|
|
|
| Authored by: Anonymous on Friday, March 29 2013 @ 11:29 AM EDT |
| Establish a precedent and then mount the hunt? [ Reply to This | Parent | # ]
|
| |
| Authored by: DieterWasDriving on Friday, March 29 2013 @ 10:30 PM EDT |
I expect that you are correct.
Target a well-known Linux-using company. Linux, because it dominates the data
center. Hit a company that is not obscure, but without big resources. Certain
not IBM, which will bring in an employee notebook and long-buried bones of the
guy that actually invented "it", and in wrote about why it was a bad
idea.
My reading of the patent is that one claim is a technique for pre-biasing a
number when it's loaded from memory into a processor register so that the
rounding occurs in the right order. For instance when loading a 64 bit memory
value into a 80 bit floating point register, you selectively extend the mantissa
with zeros or ones, depending on the rounding mode in effect, sign of the number
and subsequent mathematical operation. The expectation is that this will result
in a "better" rounding.
I don't know if it's a better rounding or not. Most people with a sophisticated
enough understanding to have a valid opinion would take a long time to that it
was.
But as a practical matter, it's not. 'IEEE' FP is pretty darn good, and every
drawback has been argued over by true experts. If your results don't match
Intel's post-1996 implementation, you are wrong. Just wrong. Broken.
Defective. Return and get your money back. The phrase "the right way, the
wrong way and the army way" has deep meaning. Consistently good enough is
far better than usually slightly better and occasional disastrously wrong. [ Reply to This | Parent | # ]
|
| |
| Authored by: MadTom1999 on Sunday, March 31 2013 @ 08:29 AM EDT |
check your maths, you forgot to multiply it all by the number of people
violating the patent which must be either 0 or the number of people violating
the patent which would not be invalidated by prior art - again 0.
[ Reply to This | Parent | # ]
|
| |
| Authored by: Anonymous on Monday, April 01 2013 @ 08:51 AM EDT |
Does Rackspace even implement the patent? All the systems I'm
familiar with (hardware and software), leave the rounding to
the end of a calculation.[ Reply to This | Parent | # ]
|
| |
| Authored by: tknarr on Monday, April 01 2013 @ 01:57 PM EDT |
I think the usual criteria for a target are:
- Uses a large enough
number of claimed-infringing items that they've got a significant liability if
they're ruled to be infringing. The idea is that they risk a lot by
fighting.
- Has deep-enough pockets that they can afford a settlement.
The idea is that they can settle without breaking the bank and at a lower cost
than fighting.
- Lacks direct involvement with the creation of the
claimed-infringing items. The idea is that they won't have ready access to the
information they'd need to successfully fight, they'll have to spend their own
money digging it up and getting the right witnesses into court and it'll be a
lot of work for them.
It's a rather slimy tactic, but it can be
effective.[ Reply to This | Parent | # ]
|
|
|
|
|
