Server upgrade. Difference between iGPU in i5 11400 vs 11500 (HD730 Vs HD750)

[rbjtech 4283]

Thanks for taking the time to put these notes together @softworkz

I think it would be a great feature in the Advanced Transcoding section to have some sort of 'Benchmark' function that gives a transcoding metric ('EmbyMark') for your system.   

This would need to download various short copyright free media files to keep the test consistent (hosted on emby.media ?) - when a few of the guys and I talked about this a while back - this was the main hurdle - where to host the files ...

[softworkz 3338]

There is movement in this area. Just very recently I had began to work on a "Test Video Factory" for generating various kinds of uncopyrighted videos for testing.

Here's an example: Emby_TestVideo_x264_r25_g75_3840x2160_r1.zip

One advantage of these is that the content can be compressed pretty well, which results in small file sizes.

With regards, to benchmarking, the simple content shouldn't affect processing performance, I'm not sure about encoding and decoding performance, though. Maybe you could do some tests to see how this compares to "normal" videos?

Should it turn out that this content would cause distorted performance results, we'd need to see whether we can find something different (e.g. morphing fractals or similar).

Finally, there are also plans for having some automated execution of transcoding, probably delivered as a plugin. While the primary purpose is performing system tests on various platforms, the results would also be suitable for measuring performance.

[neik 837]

@rbjtech, iirc you are using Windows as OS, right?

Have you ever tried Linux to compare the transcoding performance?

Edited April 11, 2022 by neik

[rbjtech 4283]

1 minute ago, neik said:

@rbjtech, iirc you are using Windows as OS, right?

Have you ever tried Linux to compare the transcoding performance?

I haven't I'm afraid - never felt the need to.

[neik 837]

8 minutes ago, rbjtech said:

I haven't I'm afraid - never felt the need to.

Thank you!

@softworkz, do you have some insights on how Linux and Windows compare transcoding performance wise? Are they comparable?

 

[softworkz 3338]

7 hours ago, neik said:

@softworkz, do you have some insights on how Linux and Windows compare transcoding performance wise? Are they comparable?

Performance-wise it's largely the same. But the driver situation and delivery is much more reliable and stable on Windows. There's additional trouble on the horizon regarding newer Intel platforms on Linux, and in this regard at least, you'll be better off with Windows.

[Q-Droid 651]

2 hours ago, softworkz said:

Performance-wise it's largely the same. But the driver situation and delivery is much more reliable and stable on Windows. There's additional trouble on the horizon regarding newer Intel platforms on Linux, and in this regard at least, you'll be better off with Windows.

Any hints as to what it may be that's coming?

[softworkz 3338]

2 minutes ago, Q-Droid said:

Any hints as to what it may be that's coming?

FIRST

SECOND

Then there are issues on the Alder Lake platform, where only a part of it is resolved and only by a kernel that is not even released yet:

https://github.com/intel/media-driver/issues/1342

THIRD

And for Jasper Lake and Elkhart Lake it is required to have a specific kernel parameter enabled and certain firmware available (for the latter, the requirements are not even clear yet).

 

(None of those problems exist on Windows)

[Q-Droid 651]

Thanks. I was following the thread for your first but once it became evident that the issue wasn't configuration I didn't have anything else to offer. Seeing that you did not have the problem on the same family makes me suspect the hardware implementation (PC vendor) and/or firmware.

Regarding the others I agree. Keeping up is a challenge and have to be ready for that kind of stuff when jumping in. Good thing I'm cheap and even when buying new I don't want to pay for the latest gen. Keeping my 8th gen until it dies or can't handle the job.

 

[softworkz 3338]

1 hour ago, Q-Droid said:

Seeing that you did not have the problem on the same family makes me suspect the hardware implementation (PC vendor) and/or firmware.

At the hardware side, it's HP Envy vs. no-name, so this might play a role, but since the same issue is seen on ADL, it's unlikely to be singular exception only.

1 hour ago, Q-Droid said:

Keeping my 8th gen until it dies or can't handle the job.

Most likely, these things will be resolved over time, but as long as there's a requirement to use pre-release kernels (or any kernel that isn't tested and officially provided/supported by the Linux distribution), it's practically a no-go.

[neik 837]

@rbjtech, the transcoding results you've presented so far are all transcoding within Emby and therefore to h264 due to Emby not supporting any other codec to encode to.

I'm wondering now if you know how the performance behaves if you select HEVC as encoding codec?

[neik 837]

Just tried to test it on my notebook (KBL-R) using Emby's convert feature but unfortunately whenever I select HEVC as codec it does software encode the file.
I thought I could work around that but having a look at the available HW encoders it seems that Emby doesn't support HEVC HW encoding at all.

If you have a hint how to do some testing I would appreciate it.

[Q-Droid 651]

14 hours ago, neik said:

Just tried to test it on my notebook (KBL-R) using Emby's convert feature but unfortunately whenever I select HEVC as codec it does software encode the file.
I thought I could work around that but having a look at the available HW encoders it seems that Emby doesn't support HEVC HW encoding at all.

If you have a hint how to do some testing I would appreciate it.

Do you mean for a specific CPU/iGPU or in general? I have a Coffee Lake (8th gen) and it does HEVC HW encoding though the results are weird because Emby sort of does its own thing. 

Edited September 9, 2022 by Q-Droid

[neik 837]

11 hours ago, Q-Droid said:

Do you mean for a specific CPU/iGPU or in general?

For Intel iGPUs, doesn't really matter what generation, just wanted to get a "feeling" for HEVC encoding performance compared to h264 encoding.

11 hours ago, Q-Droid said:

I have a Coffee Lake (8th gen) and it does HEVC HW encoding though the results are weird because Emby sort of does its own thing. 

How do you do HEVC HW encoding using Emby?
I don't seem to find a way or maybe my i5-8250U doesn't support it?! Will check that...

Edit: According to this table my CPU should be capable of HEVC HW encoding:
https://en.wikipedia.org/wiki/Intel_Quick_Sync_Video#Hardware_decoding_and_encoding

Edited September 9, 2022 by neik

[Q-Droid 651]

For starters make sure the encoders are detected and enabled.

And if HEVC HDR then make sure the hardware tone mapping is enabled.

 

This is a conversion, I don't know if streaming to HEVC is even an option. I'd say performance is so-so on an i3-8100. The "doing its own thing" part was about how the output bitrate you give it is a mere suggestion

20:40:04.256 Input #0, matroska,webm, from '/mnt2/Media/Video/Other/Test/Logan (2017)/Logan_Test8.mkv':
20:40:04.256   Metadata:
20:40:04.256     title           : Logan 4K
20:40:04.256     ENCODER         : Lavf57.83.100
20:40:04.256   Duration: 00:10:00.21, start: 0.000000, bitrate: 38699 kb/s
20:40:04.256   Stream #0:0: Video: hevc (Main 10), yuv420p10le(tv, bt2020nc/bt2020/smpte2084), 3840x2160 [SAR 1:1 DAR 16:9], Level 153, 23.98 fps, 23.98 tbr, 1k tbn, Start-Time 0.238s (default)
20:40:04.257   Stream #0:1(eng): Audio: truehd, 48000 Hz, 7.1, s32 (24 bit) (default)

...

20:40:05.349 Output #0, matroska, to '/mnt2/Media-temp/TVHTS/emby/sync-temp/359/7742/4f5a1a38-eb53-41c6-90e7-8bbc9fab3364.mkv':
20:40:05.349   Metadata:
20:40:05.349     encoder         : Lavf59.17.100
20:40:05.349   Stream #0:0: Video: hevc, qsv(bt2020nc/bt2020/bt709, progressive), 1920x1080 [SAR 1:1 DAR 16:9], q=2-31, 1000 kb/s, 23.98 fps, 1k tbn
20:40:05.349     Metadata:
20:40:05.349       encoder         : Lavc59.21.100 hevc_qsv
20:40:05.349     Side data:
20:40:05.349       cpb: bitrate max/min/avg: 7616000/0/1000000 buffer size: 15232000 vbv_delay: N/A
20:40:05.349   Stream #0:1(eng): Audio: aac (LC) ([255][0][0][0] / 0x00FF), 48000 Hz, 5.1, fltp, 384 kb/s (default)
20:40:05.349     Metadata:
20:40:05.349       encoder         : Lavc59.21.100 aac

...

20:40:04.259 Stream mapping:
20:40:04.259   Stream #0:0 (hevc_qsv) -> vpp_qsv:default (graph 0)
20:40:04.259   hwmap:default (graph 0) -> Stream #0:0 (hevc_qsv)
20:40:04.259   Stream #0:1 -> #0:1 (truehd (native) -> aac (native))

...

20:44:15.395 20:44:15.395 elapsed=00:04:11.13 frame=14383 fps= 57 q=-0.0 Lsize=   92369kB time=00:09:59.70 bitrate=1261.8kbits/s throttle=off speed=2.39x
20:44:15.395 video:63905kB audio:28152kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.339051%
 

[neik 837]

I don't have the HEVC encoders and therefore it's done via SW.
That explains it, might be an old driver (24.20.100.6286) on my laptop but I dunno how to update as the I have the latest from the manufacturer and I can't install the latest (31.0.101.3413) Intel drivers as it sends me to the manufacturere during install.

[Q-Droid 651]

Is that when using the Intel DSA support tool on your laptop? If the mfg is a dead-end and the Intel tool won't update it then you might have to find a manual workaround for a newer driver even if not the current/latest.

 

[softworkz 3338]

12 hours ago, neik said:

I don't have the HEVC encoders and therefore it's done via SW.
That explains it, might be an old driver (24.20.100.6286) on my laptop but I dunno how to update as the I have the latest from the manufacturer and I can't install the latest (31.0.101.3413) Intel drivers as it sends me to the manufacturere during install.

I would check in the list list of installed programs and see whether there's the manufacturer driver listed which you could uninstall.

Otherwise you could uninstall the driver in device manager, then reboot, then install the Intel driver (but as @Q-Droid mentioned, not via DSA but by downloading the installer from the Intel website).

[RanmaCanada 345]

On 08/09/2022 at 12:06, neik said:

@rbjtech, the transcoding results you've presented so far are all transcoding within Emby and therefore to h264 due to Emby not supporting any other codec to encode to.

I'm wondering now if you know how the performance behaves if you select HEVC as encoding codec?

The performance will be worse because HEVC is more complex than AVC and thus it needs more computational power.  For example, nvenc encode speeds for HEVC are half that of it's encode speeds of AVC.  Sadly there is no easy to find results for Quicksync, so we need to make our own.

[Gecko 61]

@rbjtech, I finally got the opportunity to change my server and go blue team. At this moment, I have two configurations ready:

i5-13500 with 32 Go of RAM (2x 16Go) (currently my preference of choice)
I9-13900k with 64 Go of RAM (2x 32Go) (because it's the only 'K' version I'm able to get at this moment, would have rather chosen 13600 or 13700 instead...)

I've read that idle power consumption does not significantly change for 13th gen for any model you choose but my question here is more about how much ram is needed for Quicksync to transcode one 4K media with PGSSUB ? Using nvidia solution, VRAM is the limiting factor and I easily consume 1,5Gb per transcode. Using the latest beta and all the new upcoming features of @softworkz like OCR and text modification, even 5 to 6 Gb of VRAM is needed on nvidia to transcode a media with subtitles.

But would be the situation on Intel side for such scenarios and your recommendations ?

Thanks!

[FredrikT 22]

18 hours ago, Gecko said:

@rbjtech, I finally got the opportunity to change my server and go blue team. At this moment, I have two configurations ready:

i5-13500 with 32 Go of RAM (2x 16Go) (currently my preference of choice)
I9-13900k with 64 Go of RAM (2x 32Go) (because it's the only 'K' version I'm able to get at this moment, would have rather chosen 13600 or 13700 instead...)

I've read that idle power consumption does not significantly change for 13th gen for any model you choose but my question here is more about how much ram is needed for Quicksync to transcode one 4K media with PGSSUB ? Using nvidia solution, VRAM is the limiting factor and I easily consume 1,5Gb per transcode. Using the latest beta and all the new upcoming features of @softworkz like OCR and text modification, even 5 to 6 Gb of VRAM is needed on nvidia to transcode a media with subtitles.

But would be the situation on Intel side for such scenarios and your recommendations ?

Thanks!

I have a i5-12500 (cheapest processor with the fastest (i770) igpu) and 32GB RAM. I run Hyper-V (virtualization) and a lot of other stuff and this machine can easily transcode 5 4k HDR to 1080 SDR. It's all done in the iGPU. I really can't tell any difference with or without subs.

The i9-13900k will for sure draw a lot more power than the 13500!! At idle (PL1), the i9 consumes twice the power of the i5. At full load, with the "right" motherboard it can consume more than twice as much energy as the i5.

I have a family member in need of a new computer, and I'm contemplating giving them my mobo/CPU/DDR4 setup and going for an i5-13500 with DDR5 RAM. DDR5 has a form of ECC so it should be more stable.

[rbjtech 4283]

On 23/02/2023 at 14:19, Gecko said:

@rbjtech, I finally got the opportunity to change my server and go blue team. At this moment, I have two configurations ready:

i5-13500 with 32 Go of RAM (2x 16Go) (currently my preference of choice)
I9-13900k with 64 Go of RAM (2x 32Go) (because it's the only 'K' version I'm able to get at this moment, would have rather chosen 13600 or 13700 instead...)

I've read that idle power consumption does not significantly change for 13th gen for any model you choose but my question here is more about how much ram is needed for Quicksync to transcode one 4K media with PGSSUB ? Using nvidia solution, VRAM is the limiting factor and I easily consume 1,5Gb per transcode. Using the latest beta and all the new upcoming features of @softworkz like OCR and text modification, even 5 to 6 Gb of VRAM is needed on nvidia to transcode a media with subtitles.

But would be the situation on Intel side for such scenarios and your recommendations ?

Thanks!

The 13500 or even 13100 is all you will ever need for emby - I'd get 32Gb+ of memory (DDR4 or 5 whatever your mobo supports) but if you are on a budget - then 16Gb is still more than ample.

VRAM is not an issue for QSV, as it just uses system memory.  On my 64Gb RAM, it allocated 32Gb of that to be VRAM .. so it's never an issue for transcoding.

No point in getting a K processor unless you plan to overclock it - but do make sure you get one with the iGPU ... 

[Gecko 61]

Thanks both of you!

I ended up choosing the 13500 with a single 32gb stick of ram (as ddr5 has ecc and dual channel)

emby will not be the only thing running on this machine but that’s plenty of horse power anyway for my home usage.

[neik 837]

48 minutes ago, Gecko said:

single 32gb stick of ram (as ddr5 has ecc and dual channel)

Single stick and dual channel? Are you sure?

[Gecko 61]

2 hours ago, neik said:

Single stick and dual channel? Are you sure?

Apparently, there is no single Channel for ddr5 but you still have some gains with 2 sticks running in quad channel or whatever it is called.