Premiere Pro CC (2018) bottleneck

[xperiense]

Hello guys,

So i've got 3 questions:

1. Is everyone experiencing CPU bottleneck with Premiere?

We mainly edit 4k h264 footage from DJI X5 RAW camera @ 60Mbps

I know premiere is more CPU intensive in every way but i hate it so i have to make the rig work better.

With the GTX 1080 i 've found that encoding times and render previews worked faster for a 1080p timeline and export.
With the Radeo Pro Duo i 've experienced kind of slower exports (without resetting profiles) and render previews to about 20 seconds difference.

Though i think that i get slightly better timeline editing performance with the Radeon.

I really don't care about exporting time (reasonable times of course), my main concern is the real time timeline editing performance which i believe that's what all benchmarks should also include but mostly they don't. Also not including AMD professional cards is also very suspicious.

What i 've seen within premiere is that the CPU is hitting 100% load quite often especially on rendering and exporting while the GPUs just occasionally being utilized for a fraction of a second at about 5-10%.

2 .Is there a way to eliminate the bottleneck on the current configuration?

After effects can't fully support the Radeon Pro Duo though i get really good realtime performance with particles and stuff. The radeon also performs far better than the GTX 1080 in Photoshop 3D mode while one of them GPU 0 being utilized at 45% max and the cpu stays at 20% max.

My next question is:

3. Will an AMD Threadripper 1950x solve my problems with Premiere and help me edit without any glitches or slow framerates in timeline based on my current config or does the problem rely on the DJI footage no matter what cpu i add?

EDIT: I know that using proxies would be a way to go but i'm trying to understand if i can get it to work without them.

My instinct says it will but i need some help to be sure.

My configuration:
X99-Ultra Gaming
Core i7 6850k @ 4.2Ghz

RAM Corsair Vengeance 32GB 3200Mhz @2933Mhz
Samsung 850 Evo Sata SSD (Boot)
Samsung 960 Evo Nvme (Scratch & Projects)
Samsung 950 Pro Nvme (Footage)
GTX 1080 Gigabyte G1 Gaming 8GB (previous)
AMD Radeon Pro Duo (Polaris) 32GB (current)

[xperiense]

This is what Samsung has replied:

The 960 EVO is a TLC based device which has something called Turbo Write which uses a cache that is placed on the drive to provide the fast speeds. When the cache is full your writing to the NAND directly which results in lower speeds. However this temporarily and once the cache is empty the write speeds go back up. Going based off the results of these screen shots there is no present error occurring with the drives and this is completely normal. The numbers in Samsung Magician also show consistent results.

[mikeklar]

Thanks for posting the Samsung Feedback

In my case, after the firmware update, it explains the need to reboot before the read / write speed results recovered

Cheers

[xperiense]

Ok so i asked on another forum about the 960 Evo and they told me that after 20GB the 960 Evo drops to 300MB because it doesn't sustain the write speeds. That's why the DiskTest.avi and the simple copy fails.

So the only option i have here is:
850 Evo -> Boot Drive
960 Evo -> Source Footage (Original + Transcoded)
950 Pro -> Projects, Exports, Previews

So i first copy all the footage from the NAS (550MB/s Read) to 960 Evo and start from there.

And at some point if i feel comfortable (money wise) i uprgade to Threadripper 1950x to be able to max out my specs for use with the rest of Adobe CC Collection as well...

I mean Adobe Dimension CC shows 45% utilization on GPU 0 when using it prior to render and then the render is mostly CPU.

Why Adobe hasn't figured out on how to utilize the GPUs?

I can only think that it would make the countless Intel CPU product lines obsolete. Although i believe that would straighten out and simplify the Market.

[xperiense]

So concerning Premiere only and based on the current footage i am guessing any higher configuration is overkill. I mean if we use proxied footage to edit there is no need for a high end cpu or gpu. If the actual benefits are only a 10% - 20% increase in exporting times and the process is not really time sensitive (delivery wise) any config like a mainstream cpu with a low class gpu like 1060 would be more than enough. Am i wrong on that?

Of course we are talking only for Premiere use on this system.

I am downloading the Benchmark right now and i will post soon.

[RjL190365]

For Premiere use, a high-end CPU may be required, depending on the video codecs and video resolution used.

As for the GPU, the reason for the Radeon Pro Duo underperforming the GTX 1080 in exports has little to do with the GPU per se, but rather in the differences between the CUDA and OpenCL renderers. Adobe's OpenCL support in the MPE still isn't as robust or as mature as it's CUDA support.

Oh, by the way, for best performance the CPU and the GPU should be matched (relatively speaking). If one significantly underperforms the other, then that underperformance component will limit the higher-end component, and may cause the higher-end component to severely underperform even in situations which do not utilize the underperformance component.

And the worst possible combination for Premiere is an el-cheapo CPU and either an el-cheapo discrete GPU or only integrated graphics: The entire system will be excruciatingly sluggish all around.

[xperiense]

What i meant as cheap cpu and graphics would be a Ryzen 1700 and a GTX 1060 not anything lower than that. If transcoding the footage is the only option here based on the stupid DJI files especially, then this combination should be enough for using premiere even if that means you get slight worse exporting times. The real time editing with the transcoded footage would be pretty much the same unless i were to use to many video channel etc.

My original question was actually if it is worth it to move to Threadripper 1950x for real time editing. If i still need to transcode this footage there is no actual benefit as there is no absolute need for that.

[Terry_Green]

i7-8700K @ Enable accelerated Intel h.264 decoding (requires restart) https://www.youtube.com/watch?v=9GiYslv-Mho

[xperiense]

First of All i don't understand a single word. This is already enabled but the GPUs don't utilize compared to the cpu and of course there is no problem playing back a single file even with lumetri (live changing while playback) and 3 effects on top (depending on the effect)

[xperiense]

Ok so Samsung really messed up with our drives. We just have to wait for an update? Is there a way to contact support for the ssd's?

But that is not the question of this topic.
Would and AMD Threadripper 1950x improve my workflow based on the current footage we edit or would i need to ingest with that CPU as well?

Would i see the GPU utilizing more with the threadripper?

[xperiense]

I am waiting for a reply from Samsung about the problems we are having. I also tried crystal disk mark and HWInfo showed that the 960Evo was going to 95 Degress Celsius therefore i stopped it.

Anyway i overclocked the Core i-6850k to 4.3Ghz with higher voltage and it gave me 4 seconds less on the H264 test.

Then i used my 4k DJI Footage on a 1080p Timeline and applied some directional blur effects.

While playback and realtime effects changing i saw the CPU staying at 20% load and the GPU0 at 55%.

Though when exporting to h264 CPU was at 100% load and GPU at 1% (only my monitor was giving the load i guess)

What the hell is up with that?

I also tried exporting to MPEG2 Bluray and the CPU stayed at 30% load while the GPU at 1% therefore the exporting time was increasing.

I am very confused. Am i missing something?