( ESNUG 588 Item 12 ) --------------------------------------------- [01/16/20]
Subject: Anirudh on Green Hills, Black Duck, CDNS Clarity and Ansys HFSS
DAC'19 Troublemakers Panel in Las Vegas, NV
Cooley: Anirudh, you put a $150 million into Green Hills? Synopsys is
putting 1/2 a billion into Black Duck? Are you guys trying to
get out of EDA?
[ Editor's Note: Green Hills makes real time operating
systems and Black Duck does security software for
ship containers and open source projects. - John ]
Anirudh: EDA is a very, very critical business, and it's our core
business. If you look at it, we have massive investments in
EDA and IP.
But when we talk to customers, they also want us to move to the
system level.
This is not a new trend, but it has picked up in the last several
years, because our customer companies are also innovating at the
system level, and they want to look at the whole system.
So, we want to see the areas which are more synergistic with our
core business that can go to the system level. So, there are
definitely two areas I believe are very synergistic with EDA and
IP that extend to the system level. When you extend to a new
area, you want to make sure that you extend with a core
competency -- with some skill set that you can offer -- because
there are other players in those spaces.
So, the core competency we have like I said is computational
software, whether it applies to Machine Learning or other things.
And, the other area that is computational is system analysis.
So we have a big effort on system analysis & we launched Clarity
as a first computational product in that space. (ESNUG 586 #5)
Cooley: Right.
Anirudh: So, system analysis is definitely synergistic and leverages our
computational expertise. The 2nd area which is very synergistic
for Cadence is embedded software -- especially embedded security.
We believe that Green Hills is a leader in that field. So we
have a big partnership and relationship with Green Hills.
EDA and IP are our core business; but as our customers do more
and more system, we want to move to these two areas which is
system analysis and the embedded software.
Cooley: So, it's adjacency you want to jump into.
Anirudh: Exactly -- to provide a better solution to our customers.
Cooley: You mentioned Clarity, and one of the things that came up was when
you did the Clarity launch was some user feedback (ESNUG 588 #4)
saying you're going against Ansys HFSS, and that basically you
have a parallelized process going against a mostly non-paralyzed
old tool.
"My Ansys AE thinks that Cadence Marketing might be confusing
algorithmic speed-ups due to massive parallelism through
distributed computing with a brand new matrix solution."
- from http://www.DeepChip.com/items/0588-04.html
Once HFSS wakes up and becomes parallelized, won't you lose your
speed advantage against them? Won't Clarity be just on par with
Ansys HFSS then? There will be no difference.
Anirudh: Clarity is not just parallelism. Parallelism is a very important
thing; we had a whole discussion about cloud, about availability
of hardware, and parallelism is a very critical piece of software
development.
But let me talk about Clarity. Clarity has 3 main things, and
parallelism is only one of the three.
The first thing is, meshing is very critical in finite element.
We are modeling 3D services. So meshing in Clarity is very good.
The second part is the solver itself, whether it runs on one CPU
or each CPU. It doesn't have to run and 1,000's of views; you
can have a new solver technology that can be fundamentally better
than what has been done in the past.
So, in Clarity there is combination of 3 things - there is the
meshing technology, the solver technology, and the parallel
technology. So even on same number of CPUs, Clarity can give
a significant speed up.
And sometimes the parallelism is useful not just for speed up.
Cooley: So if you have one CPU for Clarity and one CPU for HFSS, would
they be significantly different?
Anirudh: Yeah, that's our experience. In the last few months, we have
done a lot of engagements with Clarity and so far the results
are promising. I mean, it's still early days.
Cooley: Numbers?
Anirudh: It's promising. Sometimes even a single CPU, or the same number
of CPUs can be more than 5X. It's not just the parallelism by
itself is the differentiator. Now it is icing on the cake, but
you need the cake, also.
Cooley: Right.
Anirudh: OK, let me back up a little bit. If you look at finite element,
finite element is a well-known method; maybe 1940's was the
original finite element.
The best thing about finite element is that you can model any
3-D surface, and the physics is relatively well understood.
What happens is once you do all the measuring and model, it
becomes a very very big matrix, that's the problem with finite
element.
Cooley: Right.
Anirudh: It [finite element] is very general purpose, but then it will
run very, very slow.
What we [CDNS Clarity] are doing is, we are not really changing
the physics of it -- we are not making any approximations. But
the matrix that you get, you can solve much more efficiently.
And the reason we [Clarity] can solve it much more efficiently
is because like I said, we have computational expertise. So the
matrices we [Cadence] have solved in the past, in other domains,
are billions of nodes, 50 billion node matrices we can solve.
Whereas, if you go to the system space, even in finite element
or other system domains, millions of nodes is a big matrix,
30 million is a big matrix. We at Cadence have several orders
of magnitude advantage in computational algorithms that can be
applied to this system space. That's the most exciting thing.
And the parallelism is only one part of it. How efficiently you
do the solvers? How efficiently do you do the meshing without
any accuracy loss?
Clarity is fully accurate. There are no approximations in the
physics or accuracy of it -- you can just do it much more
efficiently.
And it's not just about speed, though speed is important. What I
found in our initial Clarity engagement is capacity.
Because with systems you want to simulate bigger & bigger things.
People want to simulate phones or cars and planes. The [extra
Clarity] capacity is critical, not just the speed and so it is
well beyond just parallelism. The parallelism is an important
factor in today's software development, of course.
This is still early days [going against Ansys HFSS], but we are
very cautiously optimistic that this is a synergistic area for
us. This is an area that the demand keeps growing because you
want to simulate more and more things, and we are reasonably
confident that our computational expertise can be applied to
this space.
Cooley: Okay.
---- ---- ---- ---- ---- ---- ----
Related Articles:
Sawicki on cloud EDA is 12 hours of burst compute on 6,000 CPU's
Mo on tweaking your PLL/DLL for SoC vs. tweak your SoC for PLL/DLL
Anirudh on Green Hills, Black Duck, CDNS Clarity and Ansys HFSS
Join
Index
Next->Item
|
|
