Category: Uncategorized

Raspberry Pi Cooling Test

Dated: 8/1/2018 Technology/Software used: Raspberry Pi 3 Model B, Raspian Stretch (June 2018)

I’ve owned my Pi for a couple years now. I’ve always wanted to overclock it. Before you overclock you have to have a good cooling solution. I wanted to take this opportunity to test some various cooling solutions and talk about the process I went through. In this post I’ll explore my results from various cooling tests I conducted.

A word of caution: Overclocking your Raspberry Pi may void your warranty, do at your own risk. Due to differences in manufacturing some Pis may overclock well, some will not at all.
The following in no way is meant to be scientifically accurate testing and your results may very.

Why do I want to overclock? Well I sometimes use my Pi to run the RetroPie operating system. I enjoy retro gaming and I would love to be able to emulate N64 better. There’s a lot of mixed opinion and the general consensus is that the Pi is not fit for N64 emulation (spoiler alert: they’re right) but I wanted to try and see for myself if it was possible to play Goldeneye on my Pi.

Ultimately I wanted to use both a small 5 volt fan and heat sink. But I was interested in seeing the thermal performance of each component individually. I tested the heat and processor throttling of a stock configuration using the sysbench application. I was inspired by Christopher Barnett’s thermal tests on his YouTube channel – Explaining Computers and I used a similar bash script to test.

Moding the Case

My case, a very basic one for the Pi, wasn’t designed to hold a fan. I cut an opening and screw holes for the fan with a Dremel tool. You can see in the photos below the before and after. I hadn’t cut acrylic plastic before. I used a regular cutting disk for this. While it did the job, I noticed the plastic was getting a bit gummy which made it hard to control. I would have benefited form scoring the plastic before cutting. If I have to do this again, I’d try a different cutting disk and maybe first score the circle before cutting.

With the prep-work done I began the test. I tried five configurations, the Pi in the closed case (before cutting the fan hole), the Pi in an open case, the fan by itself, the heat sink by itself and finally both the heat sink and fan.

Before:

After:

Why is the cooling so important?

Running electrical components too hot can cause them to fail over time. Getting too hot in one instance can also cause an immediate failure. However, many modern computers have built in thermal protection which will reduce their power in an effort to cool the unit down. The Pi is no exception. Its thermal protection kicks in at about 80 C at this point the Pi will reduce it’s clock-speed to cool itself. I wanted to see how significant the throttling is. So I tracked both the average clock-speed throughout the entire test and the temperature between each test.

This is the script I had used. It’s a bash script which in Linux is like a batch file Windows. In other words a file which automatically runs a series of commands.

#! /bin/bash
echo "Baseline"
vcgencmd measure_temp
vcgencmd measure_clock arm
echo "Test 1"
sysbench --test=cpu --cpu-max-prime=20000 --num-threads=4 run >/dev/null 2>&1
vcgencmd measure_temp
vcgencmd measure_clock arm
echo "Test 2"
sysbench --test=cpu --cpu-max-prime=20000 --num-threads=4 run >/dev/null 2>&1
vcgencmd measure_temp
vcgencmd measure_clock arm
echo "Test 3"
sysbench --test=cpu --cpu-max-prime=20000 --num-threads=4 run >/dev/null 2>&1
vcgencmd measure_temp
vcgencmd measure_clock arm
echo "Test 4"
sysbench --test=cpu --cpu-max-prime=20000 --num-threads=4 run >/dev/null 2>&1
vcgencmd measure_temp
vcgencmd measure_clock arm
echo "Test 5"
sysbench --test=cpu --cpu-max-prime=20000 --num-threads=4 run >/dev/null 2>&1
vcgencmd measure_temp
vcgencmd measure_clock arm
echo "Complete"

The script will run the test five times. We’ll look at the specifics of the each command next, but in essence it displays a base line temperature and clock speed, then runs a stress test followed by displaying the temperature and clock-speed and repeats that sequence 5 times. I recorded the temperature and clock speed after each test into a LibreOffice Calc spreadsheet for analysis.

BASH Script commands

The line that begins with the #! is called a shebang or hash-bang. That identifies to Linux where the interpreter for the script is. In this case Linux will find BASH in my /bin/bash directory.

Next is the echo keyword. This keyword is similar to print() in Python. It will display the text between the quotes on screen. In this script it is used to indicate the beginning of each test. The echo keyword is useful to communicate with the user and let them know what’s going on within the script.

After that we have the commands that actually performs the stress test. Sysbench is a piece of software which is used to perform various benchmark test to determine performance. In this script, rather than gauge performance I am using the CPU benchmark to apply stress to the system which will cause it to generate more heat and I can see how effective my cooling solutions are. The parameters passed to the Sysbench utility are to test the CPU. The CPU test utilized by Sysbench is to determine all prime numbers bewteen 2 and max-prime number in this case 20,000. This is a very intensive task for the computer. I then sepcified that I wanted it to run on all four cores of the processor. Lastly I use the >/dev/null 2>&1 parameter to direct Sysbench’s output to NULL. In otherwords to disregard the results, because in this use case I don’t care what the efficencny of the processor is just it’s heat and clock-speed.

The next commands are two uses of the vcgencmd. Vcgencmd is a utility on the Raspberry Pi to display certain information. In this case I display two things: measure_temp which shows the system temperature and measure_clock arm which displays the current clock-speed of the ARM chip or the main microprocessor of the of the Raspberry Pi. The Raspberry Pi’s microprocessor normally runs at 600Mhz but can turbo up to 1.2Ghz if needed. If the Raspberry Pi gets too hot (around 80C), the system will reduce the speed in an effort to cool it down, this is called throttling. By checking the temperature and clock-speed I can see if the Pi is overheating and throttled slower.

Results

My prediction was that the Fan + Heat-sink would provide the best cooling, however I was surprised that the Fan by itself kept the Pi running near 70 C. I was also surprised that the heat-sinks I used (small aluminum ones) barley cooled the Pi below just having the lid off. We can also confirm from the data that the Fan + Heat-sink provided the best cooling keeping the Pi below 60 C.

From this graph we can see that the Pi only throttled while running without any cooling. The other cooling solutions were able to keep the Pi running at full speed (1200 MHz) throughout each test. From this test, I may have been able to get away with just a heat sink., however because I want to minimize the overall duration of heat exposure on my Pi so I will use both the Fan and the Heat-sink.

Conclusion

Testing the Pi in this way gave me some fascinating insights into different cooling options. In my testing I determined that using a Fan + Heat-sink offers the best cooling option and also does not lead to any throttling conditions. However, I was also able to determine that a Fan by itself offers a similar level of heat control.

Regrets

As I cautioned at the beginning this was not meant to be a scientific testing of different heat control methods. One thing in particular I would have like to have done was a real use case test of the thermal performance. My motivation was to improve the N64 emulation performance in RetroPie. I was able to find a stable overclock which various recommended as potentially useful for what I wanted to do. However, I’m sad to report that in my case it did not improve the performance, which as these same forums mentioned, may not because the nature of the Pi is just very different than an N64. Even though I was not able to improve the performance I’m still glad I was able to go through with this experience.

Want to learn more about Raspberry Pi? Well I haven’t written a book on that (yet) but if you are eager to learn Excel and VBA programming check out my book – Beginning Microsoft Excel VBA Programming for Accountants has many examples of automating Excel to teach you ways to maximize your productivity! It’s available on Amazon, Apple iBooks and other eBook retailers!

Questions or Comments?

Connect with me:

Facebook https://www.facebook.com/DerekHarlanWriter/

Twitter https://twitter.com/tazbirdinvader

Dated: 06/27/18 Software Used: Excel 365, 2016, 2010

Recently, I was asked to assist with a system conversion. I was given the entire general ledger (G/L) from the old system in MS Excel as well as two worksheets for mapping the property numbers and account numbers to the new system. I’ve put a sample screen shot of the G/L so you can see the formating.

At around 26,000 line items, the lead accountant on the project considered using look up formulas. She couldn’t quite picture how to make that work with the format she had. ultimately the solution used will make use of look up formulas, in particular vlookup. But there is a step missing. The reason we can go right to look up formulas is because the general ledger is unnormalized.

Normalize Normalize Normalize

When they brought me in, I knew right away what we needed to do first. To borrow a term from Database Administration the G/L needs to be normalized. That is, it has to be recast it in a way that each row contains all the needed information for the vlookups. This G/L is grouped by Account Number. In other words, the account number doesn’t repeat for each line. To normalize it we have to put the account number on each row.

This was the plan I came up with:

Normalize the G/L

• Insert 2 Columns for the new property and account number
• Use vlookup to look up the new mapping
• Copy the needed info into a journal entry template
• Assign transaction numbers to the J/E
• Upload, post and ensure the T/B matches by period by property (her group handled this)

Code

I could have done the entire solution in VBA. But, time was an issue and this was the most direct approach that occurred to me with one viewing of the file.

This is a simple script to create. But first I’ll insert a column which will hold the account number. Next, I analyzed the structure of the G/L. Note that the lines that begin with a number are the account numbers. While lines that begin with a character are data lines. Using this information let’s build our program.

 

Option Explicit

Sub NormalizeGL()
    Dim startRow As Long
    Dim lastRow As Long
    Dim accountNumber As String
    Dim i As Long
    
    startRow = 7
    lastRow = Range("B1048576").End(xlUp).Row
    
    For i = startRow To lastRow Step 1
        If IsNumeric(Left(Range("B" & i).Value, 1)) Then
            'row is an account number
            accountNumber = Range("B" & i).Value        
        Else
            'row is data line and needs the account number
            Range("A" & i).Value = accountNumber
        
        End If
    
    Next i
        
End Sub

Running that will populate column A with the account numbers for each line. The code is intuitive. First I defined two variables to manage the length of the For loop: startRow and lastRow. The string variable accountNumber’s name is obvious. Last I defined the long variable i as our loop counter.

We assign the lastRow using a useful bit of code. What it does is it goes to the last physical row in that column and then searches upwards for a value. When it finds a cell with a value it assigns that row number to the variable. Many of my solutions use this exact fragment since it’s a very common task.

Next, we enter the For loop where the action happens. We examine the first character of each row. If that character is a number, then we assign the cell’s value to accountNumber. Otherwise we know it is a data row, so we’ll set the cell in column A to the value of accountNumber. The result looks like this:

After that, I sorted the normalized G/L by date and period. Then I removed the subtotals and useless artifacts of the report. Now it was a matter of using some simple vlookups to match the new property/account mapping. I inserted the mapping worksheets into the workbook. The mapping worksheets were simple two column (i.e. old and new) to make the vlookup easier. I’ve put an example of the vlookups below, but they are unremarkable.

=vlookup(B7,Mapping!$A$2:$B$4,2,FALSE)

=vlookup(A7,Mapping!$D$2:$E$3,2,FALSE)

I filled in each row with those formulas. Now each row has all the required information to create the journal entry. Again, for simplicity, I copied and pasted each column into the entry template. The normalized and mapped G/L looked like this:

Transaction Numbers

We’re missing one more piece. The journal entry needs transaction numbers for each group of entries. Normally I’d assigned these based on date. But in Yardi date and period are not related. You can use the same date in several periods. Indeed, this client had done this with audit entries and post period adjustments. We needed to combine the date and period into a unique value, which I call a hash. We can use the hash as a key to lookup a unique transaction number.

First I copied both columns to another worksheet. I used the Advanced Filter to remove the duplicates. If you’re unfamiliar with this:

From the Data menu, Under Sort & Filter click Advanced. It may you about not having header rows but this is unimportant for our task, you can ignore the warning.
Select the Range of dates and periods
Select the destination cells, I picked a column to the right
Check Unique records only
Click OK

You’ll see it created a new list of dates and periods with one line for each combination. In the adjacent column I added a formula to join the two into my hash value

=Concat(D2,E2)
or =CONCATENATE(D2,E2) depending on your version of Excel

then to the column to the right I put my transaction number (i.e. 1,2,3)

I created this hash value in one of the unused columns of the journal entry template. Then used a vlookup to find the transaction number. Since this template needs to upload as a csv file, I pasted the transaction numbers as values and then removed the hash column.

The entry was then uploaded and validated with no errors. This project took me about an hour an a half, but saved them one hundred hours of manual entry. You’ll note that I didn’t create a full VBA solution for this. While it is possible, I felt this was the quickest way with the fewest question marks for accuracy.

Do you want to get started with VBA? Or want to learn what a FOR loop is?

My book, Beginning Microsoft Excel VBA Programming for Accountants has many examples like this to teach you to use Excel to maximize your productivity! It’s available on Amazon, Apple iBooks and other eBook retailers!

Questions or Comments?

Connect with me:

Facebook https://www.facebook.com/DerekHarlanWriter/

Twitter https://twitter.com/tazbirdinvader

nd in your emails so others remember you. “When I go out to a professional organization, people recognize me and come up to me. It may seem gimmicky, but people love it,” she added. These steps can help you not only advance your career, but also possibly gain new cli

Source: 8 ways to be more proactive

This article lists the qualities that set top staff accountants apart.

Source: 10 competencies of top staff accountants

This is the greatest most informed article, assuming today is sometime in early 1993. Sadly terms like artificial intelligence and machine learning have become powerful buzzwords so they’re being used in situations where they are far from appropriate.

One particular flaw is where the author makes a point about difference in file formats, such as PDF docs vs text files. Extracting data from these kind of situations has long since been addressed by non-AI software such as ACL and IDEA.  In fact many Audit functions have been preformed by software for years.

Another example, the author talks about how AI is better at aligning data, this is so vague, an Excel formula can easily show if a PO is related to an Invoice and again  – software in particular Audit software has been doing this for a long time and never under the guise of ML or AI.

Where I do feel it has some value is that he recognizes that

Only human beings, such as the auditor, can tell the true story behind the data.

All in all, I think this is 20 years too late.

Source: Artificial intelligence comes to financial statement audits.

The world suddenly starts to look brighter as the larger hand hovers over the 6. 4:30pm you think only 30 minutes until I can go home for the weekend. That dreaded  popup ends your jubilation as you start to read the email from your boss. Was the Fredricks order competed? You wince in anger and some regret as you start to key your reply. “No, I thought the priority was the Johnson order, I’ll take care of the Fredricks order Monday morning – we’re close, it won’t be problem” Moments later your boss replies with a terse “Very well, however this lateness will be noted in your review.” Wishing you could protest, but you know better,  you know there’s no arguing with an AI.

More and more companies are embracing AI workers, not just in task oriented roles, but also in decision making ones. Some companies have gone as far as to implement monitoring systems to ensure compliance with the AI’s directives. HBR’s article discusses several such examples. And shows us a world where data-scientists are our ambassadors to our computerized coworkers.

While this sounds like Skynet it’s not all doom and gloom or shouldn’t be Schrage notes

CEOs should worry less about bringing autonomy to heel than making it a powerful source and force for competitive advantage.

Learning to trust the algorithm takes “humility and faith” two qualities that leaders should have in abundance.

Will you able to trust the machine?

Source: 4 Models for Using AI to Make Decisions

Some finance executives think their positions could be replaced by technology, a new poll suggests.

Source: Kiss your finance job goodbye, automation is coming.