BI running on a new i9-7980XE (18 Cores, 36 Threads) = NICE!!!

[bp2008]

Interesting numbers there. Thanks for contributing. 3434.2 MP/s is huge, but admittedly not 100% accurate since BiUpdateHelper doesn't take into account "Limit Decode" when calculating MP/s. I'm not sure if Blue Iris's internal MP/s calculation (visible in the Status window) would be more accurate or not.

Other things that can make the numbers inaccurate or misleading are duplicated cameras or other camera types that have totally different CPU loading characteristics, like analog capture, USB, or desktop capture.

 

[Tuckerdude]

Hey bp2008, I was hoping you would chime in here!

First, let me say how much I appreciate the tools and UI3 stuff that you have developed....super useful!

Second (and for clarification), I've not used limit decoding on the new rig at first just so I could see where I landed. I've gone ahead and checked that feature on a handful of cameras now, and now running in the neighborhood of 43%. But, I also wanted to be sure that I could safely utilize the web interface (and your super cool IU3) without worrying about running my machine to the absolute limit! So I feel pretty good about where I'm at on this new setup.

 

[bp2008]

Remember that Limit Decoding doesn't have any effect on cameras that are currently visible in a mobile app or web interface. So while that feature should be able to save a few watts of energy, it won't make your system be able to handle more cameras.

 

[bp2008]

I also note you had hardware acceleration enabled, though the CPU doesn't support it. I'm curious whether turning that off makes any difference for CPU usage at all.

 

[Tuckerdude]

Yep....understood! I have done tests on both old and new setups and measured the differences.

 

[Tuckerdude]

I also note you had hardware acceleration enabled, though the CPU doesn't support it. I'm curious whether turning that off makes any difference for CPU usage at all.

Yeah, I did try with and without....no visible difference that I can spot! I knew this going into it, but I guess it brings up more questions than it answers? Meaning...I wonder what this says about the new architecture in terms of overall computing power?

 

[Migel]

... I guess it brings up more questions than it answers? Meaning...I wonder what this says about the new architecture in terms of overall computing power?

computing power for video intensive tasks in particular. I've been curious to know how QS equates to raw cores. Both in terms of power and raw throughput ability.

 

[bp2008]

computing power for video intensive tasks in particular. I've been curious to know how QS equates to raw cores. Both in terms of power and raw throughput ability.

It doesn't really equate to cores, but it is extremely helpful.

My i7-3770K pulls 656 MP/s right now. It is a total of 23 cameras, 21 of which are using hardware acceleration (I disabled it on my 2 PTZs because it slightly reduced latency to use the software decoder).

With hardware acceleration enabled for 21 cameras, Blue Iris CPU usage is 27%, overall system CPU usage is 32%.

With hardware acceleration disabled for all 23 cameras, Blue Iris CPU usage is 77%, overall system CPU usage is 83%.

 

[bp2008]

Yeah, I did try with and without....no visible difference that I can spot! I knew this going into it, but I guess it brings up more questions than it answers? Meaning...I wonder what this says about the new architecture in terms of overall computing power?

That doesn't really say anything about the new architecture, only that hardware acceleration is unsupported with your CPU model, which we already knew. I was just curious if enabling the feature in Blue Iris on unsupported hardware caused it to use an Intel-provided software decoder or something, which it apparently does not.

 

[Migel]

It doesn't really equate to cores, but it is extremely helpful.

My point was, in your example, how many 'cores/speed' would it take to 21 cameras without HA and result in 32%. I know it is not 100% actionable knowledge but interesting. Something like "a x core y Mhz machine with HA will outperform or equal a a core / b Mhz machine without HA"

 

[bp2008]

My point was, in your example, how many 'cores/speed' would it take to 21 cameras without HA and result in 32%. I know it is not 100% actionable knowledge but interesting. Something like "a x core y Mhz machine with HA will outperform or equal a a core / b Mhz machine without HA"

Well it reduced my CPU load by a factor of about 2.6 to 2.85. This is on a 4-core CPU. So if there was such a thing as an 11 core (22 thread) 3rd-generation i7 running at the same clock speed as an i7-3770K (3.5 GHz, 3.9 GHz turbo), then it might run my exact Blue Iris load without HA at about the same CPU usage as my actual server does with HA. But that really is oversimplifying things. I took those measurements with the Blue Iris GUI closed and no web server connections, which means the majority of what Blue Iris was doing with its CPU time was decoding video, and that is how I was able to see a difference of almost 3x just from switching HA on and off. If I had the GUI open on my 4K monitor for example, that adds about 20% CPU usage all by itself. It would have been ~52% overall CPU usage with HA enabled and ~103% with HA disabled, resulting in a CPU load difference that is only about 2x. And at that point I would calculate that an 8 core CPU without HA would do as well as a 4 core with HA. Do a few more things that Quick Sync doesn't help with, like serving web/mobile clients, and the gap closes even further. This is why I said quick sync doesn't equate to cores.

It used to be really common for people to not even have Direct-to-disc enabled, so they would enable hardware acceleration and see CPU usage drop by really small amounts, like 80% -> 60%, because video decoding was only a small part of the pie for them.

 

[Tuckerdude]

Hey everybody....another quick update that I thought would be interesting to share:

As mentioned, on my old Core i7 setup I had enabled "limit decoding" on about 20 of my 33 cameras just so that I could run at around 50% CPU. I thought it would be interesting to exactly replicate these settings on the Core i9 and see the difference. After enable the limit feature on all the same cameras (about 20 of them), the Core i9 is running at about 25% and very steadily holding at that. So half as much as the older i7. Now I know that there are reasons why this feature is not the most popular and get the trade-offs of using it. But again, I was just curious to see the difference between the old vs the new and it's pretty significant.

Hope this is helpful/interesting for some to know!

 

[bp2008]

half as much as the older i7.

Ironically you can thank AMD for that. I'm sure Intel would have been happy to coast along with another 10 core CPU for $2000 if it weren't for Ryzen.

 

[Tuckerdude]

Ha ha....so true man! They have been keeping them on their toes that's for sure. Out of curiosity...are you running AMD?

 

[bp2008]

Ha ha....so true man! They have been keeping them on their toes that's for sure. Out of curiosity...are you running AMD?

Not for Blue Iris. I go the more efficient route (Intel w/Quick Sync).

 

[Laser200]

Hey Tuckerdude,
I didn’t see how much RAM you are running. I would think that would make a difference in performance.
James

 

[Tuckerdude]

Hiya James,

The machine has 128GB of DDR4 Ram. In what way do you think that's a factor...just curious to hear your thoughts.

 

[bp2008]

The full details are here: Blue Iris Update Helper

It is the i9-7980XE system. (screenshot of details below)



It was obviously not built to be a cost-effective Blue Iris server. Blue Iris was only using about 13 GB, which is the highest I've ever seen Blue Iris go, by a large margin. It would have performed the same with 32 GB. Maybe even with 16 GB.

 

[bp2008]

In fact, I'm curious @Tuckerdude, what purpose did you have in mind for this machine? The CPU isn't optimal for gaming, and the graphics cards, though they were high-end 3.5 years ago they certainly aren't efficient anymore. 128 GB of RAM suggests an intent to do a lot of virtualization, but serious virtualization (of servers) is usually done with actual server hardware that supports ECC memory and IPMI and stuff. This really strikes me as more of an extremely high-end workstation with hand-me-down GPUs.

 

[Tuckerdude]

Ha ha....You are spot on right bp! I didn't have "cost-effectiveness" in mind for this machine. But the truth is, this i9 build is going to replace my existing work-horse machine that I use everyday, and the extra memory gets used by lots of other applications that I run on a day to day basis. However, it has served a dual purpose in giving me some insights into how well it can handle my Blue Iris setup. So, now that I know...I will be building another one with pretty much the same specs to replace my existing BI dedicated machine.

The videocard is about to be swapped out with my Titan X that's in my current machine. I just didn't put it into the new machine yet as I needed both of them up and running to do my initial tests (and to migrate files)

Given what you've seen (the results), anything you think I should do differently? The cost is not the main concern, it's all about capacity and performance specifically for BI.

Thanks for any insights you might have