Export only uses 12% of GPU with i5-14400F and RTX 4060

[joe_0310]

Hello!  I recently upgraded (I hoped!) an older machine with spinny disks to an Acer Nitro with the following specs.  When I use Handbrake, I get about 50% GPU usage on a straight export.  With PP as you can see it's less than 10%. I've set everything I can find.

Component Details

CPU	Intel Core i5-14400F (6 P-cores + 4 E-cores, 16 threads total)
GPU	NVIDIA GeForce RTX 4060 (8GB VRAM, CUDA Capability 8.9)
GPU Driver	NVIDIA Studio Driver (latest as of reinstall, confirmed with GPUSniffer)
Memory (RAM)	32 GB DDR5-5600
OS	Windows 11 Home (64-bit), fully updated
Power Plan	Set to Best Performance
Premiere Pro	Tested on both v24.6.5 and v25.3, currently installed 25.3
Mercury Playback Engine	Set to CUDA
Display	Dual Monitors (2560×1440 each, no Advanced Color enabled)

Here are screen shots of my NVIDIA Control Panel, my export setting in PP 25.3, and my GPU performance during export.  I'm just trying to export a simple MP4 after applying a 40% Gaussian blur.  Any help, direction, whatever would be helpful.  Thank you in advance.

[R Neil Haugen]

If you do any long-GOP it is not as wise to use an Intel CPU with F in the name. As that means it specifically has no iGPU hardware, therefore no QuickSync to do hardware long-GOP encodes and decodes.

Just as a note.

[joe_0310]

I appreciate the note.  Other comments (including your own) have alluded to that.  Unfortunately, this was the build that was available.  I can replace the CPU, but it's not something I had planned to do.  Especially since Handbrake DOES take advantage of the Video Encode, I hold out hope that there is some configuration that would convince PP to at least be as tech-aware as Handbrake.   If not, then upgrade it is.  At that point the question becomes which CPU to use.  

[R Neil Haugen]

These computer builds can be so freaking complicated. On the Adobe video forums, they do have the Hardware forum that deals specifically with gear for PrPRo, AfterEffects, and Audition I think.

Also the Puget Systems page has tons of papers on their testing of all sorts of PC gear with Pr, Ae, Resolve, Nuke, whatever ... quite a resource even if you don't have them do the build.

[joe_0310]

I'll follow up on those suggestions, thanks.  First thing I'm going to do is compare my old system performance to my new system.  That may give me some insight.  And then it's on to the next steps.  I'm just disappointed because I went from a 5 year old machine with Ryzen 7 2700 X, a Radeon RX 580 and spinny disks to a shiny new, completely solid state machine and PP is actually slower.  But that's enough whining for one night.  I'll go at it again tomorrow.

[joe_0310]

Good morning!  When you say "Adobe video forums", to what are you referring? I started at the Adobe home page and I really couldn't find my way to something specific to that. Thanks for your help!

[R Neil Haugen]

I just typed "video h " in the "Search Community" line up top, and this popped up ...

https://community.adobe.com/t5/video-hardware/ct-p/ct-video-hardware?page=1&sort=latest_replies&lang...

[joe_0310]

Perfect, thanks!  I made the mistake of going all the way back to the Adobe home page, thich is why I couldn't fnid anything.  The link you provied is excellent, and it got me to the Puget Sound page which basically identified that I have a low-end card and CPU.  Probably why the machine was cheap.  🙂  That still doesn't explain why PPro doesn't work as well as Handbrake, although I've read some articles that explain the different paths that they use.  But in my real-world testing, it seems that some software won't take advantage the RTX 4060 and that the most cost effective way to start is to upgrade my CPU (as counter-intuitive as that may be).  If that goes well, I'll look at the next steps.

[R Neil Haugen]

Every app is coded differently, and players do not have nearly the same processing chain needs of an editing app. Because players are not built to run numerous layers of multiple video and audio streams at once, as an editing app must be built.

As to the use of the GPU and CPU, they are always set for different parts of the process, they are not simply extensions of the other. Completely different code and processing chains required. 

So as an app is designed, they set the code for certain things to the CPU, for others to the GPU if available, and set the limits as to how far up the line each can go in relation to the other processing chain.

So if the CPU can only do so much, the GPU will spend a lot of time pretty much idling.

[joe_0310]

I would love to be content with that response, but the sticking point for me is that Handbrake works.  It encodes at four times the speed, using nearly 100% of both the 3D and Video Encode processing of the RTX 4060.  So it's doing something PPro does not.  And so be it, I'm done beating my head against the wall.  I just wish I could get a definitive answer from someone at Adobe.  I'm going to live with it until I can break my piggy bank for the CPU.  This time, though, I am going to do my research before I buy.  🙂  I've already read a lot about power consumption (TDP and Turbo!) and fans and airflow and AIO cooling.  I think I know my target specs now for Phase 1.

[R Neil Haugen]

I know the frustration, and that it is not intuitive that a heavy NLE doesn't simply code one stream of long-GOP better than a much smaller app that happens to have been built totally from the ground up for that one little workload.

But ... just personally, the processing between Handbrake and any NLE or heavy video post app is simply not comparable. Whether you're talking Resolve, Nuke, Avid, Premiere, whatever. So says all the people that know that stuff cold.

I tend to trust the high-end folks on this rather than insist I know better. Even when it doesn't seem to be what I thought it should be.

This is like so many coming into video editing from high-end gaming, running awesome performance on a big rig that's a Ryzen CPU and an AMD GPU. And get into video post and get mad because that rig doesn't work nearly as well in video editing and effects work.

It's VIDEO!

Well, video gaming puts an entirely different processing chain and need into the system. They aren't comparable. And I stepped into that hot mess right off, as my then-computer build guy was a real hot-shot at gaming rigs, built me an awesome system.

That rig just sucked with both Premiere and Resolve. Big time. A moderately expensive lesson, that was. As a gaming rig, it had awesome test scores. And yea, it ran Handbrake beautifully also. Wasn't bad at Lightroom and Photoshop. But it couldn't handle the consistent heavy demands of Premiere and Resolve for processing video on a timeline.

Perhaps another metaphor would be trying to drive an 1,800 horsepower car built for drag racing ... to the grocery store. You might actually get there, but It just won't work well, and will take an enormous amount of fuel. But a little old Volkswagen bug works great at that. 

What's the deal? They're both cars. Right?

[joe_0310]

I appreciate that.  I really do.  (And I hope you aren't annoyed by my persistence, because you've been most generous with your time.)  But also remember, the other issue (the one that got me started down this path in the first place) is that my all around slower machine (slower CPU, slower GPU) is faster than my new rig.  And this seems to center on your point that the F series CPUs just don't "work right".  To stretch your analogy to the breaking point, it's like putting a bigger engine (the I5-14400F) with a blower (the RTX 4060) into a car, but the car decides that the blower won't work with that engine and disables it.  And while that sentence clearly identifies my lack of expertise both in things automotive and in video editing, it highlights my issue. So, lacking a clear statement, my only option is to get a new CPU.  The final data point will be whether, if I stick in an equivalent non-F CPU, I get the performance I expect. If not, at that point I'll really want guidance from the experts at Adobe.

[R Neil Haugen]

Check with the video hardware forum around here, and also with the Puget Systems website for their published information.

And maybe @RjL190365 will pop in.

[mattchristensen]

@joe_0310 could you help me understand what your concern is here? It sounds like you're comparing the GPU usage in Handbrake when doing a straight transcode, against the export from Premiere Pro when you have a blur applied.

These aren't the same operations, and so I wouldn't expect the same system utilization. Some things to consider which can affect system utilization:

• Make sure Preferences > Media > Hardware accelerated decode is turned on in Premiere Pro. You may also see sub-checkboxes for Intel and NVIDIA, though as mentioned if your Intel CPU doesn't have an included GPU on the chip, you may not see an Intel checkbox
• Compare the bitrate export settings between Handbrake and Premiere Pro. Are you using VBR 1-pass? If so, at what target bitrate and is it the same in both?
• Is your Premiere Pro sequence size and frame rate matching your source clip, and then are the export settings also matching? If not that's fine, but it could be a difference in performance compared to Handbrake which is just passing through the same frame size.

Finally, what's more important at the end of the day is not the usage % of an individual component, but the time it takes to complete the job. How much faster or slower is Premiere Pro at the export compared to Handbrake? And what if you do it without the blur? 

[joe_0310]

Thanks for the response, Matt! Understand that I'm pretty far down the rabbit hole here, so it may look like I'm off in the weeds a bit. The root cause is that my new solid state machine is slower than my old box. My old box is decent, but unless I'm missing something the new box should outperform it pretty handily. You can see the specs on the new machine in my first post, and I briefly outlined the old machine. Would you agree that with the same Premiere Pro project cnofigured the same way new box should outperform the old box?  Because it doesn't.

So I've been looking at all sorts of things, and one suggestion to make sure the RTX 4060 wsa being engaged was run it through Handbrake and that's where I found that Handbrake outperformed PP, certainly from the aspect of making use of the RTX 4060.  Absolutely it's apples and oranges, but the primary thing I was checking was whether or not the RTX 4060 was available to be used at all, and Handbrake proved that it was.  As I get some more time, I'll try to gather some more apples to apples comparisons especially of PP on the old box and PP on the new box.

[R Neil Haugen]

I'll be interested in any reply from Matt on this ... or @RjL190365 who is the 'resident expert' volunteer on hardware issues. I know from working in several video post apps that every one of them is coded to handle hardware differently.

And in this case, as Handbrake doesn't need any other resources, it is probably coded to use hardware very differently than an NLE, which has to allow for a ton of effects processing & such in the configuration for how it uses the hardware available to it.

The GPU isn't simply another form of CPU, it has very different internal math and coding processes. So the program can't simply throw one thing at whatever is available ... that has to be separate coding chains depending on a ton of other factors. 

Which we users rarely can even guess at. But as stated, it will be interesting to get a response from Matt or RJL, as I learn from both all the time.

[joe_0310]

Thanks again Matt for reaching out. What I did tonight was create a new project, input a short MP4 (about 30 seconds) and then just export it.  I did that on both machines, the old Ryzen/RX580 (Machine 1) and the new 15-14400F/RTX4060 (Machine 2). To make it as apples to apples as I could, I first created a folder on Machine 1, copied in the input file, created the project, took screen shots of settings and then the performance dashboard while doing the output. Then I copied that folder directly over to Machine 2, and opened the project, did some more screen shots, and then ran the export.  The results were that it took about 17 seconds to run on Machine 1, and a little over 30 to run on Machine 2.  The configuration screen shots were, as expected, identical.  I'm including the performance screen shots here.  I also included the GPUSniffer output on each machine. Is there something obvious I'm missing?

Machine 1:

 Machine 2:

I'm not a hardware guy or really an expert of any kind. I am a programmer, though, and I'm good at running tests. Given that, my layman's interpretation is that Machine 1 uses the GPU of the RX580 for video encode, while Machine 2 does not really use the RTX4060.  I mean it does a little (the peak is about 10%). Is there something I'm missing?

[RjL190365]

You got things wrong there. Your GPU is being slammed by simply rendering the Gausian blur effect, which along with whatever GPU-accelerated effects you might not have been aware of that totaled 84% GPU utilization (as indicated by the 3D load). By contrast, the RX 580 does not do much to process the Gausian Blur effect, and thus the 3D utilization is very low there, resulting in very poor GPU effects performance.

The NVENC encoder is stuck below 10% because H.264 doesn't demand much if any hardware encoding capability. Your old GPU showed far greater encoding utilization because the RX 580's encoder is far less advanced than the RTX 4060's.

However, the CPU utilization being below 20% in this workflow is indicative of a bottleneck somewhere in your system. Usually in the form of misbehaving background processes interfering with your new system's performance that didn't occur with your old PC.

What's worse, big-name computer brands tend to load the system with bloatware that steals performance from your new system and often cannot be easily uninstalled without destroying its existing OS installation.

[joe_0310]

Thanks for that input, @RjL190365.  The Gaussian blur was just an attempt to see what happens with more load.  I admit I don't know much so I wanted to see what happened.  But the latest comparison was with no effects of any kind.  It takes twice as long on the new machine.  And from my testing, Handbrake is faster.  I don't know enough to know whether this is a fair comparison or not, but when I turn it to as high of quality as I can, the same file exports in 10 seconds and seem to make much better use of the GPU.  My goal I guess is to see if I can get that same performance from Premiere Pro.

[RjL190365]

First thing to check:

Was your old PC a custom- or self-build? And is your new PC bloated with tons or trialware (trial programs)?

Second, AMD GPUs historically prioritize speed over quality to begin with. Did you ever inspect the image quality of your exports from both the old and the new PC?

[joe_0310]

Old PC was custom.  The new PC isn't running any trialware that I can see.  In fact, it's pretty clean, only 78 background processes and 103 Windows processes (both less than the old machine at 133 and 105).  The background processes are nearly all pretty recognizable: primarily Acer, Adobe, Creative Cloud, Intel, Microsoft, NVIDIA, Realtek, Samsung and Windows. And the sticking point is that Handbrake is quite a bit faster.  In my simplistic view, I would expect bloatware to affect Handbrake as well.

The outputs look the same. The first frames seem to be identical.  I don't see any additional blurring, everything is equally clear on both. The sizes of the two output files are nearly the same, with Machine 1 being 128MB, and Machine 2 being 135MB.  The input file is 149MB.  Interestingly, with everything cranked up the Handbrake output file is 403MB.

[Kyle Plumadore]

I would generally expect your newer machine to outperform your prior machine in PPro - but many factors can come into play when comparing across workflows and hardware. You've done a good job isolating a simple use-case though (the 30 second MP4 export), and I'm somewhat surprised to see the older machine complete the export faster. 

One thing that would be useful to try - if you are able to - is to swap the GPUs of the machines and rerun your simple export. That may help narrow down whether the difference is primarily a result of the GPU, or something about the remainder of the system config. I would try to make sure the same nVidia/AMD driver versions are used after swapping to minimize the number of variables as even driver versions can make a difference. 

[joe_0310]

Thanks, Kyle!  Swapping GPUs is not really in my wheelhouse at this time.  While I'm okay opening my new machine (I've installed memory and a high-performance SSD),  I am absolutely loathe to play with the perfectly working old machine. Again, it may be a red herring, but I continue to wonder about the fact that Handbrake is so much faster and makes better use of the RTX 4060 than Premiere.

I'm currently trying Voukoder as another data point, but the older free version seems to be doing essentially the same thing.  Another option will be to try Media Encoder. The problem of course is every one of these tests takes time.  I'm just wondering if there's any logging that can be enabled on PP to help identify why the RTX 4060 video encoder is not being used to its fullest.

[joe_0310]

I've done everything I can think of.  I tried with the older free version of Voukoder, and with a trial of the for-fee Voukoder Pro.  I tried routing through AME and directly from PPro.  In all cases the export takes around 30 seconds or so, with the 3D performance peaking at about 75% and the Video Encode at 10-15%.  Then I run the same input file through Handbrake and Video Encode goes to 100%, with 3D at around 85%, and it creates the file in about 6-7 seconds.
So I'm willing to replace the CPU and get a standard I5-14400, but I'm worried it may not perform, either, and I don't get the sense so far that there's a clear reason why Handbrake works and PPro doesn't.  Is there any way to get a more definitive answer before I drop a couple hundred dollars?