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

[LunchieTey 0]

As per the title description, I'm considering updating(changing) my Ryzen/1050ti system to an 11th gen system with no discrete GPU to use quick-sync(which the Ryzen system can't do obviously)

I need better trans-coding performance as I'm now adding external connections that need bandwidth reduced a fair bit and the Little 2200G and 1050ti don't keep up unless I unlock the GPU for more streams which I don't want to do(I want the 1050ti for something else anyway). An 11400 based system will use less power as a bonus.

I'm trying to decide between an 11400 or 11500 for various reasons.

Firstly the 11400 should use less power than the 11500 at idle but obviously has lower peak performance(the server has other tasks other than Emby)

It's also a touch cheaper and should make less heat/noise under load with the stock cooler. Heat is important as I'm adding more drives and rack mounting the server in a glass fronted rack with my NVR0, Mikro Tik Router,Modem and UPS

My only real question is with quick-sync ability. The 11400 uses HD730 vs the 11500 HD750. The 750 has 12 more EU's BUT it's my understanding that quick-sync does NOT use the execution units to trans-code and that it's a separate part of the iGPU. I'm led to believe the Quick-sync engine is the same between the 11400 and 11500.

Does anyone know if this is true?

 

[Carlo 4330]

@softworkz got any feedback on this?

[rbjtech 4223]

For the small difference in price, why risk it - I would get the 11500

https://ark.intel.com/content/www/us/en/ark/compare.html?productIds=212277,212270

[rbjtech 4223]

What did you go for in the end ?

I've now using a UHD 770 in a 12700K - and transcoding performance on it is frankly amazing.

Comparing to the Nvidia charts - it's has RTX 4000 performance on 4K Transcodes.... 

I comfortably hit ten x 4K Remux (70-80Mbit.sec) to 10 Mbit 1080p - PLUS Tonemapping !

I think the UHD 750 in the 11th Gen is way up there too - for an iGPU/trancoding engine - it's a no brainer for a media server.

I've yet to try 1080p transcoding in high numbers, but I believe we are talking 40-50 sessions before it will choke - wow .. 

edit - 1080p 10Mb/sec > SD (480p/1Mb/sec) is over 1020 fps - hahaha..

Edited November 23, 2021 by rbjtech

[RanmaCanada 337]

From what I gather on Doom9, the asics are the same across the chips of each generation, so it doesn't matter if you have a pentium gold or an i7, the quicksync asics are apparently the same.  The only difference could be inferred that the better processor allows more encodes due to it's ability to encode more audio streams with its raw horsepower.  But as I have not been able to find tests where someone tries the exact same encode across every single processor in a generation (as has been done in the past with NVENC), this is just extrapolating data from multiple posts and putting it together to make an educated guess.

[rbjtech 4223]

By each generation, do you mean iGPU generation ?  If so, then yes I think that is correct - as the 11th Gen UHD 750 is in the same ballpark as the UHD770 on the 12th Gen.

But there is a massive difference on say a HD520 vs a HD650.

I have an i7 6500K - it's still a power CPU, but it's HD520 is pretty pathetic in terms of 4K transcodes, infact it just about manages 1 with tonemapping. 

I actually said something very similar a few hours ago (I'll find the thread in a sec) - wouldn't it be great to have a wiki/KB article giving approx performance for transcoding for all the different GPU's and iGPU's for emby (not Plex..).

edit - ah yea, it was related to tone mapping - here :-

https://emby.media/community/index.php?/topic/103504-tone-mapping/

Edited November 30, 2021 by rbjtech

[neik 835]

On 11/30/2021 at 4:38 PM, rbjtech said:

I have an i7 6500K - it's still a power CPU, but it's HD520 is pretty pathetic in terms of 4K transcodes, infact it just about manages 1 with tonemapping. 

What's your experience when transcoding 1080p streams instead of 4k?

[rbjtech 4223]

On 30/01/2022 at 12:05, neik said:

What's your experience when transcoding 1080p streams instead of 4k?

.. on an i7 6500K or i7 12700K ?

The HD520 (in the i7 6500K) is 'weak' vs the HD6xx - it'll do 1x 4K stream and maybe 4-5x 1080p streams - but no more.  Tonemapping also cripples it.

From the Tonemapping work we did with @softworkz - the HD6xx is a big leap forward vs HD520 but I've never owned one - I skipped a generation and went from HD520 to a HD770 ..  

 

[neik 835]

I just noticed that I have a Kaby Lake-R (UHD 6xx) myself and went for it myself.
Don't need many transcodes at the same time 3-4 max and only 1080p and was impressed with the performance of my 15W cpu.

For testing I used a 20mbit 1080p -> 5mbit 1080p and and also a 4k (HDR) -> 5mbit 1080p and there results were always >150fps - quite impressive I must say.
Using the throttle I can imagine this being able to transcode a couple of streams.

Also did a test on a Ryzen 5500U but on 4K unfortunately it seems to only be able to transcode 1 stream (~ 35-40fps) and no tone-mapping is available (yet).
On the other side for 1080p I had similar results as with my Intel i5-8250U.

[neik 835]

@rbjtech, maybe you can answer this:
Do you know what happens with graphical subs (PGS/VOBSUB) when HW transcoding, is the HW transcoder able to burn them in or is this still done in software?
Forgot to test that one and already deleted my VM I used for testing...

[rbjtech 4223]

2 hours ago, neik said:

@rbjtech, maybe you can answer this:
Do you know what happens with graphical subs (PGS/VOBSUB) when HW transcoding, is the HW transcoder able to burn them in or is this still done in software?
Forgot to test that one and already deleted my VM I used for testing...

With the latest BETA, it's done in hardware (using QSV) - I'm not 100% sure if using NVENC

[GrimReaper 3271]

46 minutes ago, rbjtech said:

With the latest BETA, it's done in hardware (using QSV) - I'm not 100% sure if using NVENC

That's Overlaying, AFAIK Burning-in is always done is Software only, including latest Beta, no HW. @softworkz

[rbjtech 4223]

Yep - appears a Burn-In resorts to software only.

I avoid any sort of Graphical Subtitles for this reason - SRT seems to be the most flexible for any sort of playback on any devices - thus convert ahead of playback where you can.

I notice the new beta actually has some OCR features in it now - I may take a look - it appears to be using tesseract ..  

Edited February 2, 2022 by rbjtech

[softworkz 3326]

Hi all,

many things have changed and are new. Some of them are around for quite a while in the beta, but have remained widely unnoticed. But another large package of changes is just being added right now.

Please wait until the next beta before trying anything as the new things are not consistent in the current beta.

There will also be a series of videos, explaining the new features in detail.

sw

[neik 835]

2 hours ago, rbjtech said:

Yep - appears a Burn-In resorts to software only.

I avoid any sort of Graphical Subtitles for this reason - SRT seems to be the most flexible for any sort of playback on any devices - thus convert ahead of playback where you can.

I notice the new beta actually has some OCR features in it now - I may take a look - it appears to be using tesseract ..  

Have you had a look at the logs?

I just spun up my VM again to see what it does and it seems to attempt to HW transcode but for some reason it cannot allocate memory although I have 3GB of memory spare.
Log is attached.

 

1 hour ago, softworkz said:

Hi all,

many things have changed and are new. Some of them are around for quite a while in the beta, but have remained widely unnoticed. But another large package of changes is just being added right now.

Please wait until the next beta before trying anything as the new things are not consistent in the current beta.

There will also be a series of videos, explaining the new features in detail.

sw

Oh... that sounds good.
Looking forward to your surprise box!  

ffmpeg-transcode-caca37d6-9bf5-4ff9-8d7c-6424886ccfa5_1.txt

[softworkz 3326]

1 hour ago, neik said:

I just spun up my VM again to see what it does and it seems to attempt to HW transcode but for some reason it cannot allocate memory although I have 3GB of memory spare.
Log is attached.

This is known. It should be fixed in the next beta. 

1 hour ago, neik said:

Oh... that sounds good.
Looking forward to your surprise box!  

It really is one..

[lightsout 144]

On 11/23/2021 at 3:46 AM, rbjtech said:

What did you go for in the end ?

I've now using a UHD 770 in a 12700K - and transcoding performance on it is frankly amazing.

Comparing to the Nvidia charts - it's has RTX 4000 performance on 4K Transcodes.... 

I comfortably hit ten x 4K Remux (70-80Mbit.sec) to 10 Mbit 1080p - PLUS Tonemapping !

I think the UHD 750 in the 11th Gen is way up there too - for an iGPU/trancoding engine - it's a no brainer for a media server.

I've yet to try 1080p transcoding in high numbers, but I believe we are talking 40-50 sessions before it will choke - wow .. 

edit - 1080p 10Mb/sec > SD (480p/1Mb/sec) is over 1020 fps - hahaha..

Wait what?? This is insane. How is that possible, that is really impressive. I have considered adding a gpu, but at this point seems better to just jump to intel. Although my 1800x is doing the job pretty well for now. Thats really amazing considering how much power tonemapping usually eats on a gpu. Maybe this is done a different way? I thought with MadVR we needed a gpu to tonemap one stream, maybe that is because all the other added resources of madvr?

[softworkz 3326]

4 hours ago, lightsout said:

Maybe this is done a different way?

The same basic algorithms but implemented in a more efficient way.

4 hours ago, lightsout said:

I thought with MadVR we needed a gpu to tonemap one stream

I don't know MadVR at all, but you need to consider the following:

• Scaling is much cheaper than TM, so we are scaling down before tone mapping
• There's a quadratic relation between video size (measured by single dimension like height) and the number of pixels (or area)
  the latter in turn, linearly relates to the required video processing effort
• The example mentions 480p. When we compare that with a 4k video having 2160p, that's 4.5x the height and 20.25x the area of a 480p image.
  This means that tone mapping a video to  2160p (output size!) requires 20x more processing power for tone mapping (per image) than for a 480p output video.

PS: Don't try to calculate 1020fps / 20.25 - that's totally wrong for many reasons.

Edited February 4, 2022 by softworkz

[LunchieTey 0]

On 11/23/2021 at 7:46 PM, rbjtech said:

What did you go for in the end ?

I actually just put a Ryzen 3600 I had laying about in the existing system, removed the 1050ti and installed a spare 710 I had in my junk pile for video out, 1050ti is now in my 6 year Olds PC on Roblox duty

So far it's managed fine to keep friends/family encoding under control with plain old CPU encoding.

[lightsout 144]

4 hours ago, softworkz said:

The same basic algorithms but implemented in a more efficient way.

I don't know MadVR at all, but you need to consider the following:

• Scaling is much cheaper than TM, so we are scaling down before tone mapping
• There's a quadratic relation between video size (measured by single dimension like height) and the number of pixels (or area)
  the latter in turn, linearly relates to the required video processing effort
• The example mentions 480p. When we compare that with a 4k video having 2160p, that's 4.5x the height and 20.25x the area of a 480p image.
  This means that tone mapping a video to  2160p (output size!) requires 20x more processing power for tone mapping (per image) than for a 480p output video.

PS: Don't try to calculate 1020fps / 20.25 - that's totally wrong for many reasons.

Ok this is helpful. I guess the question is could this thing tone map one movie and keep it at 2160p, for instance of someone had an SDR 4k tv. Seems like it should be doable.

[neik 835]

On 2/2/2022 at 4:08 PM, softworkz said:

This is known. It should be fixed in the next beta. 

I can still see memory allocation issue (had >2gb free) with the latest .23 beta.
Did the fix make the way into .23 or will it be in a subsequent release?

[softworkz 3326]

10 hours ago, neik said:

I can still see memory allocation issue (had >2gb free) with the latest .23 beta.
Did the fix make the way into .23 or will it be in a subsequent release?

Right. It's done already, but the .23 doesn't have the updated ffmpeg yet.

[softworkz 3326]

On 2/2/2022 at 1:17 PM, softworkz said:

Hi all,

many things have changed and are new. Some of them are around for quite a while in the beta, but have remained widely unnoticed. But another large package of changes is just being added right now.

Please wait until the next beta before trying anything as the new things are not consistent in the current beta.

There will also be a series of videos, explaining the new features in detail.

Just fyi, the whole thing is not ready yet, in .23.

[FredrikT 22]

I'm more or less in the same boat as a few of you guys. I'm currently using the original Nvidia Shield (upgraded to SSD) as a Plex and Emby server. The Emby server seems to work better (after the external clients set the bit rate manually - auto doesn't work) than Plex, but much of my content is 4k HDR.

I'm contemplating building a PC for Emby for external access (I have a NAS without GPU for direct streaming on the internal LANs), and was originally looking at an i5-10400 or a i5-12400. I just realized that the i5-11400 has the UHD 750, which is a bit more powerful than the UHD-730.

I'm hoping that a i5-1140 will be a lot better than the Shield TV at HW transcoding (plus I can remotely upgrade/restart Emby) for 4k HDR on a 30 MBit outgoing WAN connection. I think I'm at two simultaneous streams at most (until I get fibre).

Edited February 9, 2022 by FredrikT

[neik 835]

18 hours ago, FredrikT said:

I'm hoping that a i5-1140 will be a lot better than the Shield TV at HW transcoding (plus I can remotely upgrade/restart Emby) for 4k HDR on a 30 MBit outgoing WAN connection. I think I'm at two simultaneous streams at most (until I get fibre).

From what I've seen so far I would strongly assume that it can easily hold up with two streams at a time - probably even a few more.
I would even my i5-8250U (older notebook cpu) would keep up with it.

Only assumption based on my results though...