RAM memory for video surveillance
Below are extracts from a study of how the number of memory strips affects processor performance. I note that synthetic tests do not always adequately reflect what happens on a real application, for example, when running a real application, it was possible to get more than 2 times the acceleration, although in a synthetic test the difference is not so noticeable. In the process of research, the old truth was revealed: under i7 processors, you need to put at least 2 memory levels in dual channel mode, and better than 4, this leads to an improvement in work in some applications 2 times or more, and it was also discovered that memory is more expensive all the same.
So, let's go - when the video viewing system was in operation, the operator started complaining about the screen braking. This happened after adding another camera. A total of 26 video streams with a resolution from 1280 to 1980, almost all have 12fps, one 24fps, plus a resize of the image under the box that it allocated to make it look beautiful. Loading is 100%, frame skipping is visible, it is impossible to work. On the network and the server just no brakes.
Working machine:
For the test, we used the x264 bench test (when you install, specify the program files directory, NOT the program filex (x86) , otherwise the script will not find the way to AviSynth ) and the averaged figure for the luxriot application itself was used for some interesting modes PCMark8
')
RAM test
Slots are numbered letters of the Latin alphabet from the processor. First, we check each bar separately. The whole table is not given because in different slots a similar result.
Now we try to rearrange the pairs of slats in slots. Most of the table is thrown out. similar result. The difference from swapping memory modules between slots (from slot A to C, from C to A) gives no more than 3% of the difference.
Now let's try to score all the bars and switch between the modes of operation. Less is better
As you can see, the processor works in a “sparing” mode and is able to digest as many more cameras. And you can simply build ordinary RAMs, or a pair of hyperx
This study can be applied to servers of other vendors. Luxriot encodes on the client side, others on the server, so these numbers should be suitable for servers from other manufacturers.
Total:
P / S / conducted an additional test with a video card. The results have not changed much! Ie dependence is on the RAM. Although it may be that the situation on a powerful video card would be different. It turns out that it is more profitable to buy a good motherboard with a good processor and memory, and then buy a video card if it is not enough, than to buy a memory worse, but with a mediocre vidyushka.
So, let's go - when the video viewing system was in operation, the operator started complaining about the screen braking. This happened after adding another camera. A total of 26 video streams with a resolution from 1280 to 1980, almost all have 12fps, one 24fps, plus a resize of the image under the box that it allocated to make it look beautiful. Loading is 100%, frame skipping is visible, it is impossible to work. On the network and the server just no brakes.
Working machine:
- Core i7 processor - 4790 (3.6 GHz)
- video accelerator intel hd graphics 4600
- Asus H97M-Plus motherboard
For the test, we used the x264 bench test (when you install, specify the program files directory, NOT the program filex (x86) , otherwise the script will not find the way to AviSynth ) and the averaged figure for the luxriot application itself was used for some interesting modes PCMark8
')
RAM test
| No. p \ p | Name | Volume, GB |
|---|---|---|
| one | AData DDR3 1333 (9) 4GX8U AD3U1333W4G9-B | four |
| 2 | Kingston KVR13N9S8 / 4-SP | four |
| 3 | MCPoint 4/1600 | four |
| four | PQI DDR3-1333U MFACR423PA0103 2Gb | 2 |
| five | kingston hyper-x HX321C11T3K2 / 8 | eight |
| 6 | kingston hyper-x HX321C11T3K2 / 16 | sixteen |
Slots are numbered letters of the Latin alphabet from the processor. First, we check each bar separately. The whole table is not given because in different slots a similar result.
| No | test | Slots | CPU aver load when running luxriot | x264 bench, fps | PCMark8, scores |
|---|---|---|---|---|---|
| one | One PQI plank | slot A | 100% | 55 | 3051 |
| 2 | Kingston One Strap | slot A | 100% | 53 | |
| 3 | One MCP plank | slot A | 98% | 56 | |
| four | One ADATA bar | slot A | 100% | 53 |
Now we try to rearrange the pairs of slats in slots. Most of the table is thrown out. similar result. The difference from swapping memory modules between slots (from slot A to C, from C to A) gives no more than 3% of the difference.
| No | test | Slots | CPU aver load when running luxriot | x264 bench, fps | PCMark8, scores |
|---|---|---|---|---|---|
| five | ADATA and MCP default settings | ADATA slot A, MCP slot B | 99% | 53 | |
| 6 | ADATA and MCP default settings | ADATA slot A, MCP slot C | 60% | 59 | dual channel mode, racing is visible |
Now let's try to score all the bars and switch between the modes of operation. Less is better
| No | test | Slots | CPU aver load when running luxriot | x264 bench, fps | PCMark8, scores |
|---|---|---|---|---|---|
| 7 | 4 Hyper-X | 42% | 60.5 | ||
| eight | 4 Hyper-X, xmp profile 1, (timings 12-12-12-20) | 37% | 65 | 3451 | |
| 9 | 4 Hyper-X, xmp profile 2, (timings 9-9-9-12) | thirty% | 67 | 3455 | |
| ten | Kingston, AData, PQI, MCP | Kigston - slot A, AData - slot B, PQI - slot C, MCP - slot D | 41% | 67 | |
| eleven | 2 Hyper-X, xmp profile 2, (timings 9-9-9-12) | 50% | 61 |
As you can see, the processor works in a “sparing” mode and is able to digest as many more cameras. And you can simply build ordinary RAMs, or a pair of hyperx
This study can be applied to servers of other vendors. Luxriot encodes on the client side, others on the server, so these numbers should be suitable for servers from other manufacturers.
Total:
- the number of operatives plays the main role
- xmp affects performance
- i7 and i5 processors need to be completed with a pair of RAMs in dual channel mode, or 4, but not with one bar
- 2 different RAMs in dual channel work better than one fast
P / S / conducted an additional test with a video card. The results have not changed much! Ie dependence is on the RAM. Although it may be that the situation on a powerful video card would be different. It turns out that it is more profitable to buy a good motherboard with a good processor and memory, and then buy a video card if it is not enough, than to buy a memory worse, but with a mediocre vidyushka.
Source: https://habr.com/ru/post/313984/
All Articles