too bad they and Apple aren't currently getting along (one of the reasons I support both Mac and PC boxes at all times)).
Also note new Creator Ready drivers from NVIDIA for Premiere/PS in Windows: https://www.pugetsystems.com/labs/a...ey-faster-in-Premiere-Pro-and-Photoshop-1392/ (maybe 8-9% improvement; bigger motivator appears to be more testing which should equal less bugs).
EDIT3: Installed the latest CRD drivers on the 6950X + GTX 1080 box: CPU usage for these X-T3 H.265 examples dropped to ~53%, GPU holds around 24%, and that's WITH the Lumetri effect enabled. That said, it looks like they are caching quite a bit (more memory usage) and it can 'trip and fall' where their pipeline system falls apart and GPU usage drops to near zero, CPU usage goes up and we get stuttering until we start/stop. It's clear that when the software is better optimized, the hardware is plenty fast.
With 1280x720 proxies:
76% CPU and 24% GPU show we've got even better performance on this much lower cost machine! It's clear the GPU is helping and also clear why it's so important to run benchmarks: higher spec machines don't always run faster. As a software developer I can explain one reason why: there can be design decisions that make the more common case, e.g. 6 cores run faster than 10 cores (note actual clock speeds in this test were similar: 3.53 vs. 3.82 (actually less HP x6 vs x10)). Similarly, the 980ti's '(ti) boosted' aspects may be helping it perform better than the 1080.
The tiny 10-12Mbps 1280x720 H.264 proxies show almost no load whatsoever lol. 1920x1080 and even 4K H.264 might also be fine as proxies for similarly spec'd machines. AME proxy transcodes were blazing fast on this box.
EDIT: after adding Lumetri effect, GPU load rose to 41%, still played fine until crossfade, then GPU dropped to 3% and performance stalled (2nd clip had no effects at all). I need to run a similar test on the higher spec'd box- don't think it would have any issue there after also handling 4K H.264 recording during playback. Switching to the tiny H.264 proxies playback was totally smooth. Thus proxies look like the best option until Adobe updates their systems to better utilize the hardware for H.265 (note some Intel chipsets provide H.265 encode/decode support too, which Adobe supports). H.264 proxies are tiny so disk space shouldn't be an issue (only issue is time to transcode the proxies on slow hardware).
EDIT2: haha awesome, the high-spec machine (6950X + GTX 1080 + 64GB RAM) easily handled the effects and transition- no slowdown or issues with the cross fade. This is now a ~2 year old box, not clear how much lower one can go to keep performance real-time (6850K + GTX 980ti + 32GB RAM is too slow). Current top spec AMD + Intel CPUs (TR/i9) and Radeon + NVIDIA GPUs can handle even more (benchmarks needed to see how much more). This looks like reasonable parts recommendations (from Dec 2018): http://www.logicalincrements.com/articles/videoediting
For PP CC, NVIDIA still leads by a fair margin: https://www.pugetsystems.com/labs/a...19-AMD-Radeon-VII-vs-NVIDIA-GeForce-RTX-1395/ . Interesting that the R VII rocks in Resolve and drops the ball in PP CC: card specific optimizations are helping/hurting. Overall NVIDIA is the safer bet for now (has been for a long time- too bad they and Apple aren't currently getting along (one of the reasons I support both Mac and PC boxes at all times)).
Also note new Creator Ready drivers from NVIDIA for Premiere/PS in Windows: https://www.pugetsystems.com/labs/a...ey-faster-in-Premiere-Pro-and-Photoshop-1392/ (maybe 8-9% improvement; bigger motivator appears to be more testing which should equal less bugs).
EDIT3: Installed the latest CRD drivers on the 6950X + GTX 1080 box: CPU usage for these X-T3 H.265 examples dropped to ~53%, GPU holds around 24%, and that's WITH the Lumetri effect enabled. That said, it looks like they are caching quite a bit (more memory usage) and it can 'trip and fall' where their pipeline system falls apart and GPU usage drops to near zero, CPU usage goes up and we get stuttering until we start/stop. It's clear that when the software is better optimized, the hardware is plenty fast.
Last edited:
G
Guest
Guest
I am primarily editing in Davinci these days. Had to get out of the Adobe eco system. Was getting too mant post freebie requests, and I had no interest in post beyond coloring. So, Davinci Resolve has been great, gives me competent editing tools, and I don’t have to switch apps to color.
Later on, i may need a BM eGPU. But if i really need horspower, i’ll probably just build a PC.
Later on, i may need a BM eGPU. But if i really need horspower, i’ll probably just build a PC.
The proxy system is integrated into Resolve- let us know how it goes with X-T3 clips. Good news for Resolve users on Mac: AMD GPU performance is really good (hardware is getting reasonably well utilized)!
NVIDIA still 'wins', but that's not the point: the AMD hardware is getting well utilized, providing excellent performance: https://www.pugetsystems.com/labs/articles/DaVinci-Resolve-15-AMD-Radeon-VII-16GB-Performance-1382/
As for building a PC, this is a min-spec I'd suggest for X-T3 H.265 and similar loads: https://www.pugetsystems.com/recommended/Recommended-Systems-for-Adobe-Premiere-Pro-CC-143/Buy_192 (you can build yourself for less with similar spec parts: quite a bit less $$$ than what I spent for the 6950X NVMe GTX 1080 64GB build a few years ago).
Last edited: