( ESNUG 585 Item 5 ) ---------------------------------------------- [11/20/18]
Subject: Costello on Montana's 5X to 20X speed-up and merger with Metrics
DAC'18 Troublemakers Panel in San Francisco
Cooley: Okay. Costello. Before we go into the Montana stuff, you guys
decided to merge with Metrics.
And Metrics' big claim to fame was "we'll sell you Verilog at
4 cents a minute". Anirudh, one of the biggest providers of
Verilog on the planet, just said 'no we'll just do you same
model on the cloud as it is". Did you just buy a pig in a
poke? Was that a disaster? That merger?
Costello: No, not at all. Because I mean look at everything. I've been
away from EDA for a while. I have a very different perspective
about it.
And probably a longer perspective. I think all of this that
we're talking about right, it's obvious, everything is going to
be in the cloud 5 years from now. Everything.
I mean fundamentally that's going to happen. And why? Because no
one can build their own cloud as cheaply as the people that are
doing it at Amazon, Google, Microsoft, etc. For that reason,
it's going to happen for sure. And having the flexibility and
scalability.
So that's a fundamental thing happening. That's one of the
reasons I thought Metrics was on the right path building a
simulator and a simulation environment that was completely
cloud-oriented.
We at Montana, needed a front-end simulation capability to go
with our new backend capability and it's a perfect match that
way. And they have the right model for the future about how
you're going to run all simulation. All of it's going to be in
the cloud, no matter what kind of engines are running in the
background.
Cooley: Okay you talked about Montana five years ago. (See ESNUG 526 #1)
Why should we... alright I'll back up. What is Montana?
Costello: At Montana, we built a new architecture for a processor that runs
Verilog directly. That was the whole idea - or System Verilog
directly. And yes, we talked about it five years ago and the
reason it took five years is that it's a heck of a problem to do
it that way.
And we started with the wrong team of people. That were very
good on the software side and not so good on the hardware side,
and that didn't work out.
And then we had a second group of people who were better on the
hardware side, but not good enough.
And then we finally got the right hardware team in place and they
were able to build that kind of processor architecture that could
directly run that.
And it's completely different. Instead of running on Intel
you're going to run on a specialized processor that runs
simulation - dedicated simulation - much, much faster. Even on
an FPGA implementation, 10-20x faster. And eventually it will
be on an ASIC.
Cooley: Have you met Anirudh?
Costello: Yeah, we know each other quite well...
Cooley: He has something called Rocketick that's like super fast...
Costello: Running on Intel processors
Cooley: And Synopsys has Cheetah...
Costello: Running on Intel processors
Cooley: Well you made the claim, you said you were... I think what you
said when I interviewed you before is you were...
what the hell... 5X to 20x faster than Cadence Rocketick or
Synopsys Cheetah. Yes, no?
Costello: Yes.
Cooley: Care to elaborate on this?
Costello: The problem had is... you know we had only done benchmarks on a
few circuits, we haven't done a full simulator yet. That's why
we did the merger with Metrics. We need a full simulation
front-end.
But in doing some of these experiments on circuits as big as
600 million gates, we run 10, 20 times faster sometimes than
the hot parallel tools on standard Intel processors. Because the
standard Intel processor was never meant to do simulation.
---- ---- ---- ---- ---- ---- ----
Related Articles:
Joe Sawicki smirks at Cadence Pegasus' 3 big critical DRC failings
Costello on SNPS PnR "still in catch up mode" in 2 years from now
Movellus Mo's 6+ GHz 100% tweakable auto-generated PLLs/DLLs/LDOs
Dean on hybrid cloud bursting EDA flows with IC Manage PeerCache
Costello on Montana's 5X to 20X speed-up and merger with Metrics
Join
Index
Next->Item
|
|
