Xilinx XPower - Accuracy Information

Apr 04, 2005 6 Replies

Hello



i'm from germany and just registered to this forum. Hello to everybod



out there



I'm working on very small FPGA Designs which are size-comparabl



(about 4-40 Look-up-Tables on a VIRTEX FPGA) to 4 Bit-adders and Bit-multipliers. I use them just for testing some things. The proble is that I use XPower to calculate the power consumption of the design and the power values are not what i expected



I have a testbench, an input stimuli and the placed and routed design



Whenever i put stimulis to the design it delivers (like expected) th "mW" values. But if i change the stimuli to other ones, which ar nearly the same, the power consumption raises to the double or stay the same. (after subtracting the quiescent part). There's (in mos cases) nothing between them



Does anybody know this problem? Have I reached the smallest possibl



XPower value, or what is it



Oh, and does anybody know, how accurate XPower should be. Is there



chance to get some information about it



Thanks for all answers


parit


everybody

designs

design.

Howdy,

Xilinx tries to make XPower as accurate as possible - otherwise why would people bother to use it? Is it possible there are bugs and you've stumbled upon one? Absolutely.

Respectfully, not knowing exactly which FPGA, which version of XPower, which things you are changing between runs (code or otherwise), or which items within the FPGA that are actually being used to implement your different designs (shown in the .mrp file for each run), it is impossible to answer your question with any accuracy.

Think of it like this: it would be the same if you just asked us why your car is _somtimes_ getting poor gas milage. But you don't give any info on the car, your actual milage, your expect milage, where you're driving, how you're driving.

Have fun,

Marc

Hi,

Marc makes some excellent points about the provision of information.

If you're targeting V4 then you are targeting one of our SX or LX devices and you are using 7.1.01i.

If you provide accurate switching information (and an important element of that is simulation information from a VCD or XAD file) then your estimation for V4 devices should be within 50% of the actual measurement. If you run a series of designs you can expect your average design error to be within 25%.

Feel free to send any other queries my way,

Brendan Xilinx Power Tools

parity wrote:

not necessarily. I'm targeting an SX55 and using 6.3sp3.

--

--Ray Andraka, P.E. President, the Andraka Consulting Group, Inc.

401/884-7930 Fax 401/884-7950 email snipped-for-privacy@andraka.com
formatting link

"They that give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety." -Benjamin Franklin, 1759

I agree that you can use ISE to target the SX55 using 6.3sp3.

But XPower is a different kettle of fish to the other tools in ISE. XPower's SX & LX support was only added after silicon-based characterisation was performed. This support for SX & LX was added to XPower in 7.1.01i.

By the way - 7.1.01i allows an FX design to be opened in XPower. That is a bug - and has been corrected in 7.1.02i. 7.1.02i is scheduled for availability in late April.

I hope this clarifies things,

Brendan

Hello

thanks for the answers. I am sorry, that i didn't give you enoug

information. I am using the Xilinx XCV 800bg432-4 and i am also usin VCD and sdf files. I know the FPGA is a really small one, but i nee it for researching purposes only. Yesterday I found some informatio in the Xilinx Answer Database. The accuracy seems to be in the rang of 10 % (PRODUCTION/FINAL)

Here is the link to the Article

formatting link

parit

F77453B674D6E88FE93ECC9C Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit

Hi,

parity wrote:

formatting link

That is correct. XPower's accuracy improves with maturation of silicon.

Brendan

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required