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

[rbjtech 5284]

1 hour ago, neik said:

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...

If it's an UHD7xx then yes, it should be able to do multiple 4k streams without issues - but do remember than all the other components in the system may need to do work as well - such as transcode HD audio for example, which is not a GPU function.

There is also a big difference between a 100Mbit/sec 4K Remux vs a 5-10Mbit 4K Stream ...

[neik 873]

On 2/7/2022 at 9:55 PM, softworkz said:

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

softworkz, did any of the updates hit beta during the last couple of releases?
I can't see anything in the changelog but it might have been changed in the background though.
Thank you!  

[softworkz 5066]

4 minutes ago, neik said:

softworkz, did any of the updates hit beta during the last couple of releases?
I can't see anything in the changelog but it might have been changed in the background though.
Thank you!  

I'm afraid, these things have been postponed until after the next release.
I'm sorry about the misinformation, I hadn't expected this.

[neik 873]

Thanks for the (quick) clarification!

[Gecko 71]

On 11/23/2021 at 12:46 PM, rbjtech said:

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

One question @rbjtech, what is the power consomption at idle for the CPU ? Let’s say from the socket wall without any drive spinning ?

 

because i’m currently thinking to upgrade from 3600x ryzen to an G7400 pentium to save money from my electricity bill and the G7400 has an UHD 710 in it.

I wonder if it could achieve the same level of 4k transcoding as you or… not a all. I don’t know what to take into consideration from the Technical sheets of intel like I used to do with the nvidia charts. But the 12700k seems overkill to simply download & transcode 4K remuxes to 4-5 people at the same time…

[rbjtech 5284]

The CPU (12700K) idle in my system is about 22W

The UHD710 in a G7400 still has exactly the same capabilities in terms of features as the UHD 770 but it has a slightly slower clock speed and half the amount of Execution and Shaders.  I believe this will just impact the 3D Performance, it should not actually impact the transcoding performance as that uses a different pipeline.

I would say that the G7400 using QuickSync will easily outperform the 3600x (using CPU Transcoding) for Transcoding - and at a fraction of the power. 

Remember though - As stated above, the CPU will still be responsible for any I/O, Audio Transcoding, Networking etc - so that will become the bottleneck with this 2 Core 4 Thread CPU - so on a G7400, I would say 2-3 x 4K REMUX transcodes will likely be the max it can handle - but I cannot say for sure. 

Edited March 30, 2022 by rbjtech

[Gecko 71]

Thank you for your detailed reply.

Did you underclocked or undervolted your cpu or is it the stock config using the power saving or balanced power plan ?
Which motherboard do you have ?

I've read that some user created a config with a G7400 that can go as low as 7 watts idling (for the entire system with no spinning disks) but 22 watts for the CPU is already twice as less as my 3600X (without taking into account an nvidia for the transcode part) 

2 hours ago, rbjtech said:

Remember though - As stated above, the CPU will still be responsible for any I/O, Audio Transcoding, Networking etc - so that will become the bottleneck with this 2 Core 4 Thread CPU - so on a G7400, I would say 2-3 x 4K REMUX transcodes will likely be the max it can handle - but I cannot say for sure. 

Correct, I tested pinning my 3600x to a single Core for Emby and direct playing a single 4K Remux is already taking 70% of it... The choice will be hard to do.
 

[rbjtech 5284]

1 hour ago, nSGecko said:

Did you underclocked or undervolted your cpu or is it the stock config using the power saving or balanced power plan ?

This is Stock on the High Performance Power Plan - Windows 11

1 hour ago, nSGecko said:

Which motherboard do you have ?

Gigabyte Z690 UD DDR4

There is actually a hardware build log here if you are interested ..

https://emby.media/community/index.php?/topic/103658-alder-lake-build/

1 hour ago, nSGecko said:

I've read that some user created a config with a G7400 that can go as low as 7 watts idling (for the entire system with no spinning disks) but 22 watts for the CPU is already twice as less as my 3600X (without taking into account an nvidia for the transcode part) 
 

Nice - I think the TDP for the G7400 is about 46W - so for a 4K Transcode, I'd imagine you'd be using at least 15-20W using QSV - with 5-10% CPU load.

If you look in the thread below - I had 7 x 64Mbit 4K Remux to 1080p QSV transcodes going - and my CPU was still only on 13% total.   If you are interested, I can look at the power usage for just one transcode if you want me to ? edit - see below

https://emby.media/community/index.php?/topic/107213-using-intel-transcoding/&do=findComment&comment=1128671

So above is a SINGLE 64Mbit 4K Transcodes > 1080p/10Mbit - GT is the GPU @ 4.8W

Above is TWO x 64Mbit 4K Transcodes > 1080p/10Mbit - note the GT is about the same (+0.1W) , CPU (IA) goes up to 37W (+5W)

 

Edited March 30, 2022 by rbjtech

[Gecko 71]

Wow all of this is really interesting !

I was looking for a rack case like this in the past few days, I'll keep that reference in mind when the time comes to upgrade

You are talking about 7 and 10 transcodes that you were able to achieve, I assume 10 was with using throttle and 7 was without throttling...
Anyway, the 7 simultaneous 4K transcode you achieved is unfortunately not my use case... When I need to transcode, it's because I'm not at home and streaming to a small device. Even using a TV from another location would trigger the transcode anyway because I lack a decent upload bandwidth. So my use case would be to also transcode Atmos/DTS-HD/... alongside the video... + overlaying French subtitle mainly in PGSSUB format, because I'm not fluent enough in English to follow a movie without activating them (French dubbing is unfortunately garbage since the 2000's).

I know that in the last beta, overlaying is done on the GPU instead of the CPU (at least using NVDEC). In the end, if you estimate that a G7400 will only transcode 2-3 streams (let's say with tone mapping and audio aswell), burning subtitles will maybe cost me one additional stream, so 1-2 at the end ? Would love to see another 12th gen user to compare that use case.

 

Currently with my 3600x + GTX 1660, 4 4K transcodes is my upper limit because of the nvidia VRAM that is very limited. The full system idle at 70 watts with only 2 drives attached. That's why I'm trying to find a more eco friendly alternative  

[rbjtech 5284]

If you want to list say 4 'scenarios' then I can test those out for you - ie 1) 4K 20Mbit Stream, DTS-HD, PGSSUBS to 1080p 4Mbit 2) ...

The Throttling will have a big impact if transcoding activities are not all started at the same time - ie as one buffer is filling (transcoding), another is just being read.  As long as the buffer re-fill is faster than the playback fps, then potentially you can run many more sessions - it's a logarithmic graph  - ie the more sessions you add, the less capacity you get for adding more.

As you noted, I think 4.7 will also bring many transcoding enhancements, the testing I am doing is on the 4.6 Release - so those numbers may go up.

I also wanted to be conservative on my estimate for the G7400 - it may well do 4 with ease, but I have nothing to back that up other than comparing to my 12700K experience - it may well do 7 (!) as it uses the same generation of integrated UHD.. 

Edited March 31, 2022 by rbjtech

[FredrikT 24]

I've just build a new workstation/server that replaced an existing Nvidia Shield. The server consists of;

i5-12500
32 GB 4000MHz RAM
Windows 11 Beta (22H2)
Emby Beta (4.7.0.32)

When I stream a 4k HDR True HD (Matrix Ressurections) to my mobile phone, and limit it to 20 MBit 1080p (as most external sessions will use this resolution), the CPU performance isn't the issue at all. But the GPU is working quite hard, averaging over 50% GPU. I'd hoped it would be less so at least two 4k streams to 1080p would be possible.

[Gecko 71]

6 minutes ago, FredrikT said:

When I stream a 4k HDR True HD (Matrix Ressurections) to my mobile phone, and limit it to 20 MBit 1080p (as most external sessions will use this resolution), the CPU performance isn't the issue at all. But the GPU is working quite hard, averaging over 50% GPU. I'd hoped it would be less so at least two 4k streams to 1080p would be possible.

I think I've read that Windows is not very optimized with the latest 12th gen for transcoding but that may be an old thread I've found. Anyway, like rbjtech said on his original topic, was that on windows, the 3D component of the GPU is not relevant while transcoding. His was maxing out at 100% at the first stream and stayed there. You should look at the decode/encode part of the GPU in the task manager

[FredrikT 24]

Just now, nSGecko said:

I think I've read that Windows is not very optimized with the latest 12th gen for transcoding but that may be an old thread I've found. Anyway, like rbjtech said on his original topic, was that on windows, the 3D component of the GPU is not relevant while transcoding. His was maxing out at 100% at the first stream and stayed there. You should look at the decode/encode part of the GPU in the task manager

I just read his thread and checked this as you wrote this, lol. I see though that he doesn't do tone mapping. Could make a difference. Here's a screenshot of the Video Decode. Seems I should be good for at least three simultaneous streams, which is enough for me

[rbjtech 5284]

24 minutes ago, FredrikT said:

I've just build a new workstation/server that replaced an existing Nvidia Shield. The server consists of;

i5-12500
32 GB 4000MHz RAM
Windows 11 Beta (22H2)
Emby Beta (4.7.0.32)

When I stream a 4k HDR True HD (Matrix Ressurections) to my mobile phone, and limit it to 20 MBit 1080p (as most external sessions will use this resolution), the CPU performance isn't the issue at all. But the GPU is working quite hard, averaging over 50% GPU. I'd hoped it would be less so at least two 4k streams to 1080p would be possible.

I think you may be missing the fact that you are transcoding this stream @ 134 fps !

So while the GPU is busy (70%) it is transcoding as FAST AS IT CAN - there is no throttling.

Therefore if you added a 2nd identical stream, @ 70% GPU, each stream would be transcoded @ 67fps - ie still way above the minimum required.

Extrapolating the 134 fps figure (with tone mapping), you are actually at 5.6 'streams' - so lets say 5 - which are identical to my findings (taking into consideration tone mapping)..

 

 

Edited March 31, 2022 by rbjtech

[Gecko 71]

28 minutes ago, rbjtech said:

If you want to list say 4 'scenarios' then I can test those out for you - ie 1) 4K 20Mbit Stream, DTS-HD, PGSSUBS to 1080p 4Mbit 2) ...

The Throttling will have a big impact if transcoding activities are not all started at the same time - ie as one buffer is filling (transcoding), another is just being read.  As long as the buffer re-fill is faster than the playback fps, then potentially you can run many more sessions - it's a logarithmic graph  - ie the more sessions you add, the less capacity you get for adding more.

As you noted, I think 4.7 will also bring many transcoding enhancements, the testing I am doing is on the 4.6 Release - so those numbers may go up.

I also wanted to be conservative on my estimate for the G7400 - it may well do 4 with ease, but I have nothing to back that up other than comparing to my 12700K experience - it may well do 7 (!) as it uses the same generation of integrated UHD.. 

That's very nice of you but I'll just bother you with 2 scenarios then

1) 4K HDR Remux Stream (> 50mbits) with DTS-HD or Atmos and PGSSUB to 1080p 10 Mbits (for my external users)
2) Same input file to 720p or 480p 1Mbits (for myself when commuting on 4G). For 480p, I simply wonder if the iGPU will kick in as it is an SD resolution ?

Maybe deactivate throttling is the best option here, but could you set your CPU to the most power saving mode you have on Windows ?
And if I recall correctly, on windows, you can define the affinity of any process you run, I know it sounds silly, but I would also be interested if you could try another run for scenario 1 by setting the ffmpeg process specifically to a single core (or two cores) (G7400 is only dual-core with no HT, so like you said, I may ran out of cpu power before the igpu gives up).

Again, many thanks for your time!

[Gecko 71]

42 minutes ago, rbjtech said:

I think you may be missing the fact that you are transcoding this stream @ 134 fps !

So while the GPU is busy (70%) it is transcoding as FAST AS IT CAN - there is no throttling.

Therefore if you added a 2nd identical stream, @ 70% GPU, each stream would be transcoded @ 67fps - ie still way above the minimum required.

Extrapolating the 134 fps figure (with tone mapping), you are actually at 5.6 'streams' - so lets say 5 - which are identical to my findings (taking into consideration tone mapping)..

Okay, I just realized how to extrapolate from that fps thing  hahaha silly me !

EDIT:

I disabled throttling and ran one 4K transcode to 10mbits (with tone mapping, pgssub and DTS-HD transcoded) : 124fps.
Adding a 2nd, 3rd and 4th stream with the same file and settings got me all streams at ~100 fps but my gpu decode engine is at 100%.
Does that means that if I had more memory available on the GPU, I could run a 5th, 6th,... stream until reaching less than 23ish fps on a single stream ?

During that test I also tried the following :
I locked my cpu at 2,2 Ghz on all 6 cores / 12T (it's my default setting). Overall cpu load is at ~50%. 100fps on all streams.
I tried pinning Emby to 2 cores only at the same frequency, only 3 streams were starting at only ~65 fps. The fourth refused to work and of course, 100% cpu load.
2 cores at 3,8 Ghz, the four streams are playing at ~80fps and the load is ~99%.
2 cores / 4threads at 3,8 Ghz (which is the closest to the G7400 as I can get), back to 90~100 fps on all streams and 60~70% cpu load.

Disabling PGSSUB did no change anything on my side in terms of FPS but the CPU load went down to 30%. I should do the same test on Emby Beta  
Changing from DTS-HD MA to DTS has nearly no impact on either FPS nor CPU load.

Edited March 31, 2022 by nSGecko

[FredrikT 24]

38 minutes ago, rbjtech said:

I think you may be missing the fact that you are transcoding this stream @ 134 fps !

So while the GPU is busy (70%) it is transcoding as FAST AS IT CAN - there is no throttling.

Therefore if you added a 2nd identical stream, @ 70% GPU, each stream would be transcoded @ 67fps - ie still way above the minimum required.

Extrapolating the 134 fps figure (with tone mapping), you are actually at 5.6 'streams' - so lets say 5 - which are identical to my findings (taking into consideration tone mapping)..

 

 

Ahh. It's doing more frames than it needs to to fill up the buffer!!! D'oh, didn't think of, or see that . Then it's really awesome, and I get about the performance I would expect. Thanks for clearing that up for me!

[FredrikT 24]

23 hours ago, nSGecko said:

Thank you for your detailed reply.

Did you underclocked or undervolted your cpu or is it the stock config using the power saving or balanced power plan ?
Which motherboard do you have ?

I've read that some user created a config with a G7400 that can go as low as 7 watts idling (for the entire system with no spinning disks) but 22 watts for the CPU is already twice as less as my 3600X (without taking into account an nvidia for the transcode part) 

Correct, I tested pinning my 3600x to a single Core for Emby and direct playing a single 4K Remux is already taking 70% of it... The choice will be hard to do.
 

Remember that you can't compare one virtual cpu on one platform to another platform. There are many factors that doesn't make this an apples to apples comparison. Such as MHz, CPU cache, bus, memory technology and speed, etc.

I went with an i5-12500 myself as this has the most powerful GPU (UHD 770) and lowest TDP 65Watts. Instead of making this a dedicated server as my old Emby/Plex servers were, I chose to make it a workstation instead so that I can work on it since it's being powered 24/7 anyways. So far it works great I have to say. It's even running a few (legacy) VMs in Hyper-V, and in a NZXT H1 (v1) it's super quiet with no temperature problems whatsoever.

Remember that subtitles can use a bit of CPU. My transcode from 4k HDR to 1080p with hardware transcoding and tone mapping used an average of 1% of my cpu resources according to Emby (see my screenshots). I tried now with subtilties, and the average CPU usage according to Emby is 3%.

[Gecko 71]

I've done the same choice as you did just before the pandemic, after weeks of researchs, I ended up buying the 3600X + GTX 1660 Super + 32 Go of RAM, putting all of this in an old Antec Case that was picking up dust and use Unraid as a 24/7 workstation  I was aiming to run one or two VMs for gaming and work and a bunch of Docker containers for Plex (at that time) + others things.

What I didn't know at that time, is that on one year, I would consume around 1500 kwh of electricity running it 24/7, mainly idling, let's be honest  Now that my electricity bill has double since that time, I'm again (and sooner than I anticipated) looking for a more economic solution, hence my presence in this topic.

For sure, the 12500, 12600 and 12700 are fantastic CPUs (and I'm jealous that I don't have one at home like you do ) but if I can reduce another 10 watts or more by choosing a more modest CPU, I'll do it (especially after comparing the price of those CPUs )

1 hour ago, FredrikT said:

Remember that subtitles can use a bit of CPU. My transcode from 4k HDR to 1080p with hardware transcoding and tone mapping used an average of 1% of my cpu resources according to Emby (see my screenshots). I tried now with subtilties, and the average CPU usage according to Emby is 3%.

PGSSUB or SRT ? On my side, the user session page is always showing me 0% for the CPU, I think it's broken when Emby is running on a docker environment. 

[FredrikT 24]

26 minutes ago, nSGecko said:

I've done the same choice as you did just before the pandemic, after weeks of researchs, I ended up buying the 3600X + GTX 1660 Super + 32 Go of RAM, putting all of this in an old Antec Case that was picking up dust and use Unraid as a 24/7 workstation  I was aiming to run one or two VMs for gaming and work and a bunch of Docker containers for Plex (at that time) + others things.

What I didn't know at that time, is that on one year, I would consume around 1500 kwh of electricity running it 24/7, mainly idling, let's be honest  Now that my electricity bill has double since that time, I'm again (and sooner than I anticipated) looking for a more economic solution, hence my presence in this topic.

For sure, the 12500, 12600 and 12700 are fantastic CPUs (and I'm jealous that I don't have one at home like you do ) but if I can reduce another 10 watts or more by choosing a more modest CPU, I'll do it (especially after comparing the price of those CPUs )

PGSSUB or SRT ? On my side, the user session page is always showing me 0% for the CPU, I think it's broken when Emby is running on a docker environment. 

I've a NAS as well, so I don't have to do the storage as well. But I feel your pain when it comes to the electricity bill!

I had a NAS with more or less the CPU you're looking at. It might be ok for Emby, but forget using that server for anything else.

The subtitle I tested was PGSSUB.

[rbjtech 5284]

So your lowest powered option is to actually just use a NAS - and as long as your media direct plays - you don't NEED transcoding and you don't need CPU threads.

Before my big upgrade, I has an i5 750 - no iGPU, no GPU - it was a headless server - it played 4K perfectly fine and I kept (I still do) a 1080p version of the same file.

The 'power' to spin a disk is exactly the same whether it's empty or full    Thus, don't transcode 4K unless you absolutely have to and bigger storage on lower powered kit, may be a much more efficient option ?

 

 

Edited March 31, 2022 by rbjtech

[Gecko 71]

Yeah yeah yeah both of you are right, I should also consider that option, increase my storage in order to get a 1080p copy of every 4K file I have... But I always found that solution to be a double pain in the a**. At least using Plex and his version system. I know that Emby also has his own conversion system, I should test it and re-consider my position.

@rbjtech, any advice on the best way to create that 1080p copy ?

@FredrikT, actually, my Synology consumes more watts than my unraid server  I'm at 90 watts idle, but it has an i3 inside and 12 disks, too bad synology doesn't allow to use the igpu on this model...

Thanks for your inputs guys !

[FredrikT 24]

1 minute ago, nSGecko said:

Yeah yeah yeah both of you are right, I should also consider that option, increase my storage in order to get a 1080p copy of every 4K file I have... But I always found that solution to be a double pain in the a**. At least using Plex and his version system. I know that Emby also has his own conversion system, I should test it and re-consider my position.

@rbjtech, any advice on the best way to create that 1080p copy ?

@FredrikT, actually, my Synology consumes more watts than my unraid server  I'm at 90 watts idle, but it has an i3 inside and 12 disks, too bad synology doesn't allow to use the igpu on this model...

Thanks for your inputs guys !

I decided to split the NAS from Plex/Emby, and have something dedicated for remote streaming. For a long time this was an Nvidia Shield. It worked ok until 4k HDR became a thing. My NAS has taken over almost all of the workload of my previous Hyper-V server, and runs various docker containers. I also got fed up with the fans of my laptop, so I replaced 1x Hyper-V server, 1x Nvidia Shield and 1x i7 laptop with an i5-12500 (it's SUPER fast!). I actually have Emby running in a container on the NAS as well for internal use only. I might get rid of it now as the i5 is plenty powerful for simultaneous internal and external use. The internal does on the fly library updates, downloads artwork, etc, etc. But it's probably overkill.

I haven't measured the watt consumption of the NAS nor the i5. Though I have contemplated building a custom 3 or 4U unraid/TrueNAS system for better performance, upgradability, fit (for my small rack), customization, etc. It might be the next project. Might use the old Hyper-V server to test unraid and TrueNAS. But I'm not a big fan of having too many eggs in one basket, though with the electricity prices....

[rbjtech 5284]

7 minutes ago, nSGecko said:

@rbjtech, any advice on the best way to create that 1080p copy ?

There are so many systems out there that do this, a recommendation is difficult.

If you want a fully automated system, then something like Tdarr is great - but if you just want manual conversions, then the likes of Handbrake, Ripbot264 even Emby itself are all good options.

The key thing to decide is your target codecs - h264 and AAC 2 Channel Stereo is a good 'baseline' to ensure no transcoding will ever be required.  If you have modern clients, then hevc is also a safe bet as long as you choose standard profiles.

But I would investigate the intel iGPU route first - I think you'll be surprised how far the iGPU has come for transcoding and combined with the Emby tonemapping etc - it works exceedingly well for just a little extra power usage over a direct play.   

[softworkz 5066]

Hi guys,

interesting conversation - a few remarks:

1. Don't trust any 0-100% indicator

Whether it is CPU usage in task manager, disk usage in performance monitor or any of those weird GPU metrics: 
The only truth you can gather from those is like a tristate "low", "average", "high". Beyond these, you can't take those values as for what we all would like them to look at.

Due to the way how modern CPUs are managing their own clock rates, switching certain cores on and off, applying boost clock rates to increase single-core performance (but in reality not applying this to a specific core, but circulating this over all instead)).

Disk I/O is highly dependent on the kind of access patterns being made, it's practically impossible to even define what "100%" would actually mean.

And for the GPU metrics: those are telling just about nothing. You can't know what these are measuring at all: Seeing one value at 30% could mean anything. It could mean that the operation is already at the maximum, because the limiting factor sits elsewhere and in the current scenario, it's just from internal processing relations that value A is 30% when value D is at 100% (a value that might not even be shown there). So those 30%, don't necessarily mean that there's capacity for doing three of those executions in parallel.

The opposite is possible as well: a value could be at 100% because that single operation is done in a - maybe super-fast or super-energy-saving - way, and the GPU does that as long as it can "afford" to do so, but as soon as the processing demand grows, it might choose a different method with higher throughput (but other downsides)..
So: 100% doesn't even necessarily mean that you're running at the max. capacity already.

2. Don't underestimate the effect of throttling

There have been mentions of throttling in a way that it can allow more simultaneous transcodings at a time, but there's a multiple times stronger effect and benefit involved when using throttling. Many of the considerations above were about system power input taken at a point in time, while the actual concern is this: Energy by time (e.g. 'Wh').

The time is equally (proportionally) important in this calculation and throttling can have a huge effect on that: When you have decent transcoding speeds, let's say 6.0x, and you have a video of 120 minutes, the video will be fully transcoded after 20 minutes. Now, assume you start watching 10 of such videos, and you'd watch each of them for 20 minutes, then your system will have transcoded each video to the end, which means, it has transcoded 1200 minutes of video.

With throttling switched on, it will only transcode up to 2 or 4 minutes going forward from your playback position. After watching those 10 videos in the same way as above, we got (20 +  4)  * 10 = 240 minutes 

=> This is what's saving energy (unlike below)

3. Never mess around with thread affinity of thread priority settings!

I can only assure you: whatever you try - it will just make things worse. Limiting cores for use by a certain process doesn't save energy, it costs energy.

4. Low TDP is not what you want

TDP doesn't tell anything about the efficiency of a CPU. It's nothing more but telling system integrators about the amount of heat that needs to be conducted away.

5. A "weak" CPU doesn't necessarily save power

Letting a CPU run at 100%is usually less effective than on average load.

When there's high pressure and queues for execution are loaded, the CPU needs to make compromises in a different way which involves more frequent and less optimized context switches for example.

Finally

I wouldn't rely on anything or anyone and would rather set up a test scenario over like 12h, connect a power meter between the wall plug and the cable and try different scenarios.
The number of influencing factors is just too high for good predictions to be made.