( ESNUG 426 Item 10 ) ------------------------------------------- [03/31/04]
Subject: Nine More Letters On The Free Xilinx XST & Altera Quartus Tools
> First off, overall FPGA synthesis market *shrank* 15% in 2002 ??!! Huh?
>
> No only that, FPGAs have been getting bigger every year -- thus driving
> a stronger demand for FPGA synthesis tools. Maybe you can do a 10,000
> gate design using schematic capture, but God help you if you're not using
> simulation and synthesis for a 100,000 gate design.
>
> - from http://www.deepchip.com/gadfly/gad031204.html
From: [ An Anon Engineer ]
John,
Keep me anon...
Perhaps another possible variable may be that each vendor has changed how
they report their FPGA seats vs ASIC seats??? I don't follow the market
at all, but as you know more than anyone, they are millions of different
ways to slice up data. Synplicity has been trying to penetrate the ASIC
market, so perhaps they "pushed" more of their 2002 seat sales into the
ASIC bucket??? I can imagine they may have several customers who use their
tools for ASIC and FPGA - maybe Synplicity counted those seats as ASIC
seats??? Maybe Mentor/Synopsys have done similar, but only pushing more
into the FPGA buckets???
Just speculating...
- [ An Anon Engineer ]
---- ---- ---- ---- ---- ---- ----
From: Billy Vitro <bvitro=user domain=cisco spot balm>
John,
When I read this, a thought occurred to me - are these *sales* numbers or
number of *users*? If the first, then with a shrinking economy, people
may have bought enough Synplicity tools and are just paying maintenance
(which wouldn't show up as big in sales $), meanwhile, there are a bunch
of new purchases of a competing tool just for validation.
Just a random thought of a brain on cold meds....
- Billy Vitro
Cisco Systems, Inc.
---- ---- ---- ---- ---- ---- ----
From: Jack Killingsworth <jjkilling=user domain=verizon.net>
John,
Perhaps the answer is that Xilinx has most of the FPGA market and Xilinx
provides their own synthesis tool (XST) free or dirt cheap. I suspect
Xilinx XST tool users may have been overlooked. Xilinx XST users may also
have been in the other (1% by sales) category and because the tools are
free or cheap the overall FPGA synthesis market ($ value) may really has
shrunk. There is also something good about competition as the differences
between Exemplar and Symplicity decreased the prices should go down. As
design services are outsourced to large design centers in the USA, India,
China there could be more efficient utilization of the tools in larger
organizations, therefore less tool cost per design.
In many cases Xilinx designers had no choice but to use the Xilinx tool
because other tool vendors were late to market with support for the latest
large FPGAs. I use the Xilinx XST tool and find it adequate. XST may not
be the best, but what many of us want is the "cheapest good enough
solution".
I think one of the reasons Xilinx sells more FPGA silicon than Altera is
because Xilinx silicon subsidizes Xilinx EDA tools to make them either free
or cheap. At Altera for a long time EDA tools were a profit center with no
subsidy from the silicon side. Another reason for Xilinx dominance is they
have generally done a better job of addressing board level architecture
issues.
- Jack Killingsworth
---- ---- ---- ---- ---- ---- ----
From: [ An Anon Engineer ]
Hi John,
Please keep me anonymous but I thought I would put in my 2 whatever.
A possible explanation for the numbers you are seeing could be that, 2002
was the year of expense shut-down. Now isn't it possible that management
took a look at software expenses and found that they could get
Altera/Xilinx tools for free while they had to pay for the Synplicity and
other tools. Even if engineers cried themselves hoarse shouting to get the
Synplicity tools, it would have been a no-brainer for management to go
with the Altera/Xilinx tools.
I would however like to point out that I have worked with Synplicity tools
and recent Altera tools. I can not compare on a one to one basis as they
were for different projects. However, except for a short learning curve
with Altera, I don't think there was any major problems in getting to do
what I wanted. Please note the requirements were different but I know in
my company for other Altera projects we went the Altera tools and have
been happy with them.
In general, I think if you are not really pushing the envelope with
regards to speed and area requirements, you can get things done with the
free tools from the FPGA vendors. So why go for expensive tools? It is
possible that Mentor retained their revenue because they could bundle
their tools.
- [ An Anon Engineer ]
---- ---- ---- ---- ---- ---- ----
From: [ An Anon EDA Engineer ]
John,
Anonymously, as usual...
You might want to consider how the data is collected and reported, and
what it is really reporting.
As I understand it, Dataquest reports seat sale revenue only, not
maintenance. You also need to look at how products are categorized by
the vendors. Is Certify an FPGA tool, or an ASIC tool? (to make SYNP's
numbers look better). If you take a look at Mentor, they have a higher
proportion of term deals that result in rolling revenue recognition
every year, even through a downturn where fewer new sales occur.
Point-tool seat sales, especiallly in the conventional perpetual license
model have taken a hammering on two fronts through 2002 and 2003:
- Firstly, as companies lay people off, they still own seats. These
seats can be reallocated to other teams so there are no new buys,
but those vendors with term deals get yearly revenue recognition
that smooths out the bumps.
- Secondly, (and I have this from anecdotal experience) Synplicity is
going back to its' "bomb the price to get the seat" mode of "market
share" acquisition (or desperation, depending on your point of view).
They dump the seat sale price, but try to hold firm on maintenance.
That's great for getting a little revenue (at high cost of sale) and
increasing the customer base on maintenance. However, it doesn't help
with regard to reporting sales numbers to Dataquest. Note that
Synplicity has pretty much the highest percentage maintenance in
the industry at 20% of list per annum so this bolsters their revenue
due to their list prices being remarkably high.
I've also seen them offer one year term deals of Synplify at under $5000.
which means they only get about $3500 recognition for seat sales. Perhaps
Synplicity have many seats, but their cost of sale and support must be
scary. They sure have a hard time making a profit.
What the Dataquest numbers do indicate is that the FPGA Synthesis market
is shrinking in terms of "Dollars Spent", and I believe it really is,
irrespective of design starts. Customers are more willing to get by
with XST or Quartus Synthesis, along with OEM copies of ModelSim, or maybe
a copy of ModelSimPE, or something from Aldec. You'd be amazed how many
designers still verify by throwing the chip on a board to see if it works.
Xilinx and Altera are pitting Synplicity and Mentor against each other
(Xilinx, after all helped Synp stay solvent about 4 or 5 years ago by
"investing" short-term in them) and they keep both starved of true
technonology access. When the data is available to enable full P&R and
static timing analysis by a third party, then you know they are really
sharing. They don't want to be reliant on the EDA industry for their
customers getting to buying silicon. Given the history of third party
FPGA Synthesis it's hardly surprising. I would have made the same
decision. Xilinx is working hard to develop XST into a fairly competent
tool, Altera have QNS (QIS?) with both learning from SynplifyPro,
LeonardoSpectrum and Precision. They are also both in the process of
"emulating" the physical synthesis ideas from Mentor's Physical Group too,
which is annoying to say the least.
IMHO, this market is structurally defective as a result of years of the
FPGA vendors trying to lever each other out of design sockets with free
tools. Many customers now just expect free tools, and place little value
in third party EDA. The truth is that with the limited data we have on
the physical make-up of modern FPGA's the third party vendors can't offer
much incremental value anyway. There's no "Primetime" equivalent for
FPGA's and you really have to wonder why.
- [ An Anon EDA Engineer ]
---- ---- ---- ---- ---- ---- ----
From: [ An Anon Engineer ]
You have left out
Scenario #3: Commodity Reductio Absurdum
FPGA RTL synthesis is now a commodity. Anyone can play.
Xilinx and Altera have increased their market share of the
FPGA RTL Synthesis market as their own tools have become
adequate for all but the large designs. So in fact the
market may have actually shrunk in dollars but the seats are
now at Xilinx and Altera. (Just a thought ... no numbers to
prove)
Please keep me anonymous ...
- [ An Anon Engineer ]
---- ---- ---- ---- ---- ---- ----
From: [ An Anon Engineer ]
I'm one who would prefer to remain anonymous...
Well, it's interesting that Dataquest didn't include any figures for the
synthesis tools built in to Xilinx XST and Altera Quartus, yet they may well
be the tools of first resort for small users and 2002 was probably the year
when those tools became something close to comparable with Synplicity and
Exemplar in terms of quality of results. My guess is that they fall off
Dataquest's radar because they (XST, Quartus) come along for the ride with
the place-and-route tools that users need in any case, and so they don't
necessarily reflect a positive purchasing decision in most cases. However,
they are a completely satisfactory synthesis solution for less demanding
users, and they certainly account for a significant fraction of the FPGA
design starts that we see among smaller users.
What's more, Quartus used to ship with a Altera version of Exemplar. I
can't remember when that stopped happening, but I wonder if the Exemplar
sales increase you saw may perhaps be Altera customers buying Exemplar
seats to fill that gap?
Finally, do the Dataquest figures tell you anything at all about how many
vendor-specific copies of Synplify and Leonardo were shipped by FPGA
vendors, bundled with other FPGA toolkits?
As usual, market analysis numbers tell only a tiny part of the story from
an engineering viewpoint. :-)
- [ An Anon Engineer ]
---- ---- ---- ---- ---- ---- ----
From: Pete Nolan <pete=user domain=nolan-jaska spot balm>
John,
The Dataquest numbers don't necessarily fail my sniff test. Perhaps your
mistake is assuming that there is a high correlation between the number of
designs completed and the number of FPGA designers using a certain
synthesis tool and the EDA vendor's revenue from that tool. This is not
necessarily true.
Synplify and Synplify Pro are mature tools and I would guess that most
users are just paying annual maintenance fees, as opposed to new users
paying the initial license fee, especially in a down economy with many
fewer startups out there. The annual maintenance is typically 20% of the
initial license cost, so that might explain the decrease in revenue.
On the other hand, Mentor is very good, probably better than Synplicity, at
extracting revenue from their existing customers. They have perfected the
scam of selling "new" tools to their existing customers every few years --
add some features, change the tool name, and sell it as a "new" tool. Then
phase out support for the old tool. So all of your customers need to buck
up the new initial license fee every few years, rather than just
continuously staying on maintenance. A good example of this phenomenon
currently in progress is Leonardo being phased out & replaced by Precision.
Another interesting stat to compare to the revenue numbers would be
the number of active licenses (both new and on maintenance) for each
of the EDA vendors in each year.
I concur with all the others about Synplify providing the best quality of
results.
- Pete Nolan
Synplify user
---- ---- ---- ---- ---- ---- ----
From: [ An Anon Engineer ]
John, you may remember the very same thing happening with VERA vs. Specman
back in 2000. After garnering 77% market share, Verisity all of a sudden
was on parity in the verification space with Synopsys at about 43%, and
the verification market shrank overall during the heart of the internet
telecom boom!
Janick Bergeron later confirmed Verisity's lead with a poll in his online
rag that showed VERA usage at a very distant third, if I recall correctly,
behind Forte!! It's a shame that the leading market survey entity in the
space can't get it at least sorta right most of the time. Anon please.
- [ An Anon Engineer ]
|
