( ESNUG 585 Item 7 ) ---------------------------------------------- [03/27/19]
Subject: Costello dissecting Montana, Rocketick, Palladium, Zebu, Veloce
DAC'18 Troublemakers Panel in San Francisco
Cooley: Well you made the claim, you said you [Montana] were...
I think what you said when I interviewed you before was
[Montana is] 5x to 20x faster than Cadence Rocketick or
Synopsys Cheetah. Yes? No?
Costello: Yes. Because the standard Intel digital processor was
never meant to do Verilog simulation.
Cooley: Anirudh. Are you regretting buying Rocketick and maybe
you should have bought Montana instead?
(audence laughs)
Anirudh: First of all, it's good to see Joe back in the game, right.
(audience claps)
And you know there are not that many startups in EDA, so you
know we want to encourage more and more startups and as you know,
sometimes we have bought enough startups in the past several
years you know. But in terms of verification I do want to put
it in perspective before I answer the question on Montana.
There is a continuum of solutions, right; all the way from formal
and static, to simulation, to emulation, to prototyping. So, I
think what Joe is talking about is simulation running on some
FPGA or specialized hardware. So that is not appropriate to
compare to Verilog simulation running on Intel. Even though we
believe we have the best...
Cooley: Well he's [Costello's] claiming that his [Montana] architecture
is tuned for Verilog saying it's going to be 20x faster than
what you [CDNS] have to offer because the best you can do is
an Intel processor.
Anirudh: As I said we [CDNS] have a continuum of solutions, and we have
a specialized processor that runs verification -- it's called
Palladium. And it's been doing it for years and years and
it's the #1 Verilog accelerator in the market.
In Palladium there are several use models. So and of course
in Verilog simulation on Intel processors you can run on
single CPU, you can run multiple CPU.
But if you go to Palladium itself there are multiple use models.
Sometimes people only think of emulation or In-Circuit Emulation,
but Palladium for a long period of time has offered what we call
simulation acceleration -- in which you can accelerate your
simulator on a Boolean processor and we get, you know, in the
range of 10x to 100x faster than logic simulation.
Now if you go full emulation you can get 1000x faster. So I
think you have to have a complete solution across all these
platforms, that's the way to solve this problem. So if you want
to compare a specific processor [like Montana] you have to
compare it to a specific [Palladium] processor that we have,
or an FPGA solution.
Costello: I agree with that. I mean that's exactly right. There is a
complete spectrum of simulation. If you're going to have a full
environment that's going to solve these problems you're going to
have to have all the weaponry. You're going to have to be able
to run on Intel CPU processors in the cloud where you're going
to have an infinite capacity. You're going to have to have the
high-end emulation or prototyping things.
And where we fit in, you know, is in that middle ground between
those two [emulation and prototyping] So, I think it's
exactly right, I agree with Anirudh.
Cooley: Okay.
Costello: And by the way, Rocketick was to push one piece of it forward.
Originally, they targeted the [Nvidia] GPU thinking 'Ooh...
that architecture!', but that architecture sucked also for
doing Verilog simulation -- it was never intended for Verilog
simulation.
So, it [Rocketick] didn't work out well -- but the parallel
part worked, so it extended the capability for Verilog
simulation in the Intel CPU environment nicely. But that
still doesn't make it [Rocketick] an emulator or something
that fits in between.
Cooley: Sawicki.
Sawicki: Yes, sir.
Cooley: Why did Mentor Veloce lose to Synopsys ZeBu in the market share
to sales to Intel?
It was widely reported in a lot of the press that Aart's ZeBu
displaced Mentor's Veloce.
Sawicki: I'm going to try to ignore part of the question, because I won't
confirm anything in terms of losing or not losing. We still have
customers in the market, some of whom I can't talk about.
But what I will say is we had one classic problem: we were about
two years late on the hardware with [Veloce] Strato.
Strato came out finally last year [2017] we've had a really good
run since then. We've taken a lot of the share back, but when
you're 2 years late...
Cooley: Against ZeBu or against ZeBu and Paladium?
Sawicki: It's really hard to isolate one versus the other. I mean we're
out there all beating the crap out of each other pretty much
on a daily basis.
But you know we had a system in the in the market it was based
upon old [Crystal 3] chip about for 2 years to 2.5 years more
than we wanted to be.
I'm really happy with some of the stuff we'll [Veloce'll] be able
to do now in terms of accelerating some parallel paths in terms
of hardware development. So hopefully we don't ever have that
[market losses] happen again. But simple, we were late on a
chip, that hurts.
Cooley: Alright.
---- ---- ---- ---- ---- ---- ----
Related Articles:
Anirudh and Sawicki on why CDNS and MENT did the Cloud this year
Costello dissecting Montana, Rocketick, Palladium, Zebu, Veloce
28nm vs. 7nm, AMS, Virtuoso, CDNS Innovus vs. SNPS IC Compiler 2
Costello on EDA ossification, cloud, and RedHawk vs. GreenHawk
Join
Index
Next->Item
|
|
