Can we reduce EDA license cost by using larger, faster servers?

[nasica88]

Sorry for my novice question. I am a mere systems engineer for EDA servers.

I am trying to figure out how to reduce EDA software license cost like Cadence or Synopsys. In an effort for that, are the following options viable? If no, can you pls tell me why?

1) Just use the latest, faster processors from Intel or AMD or even alternative processors like ARM or IBM Power.
2) Just use larger servers with more CPU cores or more sockets - we have up to 16-socket servers in the market.
3) Use MPI for parallel processing across multiple server boxes.

Again, thanks for your advices.

** In a follow-up question, how many CPU cores do you usually use for a single job? I know this question is too vague, but let me hear your usual practice.

 

[crutschow]

The software license cost is for the supplier to make money from designing and supporting the software, and he charges what the traffic will bear.

The cost has nothing to do with processor or server speed (which just determines how fast the software will run).
Why do you think it would.

 

[barry]

Sorry for my novice question. I am a mere systems engineer for EDA servers.

I am trying to figure out how to reduce EDA software license cost like Cadence or Synopsys. In an effort for that, are the following options viable? If no, can you pls tell me why?

1) Just use the latest, faster processors from Intel or AMD or even alternative processors like ARM or IBM Power.
2) Just use larger servers with more CPU cores or more sockets - we have up to 16-socket servers in the market.
3) Use MPI for parallel processing across multiple server boxes.

Again, thanks for your advices.

** In a follow-up question, how many CPU cores do you usually use for a single job? I know this question is too vague, but let me hear your usual practice.

Novice or not, this question makes no sense. This is like asking if faster cars should cost less.

 

[D.A.(Tony)Stewart]

Comparing Software License cost to Hardware Power is like comparing Apples and Oranges. It’s a wrong assumption that they are opportunities for cost reduction.

You can have a big server. The fees are based on the number of licensed users or “seats”.

 

[dick_freebird]

The only way a faster server will help you is if the engineers
using it, have the discipline to give up license once their task
is complete. (That rarely is the case)

Tailoring activity-based license checkin could help, but expect
plenty of bitching from everybody "who was just about to..."
when their license went away and now they're at the back of
the queue.

You are just a field hand on somebody else's plantation and
the rules are not set up to benefit you at all.

Now one point of leverage might be, find the license-hogging
activities and substitute tools that don't need license tokens.
Such as, get ngspice to run your Spectre netlists clean, and
you can get as much simulation horsepower as you can find
CPU cores. Find other ways than Calculator to get your info
out of the results (like see if you can force a SPICE format
output "rawfile" which would let you use various results
browsers / viewers) or use output methods which produce a
non-proprietary format, maybe "just the numbers" for the
cared-about quantities.

You might even be able to work from open schematic tools to
SPICE netlists on that side, and verify against Brand X layout
extracted netlist, in foundry-blessed Brand X LVS, eliminating
seat usage for that task-set. Presuming your expensive tools
can read the industry standard format and do their job.

 

[ThisIsNotSam]

Classical cost model for EDA is seat based. Slow or fast hardware would make no or little difference. Perhaps, and this is really unlikely, if you could reduce peak load maybe you could buy less licenses and still keep your engineers working happily. I would not count on it.

There are rumours of software-as-a-service offerings from big name vendors that have a cost model based on use time. I don't know the details. These would be cloud based and would run on Amazon/Google machines. Even if, in theory, you could save money by running things fast, you don't have much control over the hardware offered by the cloud.