There is a problem with graphing values in the Dude.

If more than 1 sample value is returned between polls, then they will be summed.

For example (with sampling at 30s intervals)
Input samples:
Time(s) CPU(%)
0 35
30 35
60 35
91 35
120 35
150 35
Graph shows:
Time(s) CPU(%)
0 35
30 35
60 35
90 0
120 70
150 35

Example of this type of glitch: Note that the temperature drops as the sample is missed from a 30s interval, then rises as it is added to the next 30s interval.
Verification by snmp polling the device shows that the temperature has not changed by anywhere near the amount indicated in the graph.

To demonstrate this issue can be tiresome , so to make it happen fast , use Reprobe :

Step 1: Create a probe that only ever returns 1. Step 2: Attach it to a device, and wait for some history to accumulate. Step 3: Hit reprobe lots. I hit reprobe 12 times, so graph shows: 12 reprobes + 1 real sample = 13

That is a great post on this problem. I hope they can find a solution. I find pings all continue to be wrong all over the place when darn near everything on my network is under 1ms and most show anywhere from 1ms to 500ms and above. It is obvious on temperature probes and many other probes like disk and such. I have found that forcing affinity to one processor does help. Also I have found that restarting the service clears things up for a while. This is an old issue and has been around since early beta 3 versions. I don't know if it was around for version 2. I started using the dude at v3 beta.
On a slightly related note I have also found that when a new device gets added false positives start appearing soon after, on windows 2k3.