[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [f-cpu] License issues GPL/LGPL and Juergen Goeritz' SoC

On Thu, 13 Sep 2001, Yann Guidon wrote:

> you fall in the same trap constantly. F-CPU's goal is not to let people
> do "benefits" on it. It is here to "help" the industry as a whole and
> harmonize some common stuff. Where people would use SUN/SPARC workstations,
> they could find F-CPU/Linux boxes for 1/4 the price and a lot of third
> parties.

Why would the F-CPU box be so much cheaper. CPU is quite small part of the
total cost of a computer. Already IA32 machines are dirt cheap, but not as
reliable as workstations. The realiability is what costs with those
workstations. IA64/Hammer etc. will take that place in the future when
64-bits are needed.

> So here, i will draw (or more precisely, redraw and insist upon it)
> a "line" : You have the right to use the F-CPU sources for any purposes
> as long as it is "out of the tarball" (unmodified). In fact, most
> IPs are like that. I remember of an example (unsure but it's a rough
> figure) for the ARM (?) core IP : ie you get the "compiled IP" for $10K
> and if you want to modify it, you have to pay $35K (?). If my memory
> serves me well, i heard these numbers at DATE in Munich).
> "please contact your local reseller for up-to-date pricing".

Quite big part of all IP can be bought as hard of soft macro and the soft
one costs much more. Usually the reason to buy RTL code is to ensure that
if the IP company is bought out or goes out of business there is something
usable for process changes etc. It's quite difficult to migrate
optimized netlist to a new process.

> The next step is : you want to modify the core. You don't need to pay more
> because it is already "free" and it can't be "more free". If i take
> the ARM guy's saying for true, this is not going to happen a lot.
> The condition for modifying the code is to redistribute it (in F-CPUland).
> Of course there is still the problem of where the "limit" of the core is,
> but it's another discussion.

I think the problem is just what you described. What if I add IP 
forwarding accelerator directly to the CPU as a command to make it network
processor. Where is the limit between FCPU and custom logic. Is it the bus
interface to the search engine or is GPL forcing actually release of the
whole search engine. That search engine on the other hand can contain very
sensitive algorithms. What if that search engine is connected to other
logic in the chip, does that have to be released also?

> What we "give" is the possibility to
> "evaluate" the core, within the limits fixed (and slowly loosened)
> by the m4 files' parameters. So if the company wants to implement,
> say, a communication controller (SCSI RAID or IP router for example)
> and wants to avoid the cost of an existing IP, a small team of
> engineers can d/l the F-CPU tarball, the LEON's, anything else,
> and make testbenches, comparisons, benchmarks...
> If it chooses F-CPU, fine. If it needs some changes,
> then it must release the new files. It is simple.

One problem is that the cost of CPU IP is unnoticable in ASIC projects.
The masks can cost $1M currently. Tools that are needed for the flow also
can cost the same amount, and count on top of that the salaries or tens of 
people. $20k for example is neglible sum compared to those. But the real
problem is what must be released after the changes.

> Forcing to release the files is seen, by you and Juergen at least IIRC,
> as a problem. It is also forcing companies to open their eyes : releasing
> their specs and sources increases indirectly the customer's self-help
> and augments the feedback (if the company is open enough). It is a garantee

I think that this just forces companies to use other alternatives. ASICs
are the core components of products and one of the most important
competitive separators. If you release the code you are very vulenarable
to competitors and you loose all the development work quite easily. For
example the main reason for Junipers success in IP router area is their
ASICs, if Cisco could copy them we would have still only Cisco and no

> Companies who heavily invested in ALPHA CPUs (sometimes when somebody lied
> when purposedly saying that ALPHA would be continued during decades) have
> now lost money and time porting their applications. Now, imagine ALPHA
> was open-sourced : the companies would be able to continue the work,
> probably through a joint-venture that spreads the costs.

Few companies have the resources to develop further on CPUs. Old ALPHA
cores will be manufactured for a long time and can be used. If the
development of top of the class CPU would be so easy, companies would not
be dropping out of that business. But it is extremely expensive to develop
top of the line processors.

> I wish the microelectronics world was not such a dirty swamp ...

It just is because the structure is so different from SW world. It just
takes almost month to make masks currently and that costs. And there is no
way to fix small bugs all the time. And also the development cycle is
quite long compared to SW.

Mr. Kim Enkovaara   | kim.enkovaara@iki.fi | Microelectronic Riemannian
Vasamatie 1 C 16    | IRC: embo            | curved-space fault in
02630 Espoo         |                      | write-only file system

To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/