Print
PDFOverview
You can use benchmarking data to decide which components you need to meet your system requirements. For more complex systems, you can use benchmarking data to determine the type of system architecture you want to implement based on the current and future requirements of your application. This document contains benchmarks of some typical dynamic signal acquisition applications.
Table of Contents
Streaming Data to Disk
This test evaluates how quickly certain controllers can sustain saving data from a live acquisition to disk.
| PXI-8105 | PXI-8351 | |||||
|
DSA Device
|
PXI-4462 | PXI-4472 | PXI-4462 | PXI-4472 | PXI-4462 | PXI-4498 |
|
Channels
|
Max fs
|
Max fs
|
Max fs
|
Max fs
|
Max fs
|
Max fs
|
|
20
|
204.8k
|
102.4k
|
204.8k
|
102.4k
|
204.8k
|
204.8k
|
|
24
|
200k
|
102.4k
|
204.8k
|
102.4k
|
204.8k
|
204.8k
|
|
28
|
190k
|
102.4k
|
204.8k
|
102.4k
|
204.8k
|
204.8k
|
|
32
|
170k
|
102.4k
|
204.8k
|
102.4k
|
204.8k
|
204.8k
|
|
36
|
160k
|
102.4k
|
204.8k
|
102.4k
|
204.8k
|
204.8k
|
|
40
|
130k
|
102.4k
|
192k
|
102.4k
|
204.8k
|
204.8k
|
|
44
|
120k
|
102.4k
|
176k
|
102.4k
|
204.8k
|
204.8k
|
|
48
|
102k
|
102.4k
|
140k
|
102.4k
|
204.8k
|
204.8k
|
|
52
|
-
|
102.4k
|
126k
|
102.4k
|
204.8k
|
204.8k
|
|
56
|
-
|
102.4k
|
-
|
102.4k
|
204.8k |
204.8k
|
|
60
|
-
|
102.4k
|
115k
|
102.4k
|
204.8k |
204.8k
|
|
64
|
-
|
102.4k
|
-
|
102.4k
|
204.8k |
204.8k
|
|
68
|
-
|
102.4k
|
-
|
102.4k
|
204.8k |
204.8k
|
|
72
|
-
|
102.4k
|
-
|
102.4k
|
- |
204.8k
|
|
80
|
-
|
96k
|
-
|
102.4k
|
- |
204.8k
|
|
88
|
-
|
90k
|
-
|
102.4k
|
- |
204.8k
|
|
96
|
-
|
78k
|
-
|
102.4k
|
- |
204.8k
|
|
104
|
-
|
70k
|
-
|
102.4k
|
- |
204.8k
|
|
112
|
-
|
68k
|
-
|
102.4k
|
- |
204.8k
|
| 168 | - | - | - | - | - | 204.8k |
| 192 | - | - | - | - | - | 170k |
| 224 | - | - | - | - | - | 130k |
| 272 | - | - | - | - | - | 90k |
Notes:
1 - Used lossless hardware compression feature of PXI-4498. Without this feature enabled, performance decreases by about 25% due to increased PCI bus traffic.
Test Details:
- Software: LabVIEW 8.5, NI-DAQmx 8.6
- Data logged as I32s
- Each test ran for a minimum of 2 minutes; tests at 102.4 kS/s or 204.8kS/s ran for 5 minutes
- PXI-8351 was configured with 2x160GB RAID-0 and MXI-Express
- PXI-8353 was configured with 1GB RAM, 500GB RAID-0 and MXI-Express
Analyzing Data
All analyses were performed with VIs from the Sound & Vibration Toolkit (SVT).
SVT Power Spectrum
This test evaluates the speed of different controllers when performing a power spectrum analysis on data from a continuous acquisition.
 |
PXI-8105 | PXI-8351 | |||
|
DSA Device
|
PXI-4462 |
PXI-4472
|
PXI-4472 |
PXI-4462
|
PXI-4498
|
|
Channels
|
Max fs
|
Max fs
|
Max fs
|
Max fs
|
Max fs
|
|
12
|
204.8k
|
102.4k
|
102.4k
|
204.8k
|
204.8k
|
|
14
|
-
|
102.4k
|
102.4k
|
204.8k
|
204.8k
|
|
16
|
198k
|
102.4k
|
102.4k
|
204.8k
|
204.8k
|
|
20
|
159k
|
102.4k
|
102.4k
|
204.8k
|
204.8k
|
|
24
|
132k
|
102.4k
|
102.4k
|
185k
|
204.8k
|
|
28
|
114k
|
102.4k
|
102.4k
|
-
|
204.8k
|
|
32
|
100k
|
102.4k
|
102.4k
|
140k
|
204.8k
|
|
36
|
89k
|
-
|
102.4k
|
-
|
204.8k
|
|
40
|
80k
|
87k
|
102.4k
|
110k
|
204.8k
|
|
44
|
73k
|
-
|
-
|
- |
204.8k
|
|
48
|
65k
|
73k
|
87k
|
90k |
204.8k
|
|
56
|
-
|
62k
|
75k
|
80k |
-
|
|
64
|
-
|
55k
|
65k
|
70k |
165k
|
|
68
|
-
|
-
|
-
|
65k |
-
|
|
72
|
-
|
48k
|
-
|
- |
-
|
|
80
|
-
|
44k
|
52k
|
- |
-
|
|
88
|
-
|
40k
|
-
|
- |
-
|
|
96
|
-
|
37k
|
37k
|
- |
110k
|
|
104
|
-
|
34k
|
-
|
- |
-
|
|
112
|
-
|
-
|
30k
|
- |
-
|
| 128 | - | - | - | - | 80k |
| 160 | - | - | - | - | 65k |
| 192 | - | - | - | - | 50k |
| 224 | - | - | - | - | 45k |
| 272 | - | - | - | - | 35k |
Notes:
1 - Used multicore technique to share processing across the 4 cores of the PXI-8353. See this example: Multicore Analysis with DAQmx Acquisition
Test Details:
- Software: LabVIEW 8.5, NI-DAQmx 8.6, Sound & Vibration Toolkit 5.0
- Block size = 4096 - 16384
- No display or data logging occurred
- Each test ran for a minimum of 1 minute
SVT Power Spectrum and Log
This test evaluates both the controller speed performing a power spectrum as well as logging the time-domain data to disk.
|
PXI-8105 | PXI-8351 | ||
|
DSA Device
|
PXI-4462
|
PXI-4472
|
PXI-4472
|
PXI-4462
|
|
Channels
|
Max fs
|
Max fs
|
Max fs
|
Max fs
|
|
8
|
204.8k
|
102.4k
|
102.4k
|
204.8k
|
|
12
|
177k
|
102.4k
|
102.4k
|
204.8k
|
|
16
|
145k
|
102.4k
|
102.4k
|
204.8k
|
|
24
|
92k
|
102.4k
|
102.4k
|
142k
|
|
32
|
73k
|
-
|
102.4k
|
110k
|
|
40
|
56k
|
79k
|
82k
|
85k
|
|
48
|
49k
|
-
|
70k
|
70k
|
|
56
|
-
|
57k
|
58k
|
58k
|
|
64
|
-
|
-
|
48k
|
50k
|
|
80
|
-
|
40k
|
40k
|
-
|
|
96
|
-
|
-
|
35k
|
-
|
|
104
|
-
|
31k
|
|
-
|
|
112
|
-
|
-
|
30k
|
-
|
Test Details:
- Software: LabVIEW 8.0, NI-DAQmx 8.0, Sound & Vibration Toolkit 4.0
- Block size = 2048 - 16384
- Data saved to disk as scaled doubles data type
- No data display occurred
- Each test ran for a minimum of 1 minute
Third Octave Analysis
This test evaluates how many channels of third-octave analysis a controller can perform over a certain number of bands.
|
||||
|
DSA Device
|
PXI-4472
|
PXI-4472
|
PXI-4462
|
PXI-4498
|
|
Octave Bands
|
Maximum #
of channels |
Maximum #
of channels |
Maximum #
of channels |
Maximum #
of channels |
|
20Hz - 6.3kHz
|
56
|
108
|
68
|
-
|
|
20Hz - 8kHz
|
42
|
76
|
68
|
184
|
|
20Hz - 10kHz
|
35
|
60
|
60
|
152
|
|
20Hz - 12.5kHz
|
28
|
56
|
56
|
120
|
|
20Hz - 16kHz
|
21
|
44
|
44
|
96
|
|
20Hz - 20kHz
|
16
|
32
|
32
|
72
|
Notes:
1 - Used multicore technique to share processing across the 4 cores of the PXI-8353. See this example: Multicore Analysis with DAQmx Acquisition
Test Details:
- Software: LabVIEW 8.5, NI-DAQmx 8.6, Sound & Vibration Toolkit 5.0
- Sampling rate set to 2.56 times the highest octave band
- No display or data logging occurred
- Each test ran for a minimum of 1 minute
- PXI-8351 tests used code to take advantage of dual core processors
Evaluating the Effect of Block Size on Fast Fourier Transforms and Octave Analysis
The following section discusses how the block size of a Fast Fourier Transform (FFT) or octave measurement affects the performance of the system. The following performance tests were performed on older PCs, but the general trend of block size versus processing time is consistent for all PCs.
FFT Analysis Comparison
Note in the above graph that the performance for a block size from 1024 to 16,384 is relatively the same. Traditionally, a 1024 block size, or 400-line FFT, is standard in most instrumentation. However, increasing the block size does not affect performance in terms of data throughput and might actually improve it in the case of changing to a 2048 block size. With a larger block size, the frequency resolution of the FFT result is twice as high. You can achieve double the frequency resolution with no negative impact on performance.
The display does not update with the same speed, however. For example, examine the 2048 and 4096 block sizes of the top line of the graph. Whether you use a 2048 block size or a 4096 block size, the number of data points processed per second remains the same. However, a 2048 block size delivers an 800-line FFT twice as fast as a 4096 block size delivers a 1600-line FFT. Therefore, increasing the block size improves frequency resolution without decreasing performance, but the number of updates to the display decreases.
Note that the top two lines use RMS averaging whereas the bottom two lines are not. When you select RMS averaging, only the magnitude information is calculated. The phase information is not calculated. When you do not select a specific type of averaging, the complex FFT is computed where both magnitude and phase are calculated, which takes more time due to the increase in computation in the algorithm.
The graph below shows the performance of octave analysis using different averaging methods over different block sizes. The x-axis shows the number of block sizes used and the y-axis shows the number of data points processed per second.
Octave Analysis Comparison
Note on the graph above that the performance is relatively consistent with a block size from 2000 to 32,000 samples. Increasing the block size in this case shows no decrease in performance. Also note that the order of increasing computationally intensive averaging methods are linear, equal confidence, and exponential. While in some cases increasing the block size does not affect system performance, it does affect the rate at which results are displayed and updated, even though the actual number of data points processed per second remains the same.
Reader Comments | Submit a comment »
Difficult to read the labels
It is difficult to read the labels on the
graphs. They are too small.
- Aug 1, 2007
Legal
This tutorial (this "tutorial") was developed by National Instruments ("NI"). Although technical support of this tutorial may be made available by National Instruments, the content in this tutorial may not be completely tested and verified, and NI does not guarantee its quality in any way or that NI will continue to support this content with each new revision of related products and drivers. THIS TUTORIAL IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY KIND AND SUBJECT TO CERTAIN RESTRICTIONS AS MORE SPECIFICALLY SET FORTH IN NI.COM'S TERMS OF USE (http://ni.com/legal/termsofuse/unitedstates/us/).