06-13-2006 08:31 AM
It seems that when I am intensively using the processor (close to 90% utilization) the clock speed on my PXI-8186 slows down. The PXI-8186 has a Pentium 4 M processor. From what I know and from experience (since my laptop has the same processor) this processor will reduce its clock speed when it reaches a specific temperature.
If this is really what is happening, I think this is a major issue. The system that I am developing is a “Hard Real-time” system. Therefore, there could be serious consequences if the clock speed is decreased during execution.
The air temperature where I’m presently testing the system is around 22 C. What would happen if the ambient temperature would be about closer to 40 C?
I do not think that a real-time system that changes clock speed on the fly is deterministic! Has anyone heard about this issue? Does anyone have any recommendations?
Thanks.
Mat
06-15-2006 03:08 PM
Hello Mat,
The Pentium 4 M processor actually has a safety feature that slows down the onboard clock just to prevent the processor from overheating and burning up. This feature was not present on Pentium 3, so processors were literarily burning up. Usually this processor is spec out to run at 100% CPU at up to 50 degrees C. Also, despite all the cooling features running application that will constantly utilize 90% of your CPU, the processor will eventually overheat. Another important factor in the whole problem is what type of chassis you are using with this controller? The cooling system on the chassis should be sufficient enough to prevent early overheating problems.
Also, how much of a slow down do you see in the clock? How are you measuring this slowdown?
This issue has not been reported before but we can definitely investigate what might be the root cause.
06-16-2006 08:13 AM
Hi,
My chassis is a PXI-1042.
“Also, despite all the cooling features running application that will constantly utilize 90% of your CPU, the processor will eventually overheat.”
I am developing a real-time control application and the utilization is fairly constant over time. This is typical of a control application where you are executing the same code in each iteration. From what you are saying it seems that I cannot implement, with this hardware, a control system that has a 90% utilization. Therefore, what is the maximum utilization at which I can be guaranteed that the system will remain within specifications? Since these PXI systems are being used for real-time application, the maximum continuous utilization should be specified. User should be aware of the potential clock slow down if they exceed a specific amount of utilization. Not knowing this could lead to serious consequences for some real-time applications.
We discovered this while benchmarking the system. We realized that after we were running the system for a while the controller was slowing down. After shutting down the system for about 20 minutes the system would run again at its original speed. We will have to better characterize the system to get some specific numbers.
I definitely think that NI should test there controllers to see what is the maximum continuous utilization possible.
Regards,
Mathieu
06-19-2006 11:14 AM
Hi Mathieu,
Please disregard the statement in my previous post in regards of the controller overheating when running at 90% utilization. That was an error on my part and I do apologize for the confusion I have created. All of our controllers are tested under 100% utilizations for number of hours and they have to pass rigorous performance tests. The problem you are seeing is definitely unusual and not expected. The controller should not slow down no matter the utilization level. The PXI-1042 chassis cooling system should prevent the overheating the processor when running at maximum utilization. It is possible that the heat sync on the chassis is malfunctioning which causes the overheating.
You can monitor the temperature of the controller using this application. Also, reference this document which links to software application that might help in troubleshooting where the overheating is coming from. Running these applications can help in root-causing the problem.
Bottom line is that you should be able to build an accurate real-time deterministic system without worries of the performance of the system clock.