10-28-2018 02:42 AM
I designed circuit in Vivado 2016.2 tools ,after synthesis and implementation. I generated power report, report contains total power consumption. What is unit of power in tool(Watt or milli watt), but tool used watt(W) for measuring the power. If unit is watt(W), the result is not matching with other power measurement tools(high error rate). If unit is millin watt(mW), the error is very less compared to other power measurement tools.
10-28-2018 09:33 AM
McPAT. How are you telling it about the characteristics of the fpga ?
so I'd say it has much less chance of providing a good estimate than the xilixn tools that do know the fpga,
10-28-2018 03:04 AM - edited 10-28-2018 03:55 AM
Hi @khyam,
If you are using Tcl Interface and if you don't specify the unit. Then default unit of power in mW (milliwatts).
if you are using Vivado IDE, tool will take unit of power as Watt.
To get accurate Power estimated results, refer Confidence level section in the output report file and please provide user input data who's confidence level is low or medium.
Regards,
Deepak D N
------------------------------------------------------------------------------------------------------------------------------------------------------
Please Reply or Give Kudos or Mark it as an Accepted Solution.
------------------------------------------------------------------------------------------------------------------------------------------------------
10-28-2018 04:37 AM
How to provide user input data, when confidence level is low or medium. I am using Vivado 2016.2, the power unit(W).
10-28-2018 05:40 AM
Hi @khyam,
Refer Chapter 3 of UG907 for providing User Input data.
For example : Referring Confidence section in output File. If you find Clock Nodes Activity, confidence level as Low, then one of the factors affecting Clock Node activity is SAIF File. Please provide SAIF File as input to Vivado IDE, refer Page 39 of UG907.
Regards,
Deepak D N
--------------------------------------------------------------------------------------------------------------------------------------------------------------
Please reply or Give Kudos or Mark it as an Accepted Solution.
--------------------------------------------------------------------------------------------------------------------------------------------------------------
10-28-2018 05:46 AM
as @ddn says , the units
On a second level,
the 'trust' in the power estimation , is dependent upon the accuracy of the numbers entered !
A lot of the power consumption of the FPGA is 'dynamic' , i.e. the more a node toggles , the more power it takes,
which is very hard for us as designers to estimate,
so take the estimate with a BIG pinch of salt,
10-28-2018 06:11 AM
It is corrent, my design is taking 10W of power in xilinx Vivado, but same design in other power estimator, it is 20mW. The error rate is very high. is it valid power value obtained from xilinx vivado.
10-28-2018 06:38 AM - edited 10-28-2018 06:42 AM
Hi @khyam
It is always advisable to do vector based power estimation with the help of saif file. This saif file will record port and signal switching rates which you can include in your Vivado power estimator tool. Refer the below link to generate saif file for power analysis in Vivado simulator:
https://www.xilinx.com/support/answers/52632.html
Further add the saif file in the report power GUI as shown below:
FYI correlation btw eport_power (in SAIF mode) and actual hardware measurements is +/-10%.
Very much valid point by @drjohnsmith
"the 'trust' in the power estimation is dependent upon the accuracy of the numbers entered !"".
Vector based power estimation is more accurate compare to vector less power estimation. Refer UG907 for more info.
Although in your case difference the power measured with Vivado power estimator and other estimators is way to high. I suggest you to use saif mode of power estimation as mentioned above.
10-28-2018 06:59 AM
Ok
thank you very much for valuable information @thakurr, @drjohnsmith and @ddn
10-28-2018 07:00 AM
Whats the other power estimator your using ?
mW seems a low number of power
10-28-2018 08:03 AM
Software based power estimator tools McPAT
10-28-2018 09:33 AM
McPAT. How are you telling it about the characteristics of the fpga ?
so I'd say it has much less chance of providing a good estimate than the xilixn tools that do know the fpga,