( SNUG 05 Item 14 ) ---------------------------------------------- [12/20/05]
Subject: IC Compiler vs. Magma Cobra vs. Cadence Encounter GXL
POTEMKIN VILLAGE -- To create the illusion of peace and prosperity in the
two troubled regions he controlled, the newly appointed Russian governor
Field Marshal Potemkin built elaborate fake villages to impress Catherine
the Great during her tours of Crimea and the Ukraine.
As a 14 year Synopsys watcher, my first impressions of IC Compiler is that
it ain't no PhysOpt.
Here are the first 9 months of PhysOpt:
ESNUG 335 #1: Bob of Nvidia gives a hands-on review of PhysOpt
Dec 1999: The first PhysOpt/PKS/Magma/SPC user tape-out count
Jan 2000: The final PhysOpt/PKS/Magma/SPC user tape-out recount
ESNUG 338 #1: Jon of Avici gives a hands-on review of Chip Architect
ESNUG 339 #2: Wall St. questions Jon of Avici about PhysOpt & ChipArch
ESNUG 340 #1: Mismatched timing engines stall Cadence PKS
ESNUG 341 #1: Nick of Ford Microelectronics on his PhysOpt story
ESNUG 342 #1: Jay of Agilent finds 3% timing correlation in Cadence PKS
Gadfly (Feb): "PhysOpt/PKS/Magma tea leaves" column
ESNUG 343 #2: Replies to user PhysOpt/PKS/Magma tea leaves column
ESNUG 344 #2: Anthony of Texas Instruments on his PhysOpt benchmark
ESNUG 344 #5: David of Matrox PhysOpt-with-Cadence-backend tapeout
ESNUG 344 #11: Wally of Mentor accidentally "outs" his TeraPlace tool
ESNUG 345 #1: Roger of Cray/SGI on his PhysOpt-with-IBM experiments
ESNUG 345 #4: Dave of Agilent on his PhysOpt vs. Qplace benchmark
ESNUG 346 #1: Anon user says Magma Sorta Worked; PhysOpt Not
ESNUG 346 #5: Bob of Nvidea on PhysOpt, Gambit, Scheme, PDEF
ESNUG 348 #3: Oops -- anon user says PhysOpt wasn't broke, PKS was!
ESNUG 349 #10: Anand of EmpowerTel details his PKS tape-out
SNUG 00 #18: More PhysOpt/PKS/Magma user tape-outs to add to the count
ESNUG 350 #2: A PhyOpt obstructions script from Synopsys Support
ESNUG 351: EE Times confirms TI PhysOpt tapout scooped in SNUG Report
ESNUG 351 #1: Jay of Agilent benchmarks PKS vs. PKS II
ESNUG 352 #5: Nancy of Sun Microsystems on hierarchical layout issues
ESNUG 354 #3: Overriding PhysOpt Steiner routing from Synopsys Support
ESNUG 354 #7: Errors on LEF2PLIB conversion for PhysOpt from Synopsys Support
ESNUG 354 #12: Anon users on signal integrity, Cadence SE-SI, PhysOpt
Notice it was the *hands-on* users and Synopsys technical support who drove
the warts-and-all PhysOpt story. It was an organic, grassroots discussion
about bugs, rival tools, and what works and what didn't work. It was lots
of messy details that takes lots of time to explain. PhysOpt traction was
ground level engineers pushing their bosses with "Hey, we gotta get PhysOpt
in house." Synopsys Marketing wasn't in the discussion.
And this happened in the iffy days when Synopsys was a green newbie player
with no credibility in the P&R market!
That was 5 years ago. Now let's look at the first 9 months of IC Compiler:
ESNUG 442 #2: Clay of Agere shares his initial IC Compiler benchmark.
ESNUG 442 #3: Magma has 8 questions about the IC Compiler announcement.
ESNUG 445 #1: Users seek dirt on Magma Cobra and Synopsys IC Compiler.
DAC 2005: Synopsys Marketing had 5 users present carefully prepared
Powerpoint IC Compiler "success stories" at a DAC lunch.
Nov. 2005: Synopsys Marketing puts out a press release with a vague
2 sentence quote from an Agere Director (who manages some
engineers who run IC Compiler) saying they have a tapeout.
Other than that one 9 month old benchmark by Clay of Agere in ESNUG 442 #2,
notice the lack of traction IC Compiler has. This isn't the discussion of
hands-on users. It's not messy details that take time to explain. It's not
grassroots. It's forced and stiff and soundbites and Synopsys Marketing.
Unlike PhysOpt, the first 9 months of IC Compiler has had all the genuine
user enthusiasm of a Preparation H television commercial.
And this is 5 years later where Synopsys is an established player in P&R.
I know Aart's marketing droids are trying to generate the public impression
that IC Compiler is selling like gangbusters, but as an outside observer
looking in, I'm simply not seeing it.
(And to be fair, Cadence Encounter GXL appears to only be a revision upgrade
of FE or possibly one of those Mike Fister closet price raise repackagings.
Magma Cobra is a massive series of changes throughout the Blast and Quartz
family of tools. So far neither GXL nor Cobra have had a single hands-on
user tapeout to vouch for them either.)
The other problem is IC Compiler sales are based on a questionable assuption.
A few months ago, Aart gave the keynote at Boston SNUG and put up a slide
showing a DC to IC Compiler only flow. During the post-speech Q&A, I asked
what happened to Astro and PhysOpt. "Why no mention of them?" He replied
that they were still his "workhorse" P&R tools and that they'd probably be
around for the next 3 years during the phase-out to IC Compiler.
During the post keynote reception where everyone was milling around in the
hallway, two engineers came up to me. The first one thanked me for asking
about Astro because he uses the tool and had the exact same question I did.
Digging through this survey's respondants, I found:
- "We are just now migrating to Astro 2004.12. We'll keep using old
due to budget issues."
- "We switched to Synopsys 3 months ago. Will keep using Astro."
- "We have used Apollo and are moving to Astro in this period."
- "Keep the old PhysOpt/Astro. We will look at the new IC Compiler but
won't consider it for production until it is well proven."
- "We are running huge designs (10 MG+) flat in Astro."
- "Keep using old PhysOpt/Astro due to budget issues. Technicaly
I believe IC Compiler is the way to go."
- "Our current subcontractor uses Astro, and will stay with it"
- "Astro's good for us. We route a chip, with 900 K placeable objects,
flat in Astro. We couldn't do that previously. We also experienced
faster run time in new versions. No need to change."
Looking at the stats, of the Synopsys P&R customers who are not switching
away, 68% of them plan to keep on using good olde PhysOpt and/or Astro.
What's questionable is assuming Aart's P&R customers will, in 3 years, all
want IC Compiler's pricey functionality. Aart's so confident that 2006.06
is the final planned rev of Astro. Here's last year's (2004) designs:
250+ nm # 3%
250 nm ## 4%
180 nm ####### 15%
150 nm ### 6%
130 nm ########################## 52%
90 nm ########## 21%
65 nm # 3%
Here's this year's (2005) designs:
250+ nm # 1%
250 nm # 1%
180 nm ##### 11%
150 nm ## 5%
130 nm ###################### 44%
90 nm ################# 34%
65 nm ## 3%
45 nm # 1%
Do you honestly think the bulk of designs is going to be 65/45 nm in 3 years?
Astro works well at 130 nm and 90 nm for most designs. Many of its existing
users won't like being forced to upgrade for what they'll see as a Synopsys
money grab. Why should I pay an extra $300 for a Tivo box plus $13 a month
for the Tivo service when I own a VCR and my current cable TV service already
offers free on-demand movies?
Oh, and about that 2nd engineer in the hallway reception after Aart's Boston
SNUG speech? He wanted me to hook him up with a current Magma user. His
company was thinking of switching to Magma and he wanted to know the gotchas
before going in.
---- ---- ---- ---- ---- ---- ----
Not everyone is free to switch to the latest at all times. Keep
using PhysOpt/Astro.
- [ An Anon Engineer ]
We won't totally switch to any of these. We will surely mix and match
Cadence and Synopsys tools to get the best QOR for each step in the flow.
- [ An Anon Engineer ]
PhysOpt is not 2x faster for us, we don't push capacity limits.
- [ An Anon Engineer ]
Was using PhysOpt/SE, now using PhysOpt/Astro and will continue this
year. Astro is heaps better than SE 5.4, way faster and actually
routes my designs when SE gives up. It is very nice that PhysOpt
doesn't crash so much anymore.
- [ An Anon Engineer ]
We actually were more of a PhysOpt/FE flow. We did see significant speed
up for PhysOpt but mostly due to running on Linux boxes.
- [ An Anon Engineer ]
We haven't used Astro. Our speedup with PhysOpt seems to come mostly
from faster computers. Keep using PhysOpt.
- [ An Anon Engineer ]
PhysOpt/Astro performance has improved. Haven't quantified it. Capacity
not an issue for our designs. X keep using old PhysOpt/Astro.
- [ An Anon Engineer ]
Keep using old PhysOpt/Astro for larger nm designs, but upgrade to new
IC Compiler for designs on the smallest dimension processes.
- [ An Anon Engineer ]
For most designs, there is a reasonable improvement for Astro.
- [ An Anon Engineer ]
(Key word being production flow) X keep using old PhysOpt/Astro.
- Lauren Carlson of StarGen Inc.
We currently use both PhysOpt/Astro and Magma, and I suspect that will
continue.
- David Smith of STmicroelectronics
We were skeptical, but indeed we have seen a 2x capacity improvement
(this week). At first glance, PhysOpt and Astro appear to be on par
with Magma.
- [ An Anon Engineer ]
Keep using old PhysOpt/Astro.
I would say in general PhysOpt has gotten better with each major release
over the past few years. We re-use many blocks from one design to the
next, PhysOpt appears to be able to do larger design blocks in less
time and with less memory than previous releases. I know that blocks
that took half a work-day before now can be done 3 or 4 times in a
work-day, and we didn't upgrade our computers.
- [ An Anon Engineer ]
Today I use PhysOpt only and I'll try to keep it.
- [ An Anon Engineer ]
We have seen no capacity issues with Astro but the tool is broken in so
many other ways though. We are really struggling to get our designs
through. Example: distributed routing crashes virtually every time
regardless of OS in an LSF environment. SNPS does NOT test their dist
routing. We have had issues from the start. Over 2 years.
- [ An Anon Engineer ]
I'd like to see some data on design sizes and related complexity (number
of clocks, IP, etc.) All indications I have is that PhysOpt is a bitter,
overpriced, disappointment. This is where Synopsys is getting it's
clocked cleaned in the industry by Synplicity and Magma.
- [ An Anon Engineer ]
I haven't been impressed with the 10 minutes required to open my unplaced,
floorplaned, Milkyway-based netlist. PhysOpt 2004.12 used to open the db
in a couple of minutes. We'll probably be using Physopt/Astro and for
some blocks, kick IC Compiler's tires (or more if we're satisfied).
- [ An Anon Engineer ]
How buggy are Astro / PhysOpt (patch release every month or 2)?
Aart and others too conned thousands of "designers" into zillions of
man-months debugging.
Astro <-> PhysOpt timing correlation is a shambles, expensive in time
and $$ when our customer says "here's 1 GHz in PC, over to you."
(PhysOpt Primetime Astro delay calculators are all different).
Astro <-> PhysOpt databases are a shambles: .db .ddc Milkyway, text
.def .pdef buggy.
Synopsys hasn't integrated Avanti after 3 years, IMHO never will.
Upgrade to new IC Compiler?
From the people who brought you the Astro/PhysOpt shambles?
- [ An Anon Engineer ]
Our foundry uses Astro. If there was an opportunity to purchase a PD
tool I would opt for Magma.
- [ An Anon Engineer ]
I don't know about PhysOpt now, but I knew PhysOpt 2 years ago. PhysOpt
was so messed up, that if Synopsys R&D spent 2 hours cleanning up their
code, I easily believe they could make it 10x faster...
- [ An Anon Engineer ]
Upgrade to new IC Compiler. We have a few seats already.
- [ An Anon Engineer ]
Astro/PhysOpt -- 2X faster, 2X capacity, 2X the bugs
upgrade to IC Compiler
- [ An Anon Engineer ]
upgrade to new IC Compiler. I heard PhysOpt runs out of steam at 90 nm.
- [ An Anon Engineer ]
Upgrade to IC Compiler. Keep Blast Fusion.
- [ An Anon Engineer ]
Upgrade to IC Compiler, and consider Magma.
- [ An Anon Engineer ]
Using an older qualified version of PhysOpt now. We will probably
upgrade to IC Compiler and maybe even use Magma and/or Cadence for
some 90 nm designs and below
- [ An Anon Engineer ]
I didn't use Astro up to now. Only PhysOpt. My kind of designs are
too small to see significant improvement. I think we will stay with
Synopsys tools so IC Compiler will be used. I also think that we will
start to use it in the next 12 months.
- [ An Anon Engineer ]
The Astro run times do seem to be more less however, they are less
compared to the 2003 suite of tools not 2004 suite. I always have
issues with upgrading to newer versions regarding timing correlation
with older versions as well as script modifications which must be
performed.
We will upgrade once IC Compiler supports multi-voltage flow.
- [ An Anon Engineer ]
We have seen Astro improve QOR as compared to last year. We use pure
Astro flow, so we would be looking into IC Compiler as the next step.
We don't use PhysOpt.
- Sunil Malkani of Broadcom
Evaluate how well is IC Compiler received.
- [ An Anon Engineer ]
Upgrade to new Cobra
- Frank Li of Thomson Inc.
We also use Cadence in our flow, considering switching those seats to
Synopsys or Magma.
- [ An Anon Engineer ]
We've switched to Magma since early this year.
- [ An Anon Engineer ]
For the moment we'll keep using old Astro.
But we are seriously looking at switch to Magma.
- [ An Anon Engineer ]
I will upgrade to the new Corbra version.
- John Schritz of Tektronix
Most likely upgrade to new Cobra, unless test metrics say it is broken.
- [ An Anon Engineer ]
current rev PhysOpt better, not quite 2X. Keep using Blast Fusion.
- [ An Anon Engineer ]
Although Astro has gained capacity it still hasn't caught up to
newer tools. Keep using Blast Fusion. We have done designs with
both SoC Encounter and PhysOpt, too.
- [ An Anon Engineer ]
Staying with Magma, not yet sure on Cobra
- [ An Anon Engineer ]
Currently using Magma Blast Fusion. For next project, my company will go
with whoever gives us the best coorporate deal, regardless of the
abilities of the tool. Currently in negotiation. Depends on outcome of
price negotiations.
- [ An Anon Engineer ]
We will most likely begin using 5.0 Cobra.
- [ An Anon Engineer ]
Astro & PhysOpt still needs to improve. There is a 4th option. Add Magma.
- [ An Anon Engineer ]
Will upgrade (primary source) - upgrade to new Cobra.
Will keep (2nd source) - keep using old PhysOpt/Astro.
- [ An Anon Engineer ]
I think we will switch to Synopsys for more designs from Blast Fusion.
- [ An Anon Engineer ]
We must support backend flows that our customers require. Our customers
include several of the top 10 semiconductor companies on the planet. To
date, we have been requested to support both Cadence and Magma back-end
flows, and as such are partners with both companies.
- [ An Anon Engineer ]
keep using old Blast Fusion. upgrade to new Encounter GXL.
- [ An Anon Engineer ]
Upgrade to new IC Compiler. Keep old SoC Encounter.
- [ An Anon Engineer ]
keep using SoC Encounter, it's what we have.
- [ An Anon Engineer ]
still use old Encounter or upgrade to Encounter GXL
- [ An Anon Engineer ]
C. keep using old SoC Encounter
- Tom Mannos of Sandia National Laboratories
My choice: upgrade to new Encounter GXL.
- [ An Anon Engineer ]
Keep using Astro. Keep using old SOCE.
- [ An Anon Engineer ]
We have been benchmarking PhysOpt/Astro over the last year and noticed
that the run times improved dramatically. I didn't put enough attention
into speed metric (it was of lower importance than area to us.) I don't
know if it was the result of a run-time setup change or a tool version
change (sorry).
We will keep using SoC Encounter, for a number of significant reasons.
- [ An Anon Engineer ]
I can't speak to Astro but the 2004 PhysOpt releases definitely were
a major improvement over previous versions. Physopt will remain here
for the next 12 months. We don't use Astro.
We will definitely continue to use SoC Encounter here and plan to
keep updated with its latest releases.
- Russell Petersen of Scientific Atlanta
Keep using old Blast Fusion. Not based on tool use, just cost-of-change.
Upgrade to new Encounter GXL. Encounter was not so great.
- [ An Anon Engineer ]
Most likely keep using Encounter, but there is a slim chance we may
switch to Astro.
- Are Arseth of Atmel
We are using SoC Encounter and will continue to use it in the future.
- [ An Anon Engineer ]
C: Upgrade to new Encounter GXL
- Brendan Barry of S3 Ltd.
Can not judge, but rather do the new Encounter GXL
- [ An Anon Engineer ]
Index
Next->Item
|
|
