Wednesday, June 22, 2011 4:45 PMOwner
I found out that you implemented EMIPs on the Virtex5 board. I am also working on that as well. I would like to know how did you measure the power consumption on this board. I checked on one of the paper from this research and they did that on the Virtex4 however the schematic seems to be different to that of Virtex5. Where is the resistor located at so that I can meausure power consumption of the FPGA chip. Any help is greatly appreciate it.
Due to the expense of these boards we have not attempted this type of experiment on the V5 at this time. We did on the V4 because we had several of these boards available from previous work so that if we damaged it beyond beyond use we would not be out a lot. We replaced the resister on the 1V rail that powered the FPGA logic. You would have to check the V5 board schematic you are using to find out where that is. Then remove the original resistor and replace it with one that will allow you to measure the voltage drop with what instruments you have.
There might be another way to get similar measurements from the V5. I have not look at it much yet but the System Monitor on the V5 might provide you with so information without modifying the board. I would investigate that avenue if there is any hesitation at changing this resistor.
rduar, you are correct. The ML50x/XUPV5 boards are simply not set up to provide detailed power measurements since they do not have shunt resistors on the power rails. I'm not certain why, since many of the other reference boards have them (ML550 & ML555 for example).
Anyways, the best I have done is to get very gross estimates. Assuming that you are not really using anything on the board beside the FPGA itself, you can get current information for the whole board directly from the 5V source. Take into account the 82% efficiency of the PTH08T220W @ 1.0V (http://focus.ti.com/lit/ds/symlink/pth08t220w.pdf). If you are using peripherals, things get more complicated since the efficiencies of the various regulators are different. In either case, if you are only trying to get relative numbers and you are OK with averaging over a relatively long time scale, maybe this will work for you too.
If this isn't sufficient and you are far braver than myself, you could theoretically insert your own shunt resistor. I don't want to take responsibility if you ruin your board, but take a look at page 25 of the "ml50x_schematics.pdf" file on the Xilinx website. In principle, you wouldn't have to disturb the traces on the board if you:
1) replaced the plug between Pin 5 of the PTH08T220W and the board with a shunt resistor
2) removed the plugs connecting Pins 6 and 9 of the PTH08T220W to the board
3) reconnected Pin 6 to Pin 9 of the PTH08T220W externally with a resistor to replace R194
4) reconnect Pin 5 to Pin 6 of the PTH08T220W externally
Again, unless you have a certain amount of confidence, this seems pretty risky. In comparison, rough estimation from the board level and XPower Estimator don't seem like such a bad option!