( ESNUG 454 Item 1 ) -------------------------------------------- [04/28/06]
Subject: Viewers react to Magma CEO's claims in "Cooley Does Rajeev" video
> While in San Jose for SNUG last month, I sat down with Rajeev for a
> quick on-camera interview and it turned into him spilling his guts
> on 3 new product announcements, on the Synopsys lawsuit, on Ambit,
> Magma, PrimeTime, Nassda, Synopsys, Cadence, Monterey, Cobra, the price
> wars, Mojave, EDAC -- and his unhappy mother. Enjoy.
>
> http://www.deepchip.com/demos/rajeev.fhtml
>
> I recommend you view it at the 150K speed; I had bandwidth issues
> trying to watch it at the 300K speed.
>
> And when you're done, I'd love to get your reactions to any of the
> topics brought up in this interview to publish in the next ESNUG.
>
> - John Cooley
> ESNUG/DeepChip.com Holliston, MA
From: Jared Bytheway <jbytheway=user domain=cirque spot gone>
Hi, John,
Most of my experience has been in design with netlist hand-off to Silicon
vendors. I watched the Magma demos and your interview with Rajeev. I am
very impressed with Magma's apparent capabilities. I have used DC and
PrimeTime for a number of years now. I have always been amazed at how
cumbersome they are to use. I have also used Synplicity; which has given
great results and easy to use.
I think with companies like Magma and Synplicity, Synopsys should be
very nervous.
On a topic in your interview, I think it's obvious why the Big 3 don't
like Magma because it's the new guy that is taking market share in droves.
The Big 3 are scared!
Another point is that I don't think price is a very important criteria
for which tools you buy. Companies that I've worked for, price is
number two behind performance.
Keep up the investigational probing -- someone needs to sort the BS from
the truth!
- Jared Bytheway
Cirque Corporation Salt Lake City, UT
---- ---- ---- ---- ---- ---- ----
From: [ Chicken Little ]
Hi, John,
I like that you asked aggressively and brought Rajeev back to your line
of questions.
Maybe will be useful to send a survey about Magma tools ONLY and ask the
users to see the movie and tell you what they agree or disagree with from
the talk with Rajeev.
Please keep my comments anonymous on ESNUG
- [ Chicken Little ]
---- ---- ---- ---- ---- ---- ----
From: Sandeep Srinivasan <sandeep=user domain=synchronous-da spot gone>
Hi, John,
Very interesting interview!! Nice work..,
- Sandeep Srinivasan
Synchronous DA Palo Alto, CA
---- ---- ---- ---- ---- ---- ----
From: Jeff Echtenkamp <echtenka=user domain=broadcom spot gone>
Hi, John,
On your questions "Is Rajeev lying on is BlastFusion a physical synthesis
tool?", and "Is unmap the first command run?" -- Raj is right, the first
command (or one of the first commands) is unmap, and everyone I know uses
it this way. This would be what I consider a true statement, and the
people who don't run unmap are probably in the strong minority.
So, does this conflict with your question about Fusion not being a physical
synthesis tool... really depends on your definition. I'm guessing what you
are saying is right as well, that most people probably use
DC -> Fusion vs. BlastRTL -> Fusion
But its a matter of what people are using DC for.
In our old Apollo based flow, the gate level netlist from synopsys was not
unmapped and it was more of a pure P&R flow, with some timing optimization.
In Fusion, the netlist from DC does matter... you want to get the right
architecture for the DW components, you want to pick the right pipelining,
etc. But, what doesn't matter is bufferring or gate sizing. With the
unmapped and unbuffered netlist, Magma does its initial placement exploration
looking at the potential to meet timing, rather than the ability of the
initial netlist to meet timing. If you think about it, since we all use
timing-driven placement, in the unmapped flow, if there is a difference
between synthesis estimates and placement, TDL might come up with a bad
initial placement and get stuck in a local minima. In Fusion's case, since
they are looking at the potential to meet timing, the initial placement and
estimate is alot closer to the final implementation.
Physical synthesis now being hype:
On your comment of physical synthesis being a buzzword that nobody cares
about anymore. I'm not sure I agree with this statement. It might be that
because of advances in synthesis and P&R, these two are getting closer to
each other so the theoretical gap between a netlist->GDS flow to an
RTL->GDS flow might be small. But, there is another thing to look at...
an RTL->GDS flow could be run by one engineer in one tool using one set
of scripts and not having to worry about correlation. A few years back,
you had 3 people do timing closure... a synthesis person, a P&R person and
someone doing post-placement optimization. When P&R tools started having
good logic optimization and correlation to PrimeTime/Star, you could
eliminate that final loop and it was now just a synthesis person and a
P&R person. When someone gets the RTL->GDS flow complete (and by complete
I mean competitve on all grounds and easy to use), in theory now we can
move down to a single person running this flow. That enables my frontend
engineer to worry about RTL architectural stuff and my PNR guy is now the
implementation guy.
Is Mojave hype?:
Anyways on your comment in your interview with Raj on Mojave and "is Mojave
an all layer, polygon based tool". I have used this tool personally on
designs and also done an comparison of randomly generator errors with it
and calibrated it to Calibre... so the answer to this is coming from my
actual experience, not from marketting slides... anyways, the Mojave DRC
tool is definitely an all layer polygon based tool (in the same league as
Calibre, Hercules, PVS, Dracula, etc) and is NOT just a metal layers-only
tool as you asked. So, Raj is not fibbing.
I think this "rumor" is someone who has Fusion experience commenting
about the internal database and DRC checking within Fusion. Fusion is a
router, and thus doesn't need to care about "base layers". The internal
DRC tool "check route drc" built into fusion is only metal layers, as
the fusion database only uses base layers (to save on data). "check
route drc" is NOT Mojave's tool... its a fast "implementation mode" DRC
similar to what is built into Astro and SOC Encounter. The Mojave
toolset is a completely separate set of tools and engines. As you ask,
Mojave CAN read in a full GDSII database and perform an all layers set
of checks on it.
On Magma tool integration and signoff:
In regards to being a closed solution, I have a tendancy to agree with
Raj on his answers. The reason the competition is "open" is the way
they grew... through acquisition. Look at Cadence for example, you have
BuildGates/Get2Chip, CeltIC, VoltageStorm, SignalStorm, First Encounter
and NanoRoute. These tools were all done by startups and brought
together by Cadence. Now, even Cadence is pitching a common database
(OpenAccess). The other alternative is LEF/DEF. People would call both
of those "open interchange formats"... The reality is they are Cadence's
formats that were openned up. Synopsys is similar... They have .lib
which for the longest time was "their" format, until they were forced to
open it up and call it Liberty. On input/output they didn't have an
issue, as you are talking RTL and gates, where there are standards
(verilog). Avanti always had Milkyway for their database, and were
"somewhat closed" if that's what you want to call it, like Magma is.
You could use lef/def/.lib with both... But they also supported their
own native models. Of course, the difference is Magma is pitching a
common memory model (eliminating the need for interchange formats). In
theory, if you wanted, you could use a DEF/LEF/GDS/.lib-based flow into
Magma's tools, just like you did with any of the others and I see no
reason why it would work.
Why no Magma point tools:
So why don't you see people using some Magma products standalone? I
think first is, if you look at things like DC, PrimeTime, and Star-RCXT
these are guys with lots of legacy, where the design wins happenned
before Magma started. Even First Encounter probably got its traction
in the market before BlastPlan became a competitive product. As Raj
mentioned, when you are a startup, you don't go in thinking of a project
where you tie, you have to win and have compelling reason to win. I'm
guessing in these cases this is what you are seeing.
Now, the flip side, is in Magma's case, once you are Fusion users, there
is an additional incentive to use their products... While I might not
choose QuartzTime or QuartzRC if I wasn't a fusion user, the ability to
have my P&R tool be able to call my signoff tool now gives me that
compelling reason and I now bite the bullet to abandon legacy. I would
agree with Raj's comments about Primetime... The reason people needed
primetime (and Star) was that initially our P&R tools (Apollo) had
really crappy STA and extraction built in. I didn't use PT because it
was awesome, I used PT because Apollo's timer sucked. Magma is closing
the gap on a lot of their tools to the point its now worth considering
dropping industry standard tools (like STAR, PrimeTime, CeltIC,
VoltageStorm/Redhawk, etc). And honestly, as time goes on, I think we
all can see these tools "point tools" being integrated is a big must,
and they need to be of the same calibre (aka signoff worthy). As a
Fusion user, I'd rather Magma spend their time achieving that goal
first, than trying to correlate and talk with other tools. As a
shareholder, I might not make this comment, but as a customer, I like
it. Granted, there is still a concern that in some cases, its nice to
do the whole "belt and suspenders" thing and have your signoff and
implementation different. (God knows who many different Primetime bugs
we found once we got Magma's timer correlating to it and could now see
their mistakes... Kinda scary)
On Magma getting the "low hanging fruit of the P&R world":
First off, we P&R guys are all religious about our tools... So I'm sure
this is a statement of a guy who guys to church at Synopsys :) Anyways,
I would disagree strongly with that statement. Honestly, all backend
tools have issues, and it's a matter of the designer's preference for
working around things. If you look in our industry, there are a lot of
different mentalities. P&R started out as glorified layout... a very
graphical thing done by technicians. Today, it's a very complex problem
often run by guys with Masters in EE. You also have ASIC houses who get
netlists blind from a customer. You get companies who do large
bottom-up SOCs. We all have our styles for a reason.
If there is a trend I see:
Cadence -> they get the guys who need capacity and to some degree a
packaged solution where I can drive it from the GUI or with very little
scripting. The other camp that is starting to like Cadence is because
of the tie-in to Celtic for crosstalk and VoltageStorm for analysis and
closure (again, my theme of being able to have a signoff tool in your
implementation tool)
Astro/IC Compiler -> frontend guys who are DC/PhysOpt users, as it's a
simple change for them. Also, backend who tend to like a nice GUI with
reasonable scripting abilities. Designs which have a lot of legacy with
DC/PC also seem to go better through this flow, so as a result "out of
the box" you might get better results than with Fusion.
Fusion -> backend guys who like programming and like control, and the
ability to tweak optimization algorithms specifically to their design
style. (A good example, in Apollo I had 3 routing commands... Global,
track, and detail... In fusion, you probably have something like 7-8
different commands with a lot more options... When the default works,
I'd rather have Apollo's as its easier to use... When things don't work,
Fusion is nicer as you can call the specific routines to clean up the
specific violations you see via code... This is great for systematic
errors) People who like a nicely integrated tool without going in/out
of the tool. The GUI isn't the greatest, but most Fusion users I know
rarely call the GUI, they do more via TCL. Magma's database access and
coding is hands down better than the competition.
I think in general, the benchmark's I've seen, Cadence and Synopsys have
their strengths and weaknesses, if you look at Magma, they might not be
the best at every point, but they aren't the worst typically either.
(Which, I guess ties in with my point about point tools and why Magma
customers seem to get the "full deal".)
- Jeff Echtenkamp
Broadcom Irvine, CA
---- ---- ---- ---- ---- ---- ----
From: [ A Wall Street Analyst ]
Hi, John,
There were some comments by Rajeev that you were skeptical of. I was
wondering what your follow-up showed. Thanks. Anon please.
- [ A Wall Street Analyst ]
---- ---- ---- ---- ---- ---- ----
From: James Beckett <jbeckett=user domain=magma-da spot gone>
Awesome. Thank you. John Cooley is by far one the best interviewers I have
ever seen. He is focused and is not afraid to ask and continue to address
issues which most will run from.
- James Beckett
Magma DA
---- ---- ---- ---- ---- ---- ----
From: [ A Synopsys Salesman ]
Hi John,
Time flew during this interview and the content was extremely insightful.
However, I agree with you and disagree with Rajeev on the "low hanging
fruit" perception. Synopsys is being used for the tougher designs and
to bail out Magma. Also, price ALWAYS comes up - "period" or "no period".
I hear it ALL the time that we can get "x" Magma licenses for the price
of "y" Synopsys licenses and therefore want to use Magma for volume.
Thank you for sharing this interview with the community.
- [ A Synopsys Salesman ]
---- ---- ---- ---- ---- ---- ----
From: [ Mighty Mouse ]
Hi, John,
Why not do separate in-depth videos of Aart, Mike, and Wally? I would
love to see those interviews, too.
- [ Mighty Mouse ]
Index
Next->Item
|
|
