-= CALLING ALL EMBY USERS =-

[Carlo 4330]

I never even saw the timezone link or just ignored it which is more likely.

Oh well, my intentions were to help.

 

Sounds like your Nvidia board is doing what you needed it to do,

[pir8radio 1292]

I never even saw the timezone link or just ignored it which is more likely.

Oh well, my intentions were to help.

 

Sounds like your Nvidia board is doing what you needed it to do,

 

Yea i was being funny...   sorry.....    I appreciate it...    Yea pretty happy with it...  I have room for 3 more, so maybe after some more hevc testing, i might go with one more just to cover a few streams of hevc 4k...

[notla49285 46]

Wow, for comparison my server seemingly can't handle two streams of the same episode within the internal network. FML.

[Fratopolis 62]

---deleted---

Edited February 3, 2019 by Fratopolis

[pir8radio 1292]

Well, I only had 35 streams going, all to the internet (remotely), no reported issues, ffmpeg logs all kept above the 30fps. This included a few direct streams. I have yet to max anything out except the GPU on 4k HEVC, and i get about 28-ish transcodes from 1080 20mbps to 1080 & 720 4mbps before the GPU fps starts to reach that of the video. I still need to graph all of this to come up with something that might be useful for the forum.

 

This is how many emby streams my server took on, you can see i usually never have more than 5-8 concurrent streams, but I was able to hit 35 on the test day!

Edited February 4, 2019 by pir8radio

[Carlo 4330]

Nicely done!

[Carlo 4330]

@pir8radio

For the tests above did you have Enable Throttling turned on?

Since that was 2.5 years ago, have you thought of trying this experiment again?

Now that we have tone mapping for 4K HDR media that should be a new test that shouldn't need as many people for the first round of testing.

What would also be interesting is take a couple of public domain videos and rename them to get 25+ or so different files. If we pre-setup who's going to participate and how many streams they will be able to use we can assign file names to play for each participant. That could help giving a more realistic HDD IO that resembles something closer to true life vs everyone using the same file that could get cached to some degree by OS caches.

It would be interesting to see the affects of bitrate and resolution coming into play as well.  Such as a 4K SDR 40Mb file being transcoded to 1080 10Mb.  Same to 5Mb.
1080p 20Mb to 1080 10mb and also 1080 5mb
1080p 20Mb to 720 5mb or similar

[pir8radio 1292]

36 minutes ago, cayars said:

@pir8radio

For the tests above did you have Enable Throttling turned on?

Since that was 2.5 years ago, have you thought of trying this experiment again?

Now that we have tone mapping for 4K HDR media that should be a new test that shouldn't need as many people for the first round of testing.

What would also be interesting is take a couple of public domain videos and rename them to get 25+ or so different files. If we pre-setup who's going to participate and how many streams they will be able to use we can assign file names to play for each participant. That could help giving a more realistic HDD IO that resembles something closer to true life vs everyone using the same file that could get cached to some degree by OS caches.

It would be interesting to see the affects of bitrate and resolution coming into play as well.  Such as a 4K SDR 40Mb file being transcoded to 1080 10Mb.  Same to 5Mb.
1080p 20Mb to 1080 10mb and also 1080 5mb
1080p 20Mb to 720 5mb or similar

1. Yes Throttling was turned on.     
2. I would be game for giving it another go..  always like to know the limits!
3. yea need to plan that one out if we wanted to do a full real test.. 

[Carlo 4330]

I've got gigabyte in my location via Verizon and also a 400Mb Comcast cable connection (Phila area), so between devices and computers depending on how you want to test I should be able to do anywhere from 5 to 15 connections myself. Just depends on the stream type and if we went to use only devices/apps or also browsers. If using browsers as well then how many tabs at once per computer. In testing my "gigabit" connection is normally only good for 850Mb upload and download according to speedtest.net.

I might be interested in doing a similar test as well with a couple of different servers to see what the limits are as well.
One is a Synology 920+ NAS with QuickSync.
Another is an old i7 with an unlocked Nvidia 1650 GPU.
Lastly my i7 notebook with both Quicksync and a Nvidia RTX 2070 (two different sets of tests).

The thing to make this useful is being able to graph usage which I don't have setup.
It would be cool to be able to graph connections/streams, HDD performance, Internet usage and anything else useful.

What did you use to graph your connections?

What would be useful is to have a discord channel or similar real time chat going on be able to communicate in real time with people and gradually increase load or ramp up.

Edited September 19, 2021 by cayars

[Painkiller8818 203]

Maybe this could also help to get a feeling for how much you should be able.

Its officially a plex GPU cheat sheet for NVIDIA GPU but maybe this could help you.

RTX not listed in there. So i also have no idea how much i can do with my i9 CPU and RTX 3060TI Mini v2 in my NUC

https://www.elpamsoft.com/?p=Plex-Hardware-Transcoding

Edited September 19, 2021 by Painkiller8818

[Carlo 4330]

4 minutes ago, Painkiller8818 said:

https://www.elpamsoft.com/?p=Plex-Hardware-Transcoding

I've seen similar graphs of usage but have always wondered if these are mathematical projected streams or if this was actually tested.
https://www.era20tech.com/the-best-gpu-for-plex-transcoding/
https://nascompares.com/best-graphics-cards-gpu-for-4k-plex-video-transcoding/

[rbjtech 4257]

From the recent tone mapping thread - the number of 4K > 1080p transcodes is pretty much directly related to the amount of RAM on the GPU.

The more RAM it has, the more transcodes it can do but it also relies on the number of 'encoder/decoder' chips the card has - this is not at all related to the number of CUDA cores.

As an example - my old GTX 1070 has 8Gb of Memory and 2 x enc/dec chips - from a purely transcoding performance, it is therefore capable of more transcodes that an RTX 2060Ti (with 6 Gb) ... it may even be equal to a 3060Ti (as that has 8Gb I believe).

So don't automatically think that because you have the latest and greatest card, you can do more transcodes - because it doesn't work like that.

[Carlo 4330]

I'm not sure that's totally true as I don't think tone mapping requires the memory you think it does.
@softworkz covers this it in this post.

 

[pir8radio 1292]

On 9/19/2021 at 1:41 PM, cayars said:

I've got gigabyte in my location via Verizon and also a 400Mb Comcast cable connection (Phila area), so between devices and computers depending on how you want to test I should be able to do anywhere from 5 to 15 connections myself. Just depends on the stream type and if we went to use only devices/apps or also browsers. If using browsers as well then how many tabs at once per computer. In testing my "gigabit" connection is normally only good for 850Mb upload and download according to speedtest.net.

I might be interested in doing a similar test as well with a couple of different servers to see what the limits are as well.
One is a Synology 920+ NAS with QuickSync.
Another is an old i7 with an unlocked Nvidia 1650 GPU.
Lastly my i7 notebook with both Quicksync and a Nvidia RTX 2070 (two different sets of tests).

The thing to make this useful is being able to graph usage which I don't have setup.
It would be cool to be able to graph connections/streams, HDD performance, Internet usage and anything else useful.

What did you use to graph your connections?

What would be useful is to have a discord channel or similar real time chat going on be able to communicate in real time with people and gradually increase load or ramp up.

yea so i saw these "Charts" and they are reasonably accurate. for gpu RAM...   but i have noticed embys "throttle" function really does help on the transcoding front.    gives you some cushion to take on more users...   

Im now using SPLUNK to do my monitoring..   Someone made me a plugin that writes current streams into the emby log and i grab that.   I also log drive, cpu, ram, GPU, and GPU RAM, etc...    I log FFMPEG logs and pull numbers from that, and i log the nvidia GPU stuff.. and those two logs look like this:     

well i log EVERYTHING...   on my server..  lol, but splunk is cool because you can tie all of those logs together to get the big picture..  For example a movie starts streaming, i can put graphs up that show Hard drive throughput, internet throughput, CPU, GPU, and RAM, hard drive and iscsi array throughput, as well as errors and users, what type of client they are using.. I a bit anal when it comes to data collection...   

Spoiler

 

 

 

 

 

 

I want to test,  i think i have the internet bandwidth to support whatever the bottleneck should be my drives or transcoding.      

 

[Carlo 4330]

Which specific OS are you running?
Those are really nice charts.

When I was running on Windows Server I was monitoring a lot of things as well but now that I'm running on Synology I'm not.  Haven't dug into to see what's needed or how to get perf metrics.

Yes after looking at those GPU charts I think they are in the ballpark.  At least they give you a relative difference between the cards which is useful when purchasing.

Of course things like burning in subtitles could throw it off a bit.  But with the way Emby is/and going to handle image based subs this should give Emby another advantage over other media platforms when transcoding.

[pir8radio 1292]

4 minutes ago, cayars said:

Which specific OS are you running?
Those are really nice charts.

When I was running on Windows Server I was monitoring a lot of things as well but now that I'm running on Synology I'm not.  Haven't dug into to see what's needed or how to get perf metrics.

Yes after looking at those GPU charts I think they are in the ballpark.  At least they give you a relative difference between the cards which is useful when purchasing.

Of course things like burning in subtitles could throw it off a bit.  But with the way Emby is/and going to handle image based subs this should give Emby another advantage over other media platforms when transcoding.

windows.    im a windows fan...   

 

[Carlo 4330]

I figured that from your image in the footer of your messages.  What I was really asking is if you are running 10 or Server (which version).

[rbjtech 4257]

8 hours ago, cayars said:

I'm not sure that's totally true as I don't think tone mapping requires the memory you think it does.
@softworkz covers this it in this post.

 

I wasn't saying the tone mapping used the memory per say, I said in that thread is where we did all the transcoding (with tonemapping) testing - you may recall I had something like 10 transcoding sessions going (4K > 1080 @4Mbit/sec) without issues - then I hit the GPU memory limit (and I seem to recall CPU was pretty max'd at that stage too).

[pir8radio 1292]

23 minutes ago, rbjtech said:

I wasn't saying the tone mapping used the memory per say, I said in that thread is where we did all the transcoding (with tonemapping) testing - you may recall I had something like 10 transcoding sessions going (4K > 1080 @4Mbit/sec) without issues - then I hit the GPU memory limit (and I seem to recall CPU was pretty max'd at that stage too).

yea I recall some memory getting eaten up on my 10bit 4k testing.     GPU memory was the first to max out before GPU processing %.

[Carlo 4330]

I really can't remember but I too thought it used more memory until I re-read softworkz's post which I linked to above.

If I remember correctly certain implementations can use a lot more memory such as VAAPI and maybe DX11VA, but QuickSync & NVENC use less. Something like that seems to ring a bell.

@softworkz can you help us out on this and how important the memory is on the GPU? Also how this memory compares on QuickSync.

[TiaanGR 2]

Hi there,

Fairly new to emby and I found a nice deal on a Dual Xeon CPU Server but I have a few doubts. I am hoping you can maybe assist as I read through your posts from the last 2.5 years.

@pir8radio I only have a few questions for you.

1. What do you think of the configuration of the attached imaged?

2. Would you suggest I add more RAM? i am not to concerned about the Storage for now as that I can upgrade as I need but a SSD for fast booting i want to do from the start.

So honestly this is what I am eyeing here and compared to what you have what do you think? Reason I ask is I am scared if I need to do Hardware Transcoding the CPU's will fail and then I wasted a lot of money on this instead of buying a new pc. The OS i want to do is Debian or Xubuntu something that is light on resources.

At most I have maybe 5 or 6 streams going but if I add more do you think it will be affected?

 

Kind Regards,

Tiaan

[pwhodges 1527]

Those CPUs are over ten years old, and have no video capability; by modern standards they use a lot of power for their modest performance.  You will probably need hardware for transcoding, and I'd want to check that the "server" case has space for a suitable video card.

Paul

[rbjtech 4257]

As Paul says - these are just very old power hungry data center servers that somebody is trying to make a quick buck on.  They frankly need to go into a skip...

Memory will be ECC memory which will likely be old DDR3 - non re-useable in anything.

A 10th/11th Gen i3/5/7 will simply outperform it for all transcoding duties at a fraction of the power as it has an iGPU (quicksync).

[TiaanGR 2]

Thanks for the feedback guys. I suspected more or less the same thing so I actually got a quote for somethig else and Damn i would rather that exstra bit and put it towards the other option of a whole new pc

[Carlo 4330]

Just to give you an idea of the "muscle" those CPUs have.  Combined they have a Passmark score of 359 with 8 total cores and 8 total threads.
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Xeon+E5606+%40+2.13GHz&id=1247&cpuCount=2

Combined they are about equal to an i3-6100 circa 2015 or a more modern 2020 Intel Pentium Silver J5040 @ 2.00GHz Passmark 3586 4 cores 4 threads
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Pentium+Silver+J5040+%40+2.00GHz&id=3665

The Pentium CPU uses 10 watts of power and the 2 Xeons use 160 watts (together).  The Pentium has an onboard Intel UHD Graphics 605 the Xeons don't.
It gets worse.  For Single Thread Rating rating score of the Pentium is 1421 while the Xeons are close to half at 813.  This can be quite important when you have something like ffmpeg that will only run some codecs on a couple of cores.

Of course that's just the CPU being talked about.  The IO will be worse on the older system as well across the board and 4GB of slow memory isn't much. Dual Xeons regardless of age don't come into their own until you feed them lots of memory.

 

I don't know if you just happen to come across that system or if you're activity looking for a new system.  Let us know and we can likely point you in a much better direction.
Also what's wrong with the current PC running Emby? What's the specs of the current machine.  Maybe an upgrade is a better solution.

It of course depends on how much you want to spend but dollar for dollar the new Alder Lake Intel i5-12600K is hard to top. It's a unique new design that has 6 Performance Cores (P-Cores) and four Efficient Cores (E-Cores) with a total of 16 threads. Each P core has a hyper thread while the E-cores do not.6+6+4=16 threads.  It actually has more physical cores than a Core i9 11900K (previous gen).  The single core performance is very close to it's big brother the new i9.  This i5 scores far better than AMD's coveted 5950X.

It can use the new really fast DDR5 RAM and supports the new PCIe 4 buss for additional speed.
The Core i5 12600K is the best CPU for gaming right now in general but surely for the dollar nothing comes close.

This is the K model which K means unlocked and can be overclocked with onboard sensors that throttle down the CPU is overclock settings would hurt the chip.  You can get another 10 to 17% out of it compared to the benchmarks ran at default speeds.  Best of all is that Intel is pricing the new Alder Lake chips very aggressively.  The i5 is currently selling for about 1/2 what it normally would sell for.  Intel is trying to stick it to AMD. The K version is usually the most expensive chip compared to other versions and right now this chip brand new is selling around $279-$299 and maybe even a bit cheaper with Holiday sales or coupons. You can get them for about $250 if you look hard enough. 

A top of the line Gamers edition Z609 Motherboard supporting DDR5 memory with all the latest unlocked bios, 4 x Dual-Channel DDR5 RAM Slots, 3 onboard M.2 PCI memory slots, extra PCI 5 slots, 40GB Thunderbolt, etc will set you back about $500.

Assuming it's not for gaming and you don't need the M.2 slots of board, don't mind the auto overclock for about 10% over defaults will cost about $160 which is a huge difference in price.  The high end boards are way overkill unless you want to use the expensive DDR5 memory and PCIe 4 M2/SSDs.

So for roughly $450-$500 you have the two main components of a fire breathing new computer that could possibly use every other component in your current system.
I forgot to mention this chip has the new Intel UHD Graphics 770 graphics on board so no need for a separate GPU for Emby use.

BTW, this chip has a Passmark CPU score of 24879 at default clock rates so with just auto-overclocking turned on you are well over 25K.

I don't know if this type of setup is out of your price range or not.

$199 HP EliteDesk 800 G2 Desktop, Intel Core i5 6500 3.2Ghz, 4GB DDR4 RAM, 500GB Hard Drive, Windows 10 Home, Intel HD Graphics 530
Passmark 5044
https://www.newegg.com/hp-elitedesk-800-g2-business-desktops-workstations/p/1VK-001E-1WCE8?Description=hp elitedesk 800 g2&cm_re=hp_elitedesk 800 g2-_-1VK-001E-1WCE8-_-Product&quicklink=true

Maybe you might even want to look at small form factor PC such as
$199 Intel Quad Core i3-6100T, 8GB DDR4 Ram, 256GB SSD with HD 530 Graphics
Passmark 3579 (check on the HD 530 for transcoding in the forums)
https://www.amazon.com/dp/B082MPZ839/ref=as_li_ss_tl?SubscriptionId=AKIAJO7E5OLQ67NVPFZA&ascsubtag=406899790-2-291805203.1640205636&tag=shopperz_origin1-20

$249 Beelink GK55 Mini PC Intel Processor J4125(up to 2.7GHz) Windows 10 Pro,8G LPDDR4/256G SSD High Performance Business Mini Computer,4K UHD,2.4G/5G Dual WiFi,BT4.0,Dual HDMI Ports with Intel UHD Graphics 600 graphics
This is the same CPU/GPU used in the Synology NAS most people go for on the forum since it has the 600 graphics
Passmark score 3035
https://www.walmart.com/ip/Beelink-GK55-Mini-PC-Intel-Processor-J4125-up-2-7GHz-Windows-10-Pro-8G-LPDDR4-256G-SSD-High-Performance-Business-Computer-4K-UHD-2-4G-5G-Dual-WiFi-BT/627841536?wmlspartner=wlpa&selectedSellerId=101084798

$349 CUK Slim SFF Professional Mini PC (Intel 10th Gen i3, 8GB RAM, 256GB NVMe SSD, Integrated Intel UHD Graphics 630, Windows 10 Home) Business Computer (Made_by_HP_)
Passmark 8828
https://www.amazon.com/dp/B09BGJMTHG/ref=as_li_ss_tl?SubscriptionId=AKIAJO7E5OLQ67NVPFZA&ascsubtag=799043500-2-291805203.1640205636&tag=shopperz_origin1-20

$479 Beelink SEi8 Mini PC 8th Generation Intel i5-8279U Processor,（up to 4.1GHz） Windows 10 Pro Mini Computer with 8G DDR4 RAM/256GB M.2 NVME SSD,Supports 6MB Smart Cache,4K@30Hz Dual HDMI,WiFi,BT5.0, Intel Iris Plus Graphics 655
Passmark 8129
https://www.amazon.com/dp/B08SGCQRYX/ref=as_li_ss_tl?SubscriptionId=AKIAJO7E5OLQ67NVPFZA&ascsubtag=330532379-2-291805203.1640205636&tag=shopperz_origin1-20

$469 Intel NUC 11 Performance mini Desktop Computer, 2.4 GHz Intel Core i5 4-Core (11th Gen), Intel Iris Xe Graphics
https://www.bhphotovideo.com/c/product/1620548-REG/intel_rnuc11pahi50001_nuc_barebones_i5_1135g7_tall.html/specs
Passmark 10184