( ESNUG 588 Item 10 ) --------------------------------------------- [01/16/20]
Subject: Sawicki on cloud EDA is 12 hours of burst compute on 6,000 CPU's
DAC'19 Troublemakers Panel in Las Vegas, NV
Cooley: Joe... Sawicki. (Again.) You issued an invitation in DeepChip
to the person to your right, Anirudh, to come "See how AMD
scaled Calibre to 4,140 CPUs in 10 hours." (See ESNUG 587 #2)
Sawicki: [ grins and chuckles ]
Cooley: Is there anything you want to share with Anirudh here before
he goes to your panel or goes to your talk?
Sawicki: Well, if I tell him (Aniridh) he's not going to come. You know,
the fun with DeepChip aside. The cool thing about that one is
if you look at cloud, cloud EDA is running differently than
it's running for the rest of the planet.
This is not about a database customer management solution where
you can have someone that will do a single installation, and
you run your localize databases.
Cloud for us in EDA is really focusing in initially on burst
compute. You have this need, whether it be for physical
verification (DRC/LVS) or functional verification (Verilog),
people cannot build data centers big enough to give them
the turnaround time that they want to get.
And so having this ability to pull together as a smaller
company with maybe 100's of engineers, you can all of a
sudden pull together these 4,000 to 6,000 CPU clusters,
and pound them on your problem, whether it's functional
or physical.
Cooley: And you only need it for a week...
Sawicki: It's not even needed for a week. For functional, maybe
you need it for a week. For physical verification, you
only need it for 12 hours or so.
Everyone would ask us about Cloud: "What are you doing
about cloud?"
We'd ask them: "What do you want to do?"
And they would go, "Nothing ... But it's a cool word...
What are you doing about it?"
[audience laughter]
Now we have most of our customers are saying, "Well, look.
I want the capacity when I go to tapeout and I've got this
chip -- that I know I can get the capacity without having
to steal it from other projects and I'm going now to the
cloud."
So that's what that project was about. We've done DRC/LVS
validation on the cloud ... we did the press release
of TSMC, so we can get their data onto the cloud, their
proprietary design kits.
... I don't remember if we got any other press releases out
the door yet, but we've done validation with another major
customer who does large CPU type designs. It should be out
if it's not yet... Where a giant chip out there is getting
Calibre DRC verified, scaling out the 4,000 CPUs, all burst
mode running from the cloud.
[ indecipherable voice off microphone ]
It's out?
Okay, the press release is out.
Cooley: So what is that large company, Intel?
Sawicki: No, it's AMD.
Cooley: Okay.
Sawicki: We actually ran AMD's processor on AMD processors in the
Microsoft Azure cloud network. And so this aspect where
all of a sudden your IT infrastructure has a chance at
completely changing now, in terms of how people manage
these hybrid cloud environments all around dealing with
these incredibly difficult-to-design chips and the
compute capacity is needed. It's a cool thing.
Cooley: Cool.
---- ---- ---- ---- ---- ---- ----
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
|
|
