All this work... for what??

[33838]

Why don't you use a corkscrew or make a cap to avoid spilling that liquid

[Luscious]

There's definitely an argument to be made for chips with faster clocks versus more cores... and why Intel has those 6144 and 6244 Xeons out there. Two of those inside a workstation can really make a system fly.

 

That said, I think anything between 8 and 18 cores that can be overclocked is a great sweet spot, and that's exactly where products like the Dominus Extreme will be able to flex their muscle. It ain't cheap, but you get what you pay for and can beat a traditional dual Xeon platform in many ways. It also opens up custom water cooling - something tough to do on rack mount and 4U tower cases.

[dgsddfgdfhgs]

failed clickbait.

I have no idea what this is about. 

[JoostinOnline]

25 minutes ago, dgsddfgdfhgs said:

failed clickbait.

I have no idea what this is about. 

Dislike the video if you don't want them to keep it up.

[dabockster]

Quote

Core i9 9990 XE

That's a funny way of saying Ryzen 3950X.

 

Like seriously, Ryzen are consumer chips, far more plentiful in the world than the 9990, and can match/beat it in core count and frequency. I want to see a match-up between this Intel render server and a Ryzen powered equivalent.

[ZXR]

12 hours ago, unijab said:

Inb4

 

3rd gen ryzen comments

too late

[OSKrueger]

Seeing the improvements in the encoding time and hearing this(16:02), all I could think about was:

 

 

source: https://www.xkcd.com/1205/

[ReaverM1]

Any way I could get my hands of the clip that was used to transcode as well as the exact settings chosen in Adobe Encoder? Would like to compare performance. Thanks in advanced!

[ahuckphin]

Is he running windows 8.1?

[dgschrei]

OK I just had to come over to the forum for this:

 

Mistake #1 using AME.

 

Whatever I do it is consistently the slowest possible encoder I can get my hands on. If you have any chance of not using it take it.

 

Now I'm guessing you guys are stuck with it because you want to export Premiere project files. so assuming that is not an option:

 

Mistake #2: What actually seems to be killing your performance is rendering the videos over the network.

 

When Linus was testing the 14 core on the installed machine the project was accessed over a network drive . When he was doing the initial tests with the 16 core the video was rendering from C:\....  (I freeze framed the video to verify)

 

Why is this a problem?

If the encoder is not optimized for that it might read data in in tiny junks and also write the result out in tiny junk. Every time it does that that operation will suffer the full latency of the network. Now while this might be ~ 1ms in your LAN this still adds up VERY quickly and utterly kills your performance. Hence the way lower utilization on the render server with the 14 core CPU installed.

 

We had a similar problem in our company fetching all the build results (think tens of thousands of dlls and config files) from a remote server. It actually turned out that it was faster (about 2x-3x) to download a zip file of the build result and then unpack it locally, then it was to just copy the folder, because the latency of all those small accesses was killing the speed and we couldn't saturate the network that way.

 

So the solution might be to first copy the project to be rendered out locally onto the render machine and then encode it on there. That would get rid of all the latency and might give you a decent speedup. And even if it isn't faster for one job it will definitely increase your overall throughput.

 

 

P.S.: I still would advise to use some of the coding know how in your company to have someone just whip up an encoding script that just uses ffmpeg for the encoding if at all possible. In my experience that is the most stable option that produces the best results both in performance and quality terms. ffmpeg's h264 encoder x264 also should scale way better onto many cores (especially since you guys render 4k) than the sorry excuse for an encoder Adobe ships with AME.

[MmRaZ]

15 hours ago, Cyanara said:

Did that last motherboard actually get its BIOS checked and updated? When I built a 9900K system on a motherboard with an older BIOS it still worked but the clockspeed was stuck much closer to 4Ghz than 5Ghz until the BIOS was updated.

Linus did say the BIOS wasn't updated in a while, that could be the issue but it seemed like it was hitting 5ghz while doing some tasks.

[MmRaZ]

28 minutes ago, dgschrei said:

OK I just had to come over to the forum for this:

 

Mistake #1 using AME.

 

Whatever I do it is consistently the slowest possible encoder I can get my hands on. If you have any chance of not using it take it.

 

Now I'm guessing you guys are stuck with it because you want to export Premiere project files. so assuming that is not an option:

 

Mistake #2: What actually seems to be killing your performance is rendering the videos over the network.

 

When Linus was testing the 14 core on the installed machine the project was accessed over a network drive . When he was doing the initial tests with the 16 core the video was rendering from C:\....  (I freeze framed the video to verify)

 

Why is this a problem?

If the encoder is not optimized for that it might read data in in tiny junks and also write the result out in tiny junk. Every time it does that that operation will suffer the full latency of the network. Now while this might be ~ 1ms in your LAN this still adds up VERY quickly and utterly kills your performance. Hence the way lower utilization on the render server with the 14 core CPU installed.

 

We had a similar problem in our company fetching all the build results (think tens of thousands of dlls and config files) from a remote server. It actually turned out that it was faster (about 2x-3x) to download a zip file of the build result and then unpack it locally, then it was to just copy the folder, because the latency of all those small accesses was killing the speed and we couldn't saturate the network that way.

 

So the solution might be to first copy the project to be rendered out locally onto the render machine and then encode it on there. That would get rid of all the latency and might give you a decent speedup. And even if it isn't faster for one job it will definitely increase your overall throughput.

 

 

P.S.: I still would advise to use some of the coding know how in your company to have someone just whip up an encoding script that just uses ffmpeg for the encoding if at all possible. In my experience that is the most stable option that produces the best results both in performance and quality terms. ffmpeg's h264 encoder x264 also should scale way better onto many cores (especially since you guys render 4k) than the sorry excuse for an encoder Adobe ships with AME.

Maybe Linus can have another video testing this? I love these videos on optimizing their editing/rendering workflow!

[AyoKeito]

1) iGPU may increase video decode speed (for AME), resulting in faster exports. I've definitely seen iGPU decoder being used in my renders. Install iGPU drivers for 9900K and (obviously) enable it. You don't have to switch from CUDA engine to make use of it.

2) AME is not the best encoder. Try using ffmpeg.

3) You may use NVIDIA GPU to encode your video using NVENC if time is very important. It will finish your renders in a minute or two. Use Voukoder plugin for Adobe or ffmpeg to do that. CUDA is way better than Intel's hardware acceleration and has much more settings, especially for H.264

Personally i gave up on AME long time ago. I just export to FFV1 using Voukoder plugin and re-encode video into final format using ffmpeg on my Linux server. That also eliminates any Windows issues out there. You can also try uploading FFV1 and see how fast it goes. It's a lossless codec with compression.

[Zmezoo]

I thought I would post a link back to this video to help explain why they are using the codecs and software there using, because it isn’t without thought.

 

https://youtu.be/YQ96fLCGFRw

[JoostinOnline]

When I saw the WAN Show was called "We made a video you guys hated", I assumed it was about this.

[xprototype]

Watched this video and all I wanted to find out is how many cpu cores and RAM are suited for rendering content in the Adobe suite, More so Adobe Media Encoder? Been trying to find the best cpu core count for performance and video output settings. I currently run x2 Intel Xeon 2699v4 22 core CPU's and no matter how many cores I assign in unraid (yes I'm using a 1060ti gpu) it always takes a long time to even render 1080/4K @60fps.

Hopefully any guru's in this space can advise the correct core count and ram for maximum rendering performance?