6
\$\begingroup\$

I am new to this site and had one question. I know typical CPUs have power consumption (TDP) in range of 100-200W, for example Intel Core2. I wanted to know what is typical power consumption of FPGAs. I saw this paper, where it says power consumption of Xilinx xc5vlx330 is 30W, but it gives no reference. I wanted some authoritative reference of any FPGA board (preferably high-performance FPGA) and any company. I searched online where they provide datasheets, but they do not clearly tell the power value, but only direct to power estimation tool.

\$\endgroup\$
1
  • 4
    \$\begingroup\$ depends what they are doing... \$\endgroup\$ – vicatcu Feb 26 '13 at 22:10
4
\$\begingroup\$

The reason they direct you to a power estimation tool is because the power consumed depends very heavily on how hard the device is working.

Intel know that their processors will be clocked at the max rated frequency (at least some of the time) and that all the processing units will be used (at least some of the time).

Xilinx have no idea what their customers will do. It may be that the design is slightly too big for a small device, so needs the larger device, hence only (for example) 60% of it is used and the other 40% lies idle, not consuming dynamic power. Most time customers cannot clock the entire device at anything close to the theoretical fmax, so that reduces dyanamic power also.

If you were to calculate a theoretical max power dissipation, it wouldn't help much as you'd then end up overspecifying your power and cooling subsystems by a factor of (guessing) 2x to 4x!

\$\endgroup\$
6
\$\begingroup\$

In general, the amount of power an FPGA "consumes" is related to the function it is performing and the frequency it is operating at. That's why you can't find an authoritative number; because "it depends". I suggest that you play around with the power estimation tools, figure out what the input fields mean, and then you'll have a much better understanding of the factors that contribute to the overall power consumption of the FPGA while it is running a particular design.

\$\endgroup\$
4
\$\begingroup\$

The power estimation tool provided by the FPGA manufactuers is a very good way to obtain an estimate. Obviously, the more information you can provide, the more accurately the results reflect the power consumption you'll see. Otherwise, there are too many parameters that influence the power consumption.

I highly recommend that you try it, even if to see what are the major factors that go into this (Clocks, number of I/Os active, etc).

\$\endgroup\$
-3
\$\begingroup\$

I'm a bit late to this, or maybe it's answered a bit better elsewhere.

But the truth is:

Completely ignore these answers. These people have no idea what they're talking about. Sure, they can spout off some nonsense like "berrraa I'm super professional senior project manager at [X] company"

But the truth is, they have no idea. They tried to search just like you did and got the same answer.

If you're actually interested in FPGAs, the sad truth is that you'll end up needing to use whatever manufacturer's IDE/toolkit to find out. Their particular software toolchain will always give you the best estimate.

The downside of using FPGAs is that it's nearly impossible to create an open source project, mostly because these manufacturers don't allow the use of their libraries in public format. In addition, they contain additional libraries that can "boost performance"

I've done about a dozen projects now, with all revolving around the same core concept. And I can say these companies do this on purpose to really try to sell a product. Nothing more. Nothing less.

While you can definitely get into some open source stuff on FPGAs, you'll quickly find that anything worthwhile would be completely gutted because you'll always have to leave a lot of performance in the box you got the ic from.

The best solution: start with a given family of performance and try to scale up in the same family, but be warned: often times there is not a drop in replacement for the ic you're currently using.

\$\endgroup\$
4
  • 2
    \$\begingroup\$ Welcome to the site. Unlike a traditional discussion forum, we follow a strict question and answer model here, and answers are expected to respond to the question. The point that one needs to use the manufacturer's modeling to get an accurate power figure is of course correct, but contrary to your false accusation against other answers, that's the very same point those existing answers make!. You then spend most of this post ranting about vendor business practices - even when one shares your displeasure, such a rant is not an answer. \$\endgroup\$ – Chris Stratton Oct 12 '20 at 16:48
  • 2
    \$\begingroup\$ So basically, you've ranted about the industry, made false accusations against other users, and in the end your only actual technical response to the question asked is to repeat the essence of answers posted and accepted seven years ago. That's not a positive contribution to this site. \$\endgroup\$ – Chris Stratton Oct 12 '20 at 17:07
  • \$\begingroup\$ A warm welcome to the site. It's an odd set of claims and my own design experience is quite the contrary. In any event, there's no apparent benefit to the OP or future readers in this Answer. It may well be deleted and you have the option to delete it yourself to remove any downvotes from your Reputation. Again, welcome. \$\endgroup\$ – TonyM Oct 13 '20 at 11:12
  • \$\begingroup\$ Jake - your answer does not address the question which in summary is "How can I assess the maximum power usage of a given FPGA?" | Despite your assertions the other answers address the question competently. Essentially FPGA power use depends very heavily on application and applications of a given device vary widely so power use does too. All manufacturers provide tools to assess indicative power use for an application (which can be run without buying anything). || Your comments on other issues are peripheral to the actual question. \$\endgroup\$ – Russell McMahon Oct 13 '20 at 23:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.