( ESNUG 463 Item 8 ) -------------------------------------------- [03/16/07]
Subject: Answers from Gary Smith, Gary Smith EDA
> Gartner shut down the EDA analysis service because it was not a money
> maker. The large EDA vendors are setting up their own "spin-alysis"
> departments. How are you going to make it now? Why will the EDA industry
> pay you now when they didn't pay Gartner then?
There were multiple problems. One was our prices were too high. There are
over 450 EDA Companies and there were less than 10 of them that could
reasonably be able to buy our service. We've fixed that.
---- ---- ---- ---- ---- ---- ----
> Gary Smith: Now that you are running your own business, will you now have
> your predictions change to something real? Like - ESL is never going to
> be a real market?
Tell you what. When somewhere between 2008 and 2013 you can no longer get a
job as an RTL engineer (or marketing type) give me a call and I'll help you
get a job in the Design (ESL) world.
---- ---- ---- ---- ---- ---- ----
> Since the demise of Dataquest, marketing pukes within EDA companies now
> feel they have license to be even more creative with their EDA marketing
> statistics... do you have recent examples of such gross inaccuracies or
> exaggerations? Be specific.
EDA vendors have always done this. We scrub the hell out of the numbers to
come up with our "estimates". The 2 that were the worst was
1.) Cadence reporting 126% of revenue
and
2.) Synopsys reporting 5 quarters of revenue
Keep in mind we publish product revenue. Companies do consulting, have one
time revenue hits, add revenue from acquisitions (we fold in the acquired
company's past revenue so you see "real" growth), etc. You can hide a lot
of things in your total sales figures.
---- ---- ---- ---- ---- ---- ----
> Gary Smith - Now that you're free from Dataquest how is business going?
Going well. It helps a lot if you listen to the people that want to buy
your service. We were immediately contacted by start-ups. They really
helped us put together offerings that were affordable. About the same time
we started seeing the financial community wanting to sign up. What was
great was the increased interest from the user community. Our former
semiconductor clients have returned and we've had a sharp increase from
other User companies. We of course have our loyal EDA clients back.
---- ---- ---- ---- ---- ---- ----
> In your opinion is there any future for "open source" EDA? Are we going
> to see in the future complete systems like Linux?
That falls into the "define EDA" category. I've been accused of having far
too big of a definition of EDA, because I include all of the design tasks
including embedded software and eventually mechanical design. I don't
think in EDA, as it exists today, there is much room for open source tools.
However in the ESL market there may be. Keep in mind most of the ESL market
will not address IC Design. The largest portion of ESL will use processors
and FPGAs as their hardware solution. That is where the large increase of
seat count comes from. With enough users you have a much better chance to
market an open source solution.
---- ---- ---- ---- ---- ---- ----
> Gary Smith - Some have stated one of the reasons why Gartner dissolved the
> EDA group was because EDA companies did not invest in EDA research. Do
> you think this is true and if so you think the attitudes amongst the EDA
> companies have changed since then?
I'd have to say I thought our high price point was more to blame for all but
the larger EDA vendors. Interestingly enough we have not seen a big rush of
EDA companies wanting to sign up for our service, so it could be true.
We're getting financial guys, semiconductor guys, and EDA start-ups signing
up this this. But other than my longstanding big EDA clients, we're not
seeing a large inrush of the medium-sized EDA companies.
---- ---- ---- ---- ---- ---- ----
> Why is Gary the last analyst to cover EDA? Why EDA such a weak business?
OK, it'll sound like I'm bragging, but EDA is one of the most technical of
the electronic markets. You need the technical expertise (it helps if you
come from the user community) and a strong marketing background. Having
been around for almost 40 years helps with the history. Add to that it's a
lot of work (definitely not a 40 hr a week job) and you have a hard slot to
fill. If you don't love the technology and if you aren't a little bit crazy
you probably won't take the job.
---- ---- ---- ---- ---- ---- ----
> Why don't we have Ajay Daga the CEO of Fishtail joining the conference?
Hell, if I know. Lack of marketing probably.
---- ---- ---- ---- ---- ---- ----
> Question for Synopsys/Cadence/Mentor: Will the industry support a uniform
> coverage database standard to be shared across all tool vendors, inclusive
> of code/line/toggle coverage, functional coverage, formal reachability
> analysis, etc? The history: Synopsys, Cadence, and Mentor all appear to
> be developing their own, propriety coverage databases.
This is interesting. The question is why should they? The big three:
Synopsys, Cadence and Magma in the back end and Cadence, Mentor and Synopsys
in the front end have never been technically closer in performance as they
are now. We have had 5 years of a brutal price war that's left EDA vendors
profits at 50% of what they should be. Although the IC CAD market is now
growing again, because of the DFM issues, the RTL Design market will
probably be flat to down (depending on the adoption rate of the ESL tools)
just as the Gate Level market of the 1980s became flat-to-down due to the
adoption of the RTL tools.
Tactically the best way to regain profitability is to lock in your present
customer base with proprietary tools. After all, the price war did not
significantly change the market share of the warring parties. However I do
think that the industry is maturing and it appears that most of the large
EDA vendors are listening to their customers rather than retreating behind
the walls of a proprietary castle.
But don't be surprised if in the end we have 3 separate proprietary tool
sets, especially in the RTL Design market.
---- ---- ---- ---- ---- ---- ----
> Is anybody buying your service other than the DVCon panel committee? Why?
Yes, they're buying it to make money.
---- ---- ---- ---- ---- ---- ----
> Why did the the IC implementation industry (other than Plato) not see the
> switch to multi-CPU machines happening? Why didn't they proactively
> pursue it so that when AMD/Intel had machines, you would have code which
> scales linearly with # of CPUs? Gary, why didn't you tell EDA vendors to
> pursue this?
This gets into the present SW crisis. To make a long story short there are
two related problems. First is that once you exceed 4 microprocessors you
no longer have a general purpose machine. After that you enter the world
of super computers. There are 2 classes of supercomputer programming tasks;
Embarrassingly Parallel Programs and everything else.
In everything else you run into Amdahl's Law. So we can only address the
Embarrassingly Parallel Programs, like some of the DFM problems, with a
multi-core supercomputer -- as with the Mentor/Mercury partnership. We can
so some SPICE work by throwing processors at it. But after you move away
from the Embarrassingly Parallel Programs you're in deep trouble.
The second problem is we now have a *sequential* programming infrastructure
built around C. We need a *concurrent* programming infrastructure built
around ????? The company that does it will be worth $2 billion.
---- ---- ---- ---- ---- ---- ----
> Gary: "We have gone from having 1 verification engineer for 2 design
> engineers to 4 verification engineers for 1 design engineer - when and
> how is that ratio going to change?"
If by we you mean your group, I'd say you're in trouble. The percent of
effort going into verification has been fairly consistent, ranging from 35%
to 70% (depending on the complexity of the SoC), for the last 7 years.
Since the gate count has been skyrocketing, for that period of time, we are
doing an OK job (not great but OK). Now the dream of verification shrinking
back down to 15% of the effort is just that, a dream. We should be able to
get it down to 35% to 40%. The "how" is twofold.
First is the automation of the RTL Verification flow. I think Cadence,
Mentor and Synopsys are doing a fairly good job at that. The verification
effort wouldn't have remained flat if they hadn't.
The second step is the development of a solid ESL verification flow. Most
of the "RTL forever" types forget that the majority of the verification
effort is shifting to software. After that it really gets exciting as we
address the validation problem. That's what the Rosetta language is about.
---- ---- ---- ---- ---- ---- ----
> The joint Si2/Accelera low power workshop last October left no ambiguity.
> All the customers reported clearly that the most important feature of a
> low power standard is that there is only one standard. Forget Si2, forget
> Accellera, Cadence and Synopsys can make this happen if they want to.
> Why don't they want to?
As I said above, they don't want to because they want to make money. That's
been really difficult the past 5 years. The question is will the white wolf
win or will the black wolf win?
---- ---- ---- ---- ---- ---- ----
> I have just one line of questioning, for Gary Smith. It is how can we
> trust his "Hot@DAC" list when you are a consultant-for-pay? Will he
> select anything other than his customers and his favorites ESL and DFM?
Let's see; in the past the list has been mostly start-ups, that couldn't
afford our service and the Big Four, of which half didn't take my service.
So I guess as a consultant-for-pay I wasn't doing a very good job was I?
Now, at Gary Smith EDA the start-ups can afford my service so you better
keep a close eye on me to make sure I stay honest. ;-) Another take is
that most of my clients are Users and Financial types. The "Hot@DAC" list
is actually something I do for the engineering community so I try to
list what is hot in the design world.
---- ---- ---- ---- ---- ---- ----
> The internet has changed the market for business info. Many day-to-day
> info needs can be satisfied with web searches. Other than to back a VC
> funding event, why would anyone need the services of firms such as yours?
The web is wonderful. But it has one detrimental attribute. You have to
do directed searches. It ends up the most important information is in
adjacent areas. That's why magazines and newspapers are so vital to real
knowledge. Every breakthrough I've made is because I've known something
outside of my area of specialization. The job of a specialist is to know
more than is what is on the web. I love the web. It keeps me from having
to answer simple questions that can now be found in a web search.
That's not counting the "quality" of information that's out there.
Keep in mind one of the major reasons people asked me to keep my service
going is to keep the EDA vendors honest.
---- ---- ---- ---- ---- ---- ----
> In your opinion, which of the EDA editors provides coverage that is the
> most useful to the EDA users? To the EDA vendors?
Richard Goering is the dean of the editors. Mike Santarini and Ron Wilson
(when he writes on EDA) are doing a great job at EDN. Cooley is the master
of providing the user's view on the industry. Gabe has been doing a great
job keeping our nose to the grindstone and Max spins tales that stretch our
imagination while educating us. Peggy does a good job taking a different
slant at the market. Tets' analog background (plus time in the tunnels of
Vietnam) gives him a needed perspective. Dave Maliniak and Ann both do
their best to keep EDA in the news and their message of the importance of
EDA is vital to our community.
Index
|
|
