( ESNUG 588 Item 9 ) ---------------------------------------------- [11/15/19]
Subject: Costello on Metrics cloud Verilog is at right place, right time
DAC'19 Troublemakers Panel in Las Vegas, NV
Cooley: Costello, last year you announced that you were merging Montana
with Metrics.
And that was going to be super great because Metrics has all
this cloud stuff and you have a Verilog simulator in the cloud
and it's super great...
But oh crap, now the rest of the world is going to the cloud...
What advantage does Metrics have anymore? Did you make a mistake
merging with them?
Costello: No, it's the best move ever made, right? Honestly, it really is.
And I think that yes, everybody's moving to the cloud. The good
news for Metrics, and for other people that are pursuing the
cloud is that EDA still hasn't really made the move to the cloud.
We're starting, you know, Joe Sawicki was talking about, some of
the cloud DRC things that they're doing...
Sawicki invites Anirudh to Calibre 4,140 cloud CPUs lunch
with this burst mode and that's great. That's important.
But there's a fundamental -- if you're going to move to the cloud,
there's a lot of pieces to it. There's a business model piece to
it and there's a technical piece to it.
Let's start with the business model. The business model side
gives you two things. Number one, it gives you a much different
business model because it's a SaaS model and we have not done
that in EDA.
The rest of the world has done that, but not EDA. With the SaaS
model, you buy what you need, when you need it.
I'll give you a great example -- one of our customers that we
just signed. They are 2,000 core simulation company, not huge
but not tiny.
Cooley: Right.
Costello: On average during the year they use 400 cores. At tapeout
time, they need 8,000 to 10,000 cores.
So why the hell do they have 2,000 cores that they bought?
Well everybody's the same way.
The license model; whether you say you're cloud or not; the
license model is a terrible model for the today's compute
needs -- especially for things like chip verification and
other things that are compute intensive.
Because ... I call it the Stopwatch Model. It's right only
twice during the design cycle, you know on the upswing and on
the downswing.
And you've got to catch it fast, because it goes right
through it on the upswing.
So, you're never buying the right amount of licenses and
you're paying way too much for it.
So, number one, you're going to save a huge amount of money
on the software by buying what you need when you need it.
Number two, your designers will be more excited because they're
going to have what they need at the critical time at tape out.
They can go to 8,000 or 10,000 or 20,000 licenses if they want,
to get their design done in a reasonable amount of time.
The second thing they'll save on is hardware, right? The public
cloud is going to be for them much cheaper over the long run
than having their own data center.
So, they'll save on software, they'll save on hardware, and on the
flip side, they'll have short design cycles because got they'll
get all the compute cycles that they need when they need them.
And that's the fundamental change that's got to happen.
On the technical side, Cloud Native applications are very
different. Again, the rest of the world is moving dramatically to
a different approach to building Cloud Native applications, all
containerized from scratch, built from scratch.
That's going to have to happen in the EDA space, too. By the
way, in other segments, the incumbents have been scrambling to
try to figure out how to rebuild all of their software in a
containerized fashion, so you can build real true Cloud Native
flows that deliver the performance and the pricing that's right
for the customer going forward.
So the EDA revolution in the cloud, it started because people
are starting to do pieces of it and the burst mode -- but the
fundamental business model of a SaaS model, pay-for-what-you-
want-when-you-want-it, with a cloud native is coming in EDA...
Cooley: Yeah, you were doing 4-cents-a-minute, if I remember, was the
price for Metrics cloud simulators.
Costello: No, our price isn't 4 cents a minute. That was you know like
the list price for an EDA company's tools. (laughter)
Cooley: Alright.
---- ---- ---- ---- ---- ---- ----
Related Articles:
Naveed and Costello discussing how SiFive/RISC-V will crush ARM
Raik on OneSpin's four formal technologies the others don't have
Costello on Metrics cloud Verilog is at right place, right time
Join
Index
Next->Item
|
|
