AMD Versus Intel

[bp2008]

I've had an AMD Ryzen 7 1800X for quite a while now and have been putting off testing Blue Iris on it. Yesterday my Intel I7-8700K arrived, and this gave me just the excuse I needed to compare their performance.

Predictably, Intel destroys AMD when hardware acceleration is enabled (Blue Iris only supports hardware acceleration with Intel CPUs):

However, with hardware acceleration turned off on the Intel system, they are on basically even footing:

These stats were captured by BiUpdateHelper on two systems running identical Blue Iris configurations. Same cameras, same frame rates, same configuration.

 

[fenderman]

@bp2008 Did you test power consumption?

 

[bp2008]

Damn, I knew I was forgetting something. I did not.

 

[bp2008]

I've made some power consumption measurements using a kill-a-watt meter.

There are some differences between the systems. Both use a 256GB SSD for the OS/boot disk, but the Intel has an additional 2 TB Hitachi hard drive. The AMD system has a GeForce GT 710 (graphics card), since it doesn't have integrated graphics.

I connected both to a 4K display for the "GUI Open" tests, and the live preview rate was unthrottled, making it a very CPU-intensive task.

It is the same 23 cameras as before, but I've increased frame rates somewhat since last time, resulting in 801.1 MP/s of video data.

CPU	Workload	Power Consumption (Watts)	Blue Iris CPU Usage %	Overall CPU Usage %
Intel(R) Core(TM) i7-8700K	IDLE	31	0	0
Intel(R) Core(TM) i7-8700K	Blue Iris (Service)	98	31	32
Intel(R) Core(TM) i7-8700K	Blue Iris (GUI Open)	129	73	75
Intel(R) Core(TM) i7-8700K	Blue Iris (Service) With Hardware Acceleration	79	15	18
Intel(R) Core(TM) i7-8700K	Blue Iris (GUI Open) With Hardware Acceleration	119	49	56
AMD Ryzen 7 1800X	IDLE	40	0	0
AMD Ryzen 7 1800X	Blue Iris (Service)	99	29	30
AMD Ryzen 7 1800X	Blue Iris (GUI Open)	130	49	51

Edit: I had some numbers mixed up in the table above.

 

[bp2008]

I'm pleasantly surprised by how well the Ryzen box handled having the GUI open (this was actually the result of the AMD/Ryzen box having a dedicated GPU while the Intel box did not during those tests). Next, I guess I'll have to transplant a graphics card into the Intel box since that should reduce "GUI Open" CPU usage. I wonder what effect it will have on the power consumption.

 

[tigerwillow1]

Thanks for posting the power numbers. I know it took a good bit of time and effort. Having used an older AMD Phenom system until recently, I'm impressed with how well power consumption has been cut in the newer systems. The old AMD system pulled 100 watts at idle and about 120 watts heavily loaded. I set up two newer Intel-based systems and measured the idle power draw. A Dell tower with an i7-4770 pulls 30 watts at idle, and an HP desktop with an i5-4590, 15 watts at idle. I had a hard time believing it! This is with SSD and internal video. Benchmark-wise, the i7 system is about 6x faster than the old system, and the i5 system about 3x faster.

 

[bp2008]

I re-ran the Intel tests with dedicated graphics installed. I had two cards on hand so I tested each of them. One, the fanless GT 710 from the AMD build. The other, the GTX 950 from my old Blue Iris server that I was using for its HDMI 2.0 output. The GTX 950 is totally overkill, even requiring additional power from the PSU. A GTX 1030 would be a modern compromise, but I don't have one.

GPU	Workload	Power Consumption (Watts)	Blue Iris CPU Usage %	Overall CPU Usage %
GTX 950	IDLE	43	0	0
GTX 950	Blue Iris (Service)	114	33	34
GTX 950	Blue Iris (GUI Open @ 4K)	150	53	54
GTX 950	Blue Iris (Service) With Hardware Acceleration	90	15	20
GTX 950	Blue Iris (GUI Open @ 4K) With Hardware Acceleration	131	33	36
GTX 950	Blue Iris (GUI Open @ 4K plugged in to motherboard) With Hardware Acceleration	129	46	52
GT 710	IDLE	34	0	0
GT 710	Blue Iris (Service)	105	33	35
GT 710	Blue Iris (GUI Open @ 4K)	132	53	54
GT 710	Blue Iris (Service) With Hardware Acceleration	83	15	17
GT 710	Blue Iris (GUI Open @ 4K) With Hardware Acceleration	113	31	34
GT 710	Blue Iris (GUI Open @ 4K plugged in to motherboard) With Hardware Acceleration	123	47	52

As expected, with the GUI closed, there is no benefit to having a dedicated GPU.

With the GUI open, CPU usage is significantly reduced by connecting the monitor to the graphics card's video output. Both graphics cards produced a virtually identical CPU usage drop, despite one of the cards (the GTX 950) being newer and much more powerful than the other.

These results also show that it is important to use a very-low-power graphics card, otherwise the additional power consumption of the GPU will negate your energy savings, as seen in all cases with the GTX 950. In fact, the only test case where adding a graphics card improved power consumption was the test with the GT 710 with the GUI open and hardware acceleration enabled in Blue Iris. Remember, I was using a 4K monitor! If it had been a more-common 1080p monitor, then the graphics card would have been much less beneficial.

 

[bp2008]

I made some charts. Since these are showing resource consumption, shorter bars are better.

These first two charts compare the Intel and AMD computers. Both systems were tested with 80plus Gold power supplies, an SSD for the OS, and a Nvidia GeForce GT 710 graphics card installed.

The Intel machine has a 2 TB HDD that the AMD machine does not have, which increases the reported power consumption of the Intel system by about 5 watts. Essentially, Intel and AMD are on even ground in terms of power consumption when running the same load, but in these charts it will look like Intel does a little worse because of that hard drive.

Note how Intel takes a clear lead in all aspects when Hardware Video Acceleration is enabled.

This last chart is kind of hard to follow, but shows the effect on power consumption and CPU usage when the Intel system had installed:

* No graphics card
* GeForce GT 710
* GeForce GTX 950

Again, shorter bars are better.

Some takeaway points from the GPU tests:

1. Dedicated GPUs consume additional power.
2. Dedicated GPUs only help when the Blue Iris GUI is open.
3. It is possible, in the right circumstances, for a very efficient dedicated GPU to save more energy than it costs to run.
   
4. The very low-end GT 710 has the same effect on Blue Iris as the newer and more power-hungry GTX 950.

 

[awsum140]

Wonder how those charts would look with a 10XX series card? I'm running a 1060 and a 970, no onboard graphics so I'm stuck with cards and no HWA. Power consumption is something I live with to be able to do something other than BI on the 24/7 machine (blasphemy I know), but it would be interesting to see.

 

[pcunite]

@bp2008

I would like to better understand the Quick Sync vs Core comparison.

How would you feel about having forum members donate to raise funds to test these various platforms? We need a Blue Iris expert who can flush this out. I would like to know: power draw, CPU utilization, and a metric of performance (frames per second encoded?). Keep it simple. This way performance and cost could be understood.

Suggested processors:

• i7-800K 6 core with Quick Sync
• Threadripper 1950x 16 core
• Core i9-7980XE 18 core
• Epyc 7401P 24 core
• Epyc 7551P 32 core

Maybe even create a test EXE that can simulate different workloads: 1080p from 30 cams, 10 remote viewers watching cams, etc.

 

[fenderman]

@bp2008

I would like to better understand the Quick Sync vs Core comparison.

How would you feel about having forum members donate to raise funds to test these various platforms? We need a Blue Iris expert who can flush this out. I would like to know: power draw, CPU utilization, and a metric of performance (frames per second encoded?). Keep it simple. This way performance and cost could be understood.

Suggested processors:

• i7-800K 6 core with Quick Sync
• Threadripper 1950x 16 core
• Core i9-7980XE 18 core
• Epyc 7401P 24 core
• Epyc 7551P 32 core

Maybe even create a test EXE that can simulate different workloads: 1080p from 30 cams, 10 remote viewers watching cams, etc.

Lol you are looking at over 6k in gear...if you are stressing an i7-8xxx then you need to move to an Enterprise solution..

 

[nejakejnick]

Have you tried the third option of HW acceleration: " Yes+VideoPostproc"?
I found that it lowers the total CPU consumption even lower (must also count kernel time for this, as it is half of the total usage). Not sure what are the disadvantages, but if I remember correctly that option did not work while also using NVIDIA card.

 

[fenderman]

Have you tried the third option of HW acceleration: " Yes+VideoPostproc"?
I found that it lowers the total CPU consumption even lower (must also count kernel time for this, as it is half of the total usage). Not sure what are the disadvantages, but if I remember correctly that option did not work while also using NVIDIA card.

Vpp is not recommend with high bitrate throughput...just use yes, no vpp

 

[nejakejnick]

I have 4500-5500kB/s, 350MP/s and VideoPostproc saves me total CPU from 30% to 20% on i5-3570k@3,7GHz. So not sure if I qualify for high bitrate throughput.

 

[pcunite]

Lol you are looking at over 6k in gear...if you are stressing an i7-8xxx then you need to move to an Enterprise solution..

I really enjoy building. If an Epyc processor works well, why not? What Enterprise solution are you speaking of?

This system comes to $2,990. Add as many 12TB drives ($490) as you need.

• $1,200 EPYC 7401P 24-Core
• $650 GIGABYTE MZ31-AR0
• $430 32GB RAM
• $150 Power supply
• $150 Samsung 960 EVO M.2
• $270 Case
• $140 Windows 10 PRO x64

 

[fenderman]

I really enjoy building. If an Epyc processor works well, why not? What Enterprise solution are you speaking of?

This system only comes to $2,870, which is a far cry from $6,000

• $1,200 EPYC 7401P 24-Core
• $650 GIGABYTE MZ31-AR0
• $430 32GB RAM
• $150 Power supply
• $150 Samsung 960 EVO M.2
• $150 Case
• $140 Windows 10 PRO x64

the 6k was for the parts listed for comparison...
I can think of much better ways to enjoy the 1k+ extra you spend for building a system...the epyc will be a powerhog and will result in a perpetual tax that you pay to your electric company...makes no sense...
There are lots of great enterprise/commercial vms solutions, like avigilon (just acquired by Motorola the other day) or on the cheaper side network optix (sold by dw spectrum in north america with free lifetime upgrades to new versions)...

 

[bp2008]

I have 4500-5500kB/s, 350MP/s and VideoPostproc saves me total CPU from 30% to 20% on i5-3570k@3,7GHz. So not sure if I qualify for high bitrate throughput.

That is 40 Mbps (give or take) so I'm impressed if VPP is indeed being beneficial. The last time I tried running VPP, I could barely even move the mouse cursor to turn it back off. Granted I had at least twice that bit rate going at the time.

 

[pcunite]

@fenderman

A custom server that we control is very important to us. An extra 200w power draw is within our operating budget. Performance is very important. Given that, I could run Avigilon Standard or Network Optix software? Sounds like a win. I guess you're saying that these two software packages are more CPU efficient over Blue Iris? That way I could build less of a server?

 

[fenderman]

@fenderman

A custom server that we control is very important to us. An extra 200w power draw is within our operating budget. Performance is very important. Given that, I could run Avigilon Standard or Network Optix software? Sounds like a win. I guess you're saying that these two software packages are more CPU efficient over Blue Iris?

you have the same control over a 500 dollar server as you do over a 3k server....a 200w extra draw equates to 200-400 per year depending on your rates....if you are willing to spend that kind of money + 3k to build a system you should not be looking at blue iris...
Avigilon is lighter and has lots more features useful for your commercial needs...

 

[asilva54]

man that cpu usage chart BP provided makes you just think...what if VCE or w/e its called was supported.