( DAC'13 Item 1 ) ----------------------------------------------- [09/27/13]
Subject: Cadence Tempus vs. Synopsys PrimeTime as #1 hot tool at DAC'13
ALL EYES ON TEMPUS: More folks wrote in about the new Cadence Tempus timing
sign-off analyzer this year than I've seen for any other tool in 19 years of
running this DAC Trip Report (since 1994!). Users very desperately want a
competitor to PrimeTime. Of course, they want a faster timing tool, but
many give the impression that just matching PrimeTime is enough.
"My company wants it to succeed. We really hope it works.
Anything to break the stranglehold Synopsys has on us."
A few engineers expressed cynical doubts over how much Tempus is "new" SW
versus it being repackaged old (but updated) Cadence ETS SW.
And almost everyone wants proof that Cadence will actually deliver on these
new public sign-off claims.
"What were the 3 or 4 most INTERESTING specific tools you
saw at DAC this year? WHY did they interest you?"
---- ---- ---- ---- ---- ---- ----
I haven't actually used Tempus. Below is what Cadence said at DAC:
- New sign-off STA engine developed by former Magma R&D folks.
Claims to be 10X faster than current versions of PrimeTime.
- Claims to do multi-threading, incremental timing (equivalent
to PrimeTime's HyperScale in functionality) and MMCM better
and faster than PT.
- 2-3X speed-up over PT in each of these areas of functionality.
- Supports distributed processing: netlist partitioned and farmed
out to multiple machines on LSF (typically under 8). Cadence
claims they get an additional 2-3X speed-up through DP.
- Supports SOCV (TSMC standard that is equivalent to POCV from
SNPS). Tempus already supports load and slew-dependent timing
derate tables.
- SI timing with 50 M instances in 8hrs (both multi-threading and
DP enabled). Memory footprint is smaller than non-DP (50 GB).
---- ---- ---- ---- ---- ---- ----
Our CAD department is extremely interested in Tempus being real.
---- ---- ---- ---- ---- ---- ----
Cadence's new Tempus timing analyzer seemed most interesting to me,
as we do multicore designs that stress EDA tools to their limits.
- The parallelism in Tempus and the ability to partition large
designs across multiple CPUs and servers is a step in the
right direction.
- Cadence says it performs significantly faster than Synopsys
PrimeTime. (We haven't validated the runtime.)
My general reaction is that after seeing Synopsys doing better than
Cadence in many areas, it's good to see Cadence doing better in some.
This leveling is important. Competition is good.
---- ---- ---- ---- ---- ---- ----
Tempus will cut long PrimeTime design cycles into more feasible
shorter cycles before tape-out. When cycles are too long, our
designers like to take risky shortcuts because of our company
TTM pressures.
Tempus appears to be an STA engine on steroids.
---- ---- ---- ---- ---- ---- ----
From the Tempus demo I saw, the advertised gain in speed (about 10X)
is impressive -- however it looks to be a combination of smaller
gains, in various places of the whole runtime Cadence flow.
Not sure how Tempus compares with PrimeTime, need real benchmarks to
assess it. It is also important to remember that overall efficiency
of STA activity relies also on your debug interface to rapidly spot
the root cause of a timing problem.
Tempus looks like an interesting step forward, but need more data to
really assess its value.
---- ---- ---- ---- ---- ---- ----
Cadence Tempus looks good. I appreciate any competition for Synopsys
PrimeTime and hope to have a chance to do an evaluation soon.
---- ---- ---- ---- ---- ---- ----
I hope Cadence Tempus isn't one of those typical Cadence vaporware
announcements. We really do need a working alternative to PrimeTime.
This lack of competition in timing sign-off is killing us.
---- ---- ---- ---- ---- ---- ----
The Tempus demo was mainly highlighting speed gains. The real
question is how Tempus compares with PrimeTime. For my company we
need to do real benchmarks to assess it on our design styles.
Keep in mind that for STA, speed is only one parameter of overall
efficiency -- usability, design and constraints analysis features,
and debug capability are also key.
It's not clear if Tempus is an evolution of existing ETS or a
complete new development.
---- ---- ---- ---- ---- ---- ----
My company is sick of the iron grip Synopsys has in timing.
We must undo this PrimeTime knot Aart has tied us up with.
---- ---- ---- ---- ---- ---- ----
I'm a current user of Cadence ETS timing analysis and am interested in
getting my hands on Tempus. Tempus looked impressive on paper at DAC
and I want see how it performs in real life situations of our daily
designs.
I want to test it's MMMC scenarios and validate the fast runtime claims.
---- ---- ---- ---- ---- ---- ----
The following is my impression purely based on Cadence's presentation.
1. Fast runtime
2. Good correlation between PNR tool and Tempus to reduce the
iterations before taping-out
3. Good capacity and concurrent run
---- ---- ---- ---- ---- ---- ----
Cadence Tempus Timing analyzer:
I was very impressed with the demo. Has a pretty good infrastructure
to deal with the run-time bottleneck and capacity for large SOC STA.
Would like to see more Tempus SPICE-level correlation numbers for
both delay noise numbers.
---- ---- ---- ---- ---- ---- ----
I am in purchasing and therefore tool feedback from my side is not
very helpful.
What I like is Cadence is starting real competition to PrimeTime.
---- ---- ---- ---- ---- ---- ----
I took away that if you have a huge design Tempus in the tool for you.
If you're doing small embedded systems, it probably has little to no
benefit vs. PrimeTime.
---- ---- ---- ---- ---- ---- ----
I found the Cadence presentation very interesting.
Tempus appears to be quite different from PrimeTime, so it's hard to
get a good sense of how it compares head-to-head without actually
using it.
---- ---- ---- ---- ---- ---- ----
Tempus claims of runtime improvement over PrimeTime look enticing
(but not validated), but due to the large effort of porting an existing
timing flow, a substantial financial incentive would need to be given
to us to even try it out.
If I were at a cash-strapped start-up, I'd consider such an option.
---- ---- ---- ---- ---- ---- ----
We will look at Cadence Tempus sometime early next year.
My conclusions from the marketing pitch were:
- Lots of big promises
- Comparison against PrimeTime leave much to be desired as old tool
PT version used as a performance reference
- New timing engine/kernel yet same results as ETS - something is
fishy!
- Tempus sounds like a new version of ETS with proper multi-threaded,
distributed implementation added in. It's ETS renamed and yes,
customers will have to buy it - again.
---- ---- ---- ---- ---- ---- ----
We attended the Tempus presentation at the Cadence booth. It was
short and didn't go into a lot of details. From what I saw, there
is not really anything that looks ground breaking to me.
It looks like most of the PT features are supported, which is good,
but I didn't see anything interesting beyond that.
Tempus showed some speed-up numbers over PrimeTime.
---- ---- ---- ---- ---- ---- ----
We can't tell what percent of Tempus is new code and what percent is
reused ETS code. We hope it's 90% to 10%. We fear it's 10% to 90%.
---- ---- ---- ---- ---- ---- ----
My impression from the Tempus presentation was that some efforts had
been made to make it faster by wrapping the existing ETS technology;
but the core tools themselves are pretty much the same.
I had a very bad experience a couple of years ago trying to use the
Cadence ETS distributed processing. It corrupted the results and
could have been embarrassing. Whilst many who look for advances in
STA are looking at run times as a main issue -- the designs me and
my team verify tend to be less huge. My STA requirements are instead
confidence in results, usability (especially ease of use), clear and
easy to interpret reports, and robustness.
But there is another important consideration. If I report timing
from Cadence ETS the majority of my partners will inevitably ask
"but how does that compare in PrimeTime?". Until the most popular
STA tool is Cadence ETS, I will still need to complete my work with
PrimeTime analysis.
In summary, even with a much better STA tool from Cadence I would be
slow to adopt it myself. I don't see a scenario where I could use
Tempus if I need to report the results outside my company.
---- ---- ---- ---- ---- ---- ----
a. Cadence Tempus: Very interesting to see how it handles these big
designs with no change to the accuracy with which Tempus handles
the designs. I was impressed with the accuracy, speed and capacity.
---- ---- ---- ---- ---- ---- ----
Cadence Tempus:
Interesting as has multi-CPU, multi-machine capability.
But need to see if they can deliver to the promise.
---- ---- ---- ---- ---- ---- ----
In my opinion the most interesting tool is timing analyzer Tempus from
Cadence. It's an ambitious project with bold claims; a brave attempt
to compete with PrimeTime.
If the tool is a success it will be extremely helpful for timing large
chips. Impressive (if true) is Tempus is uses multiple cores on
multiple computers to time flat non-hierarchical circuits. Claims
scalability and speed of path analysis.
Another interesting tool at DAC, as usual, was Synopsys PrimeTime. I
always like steady progress of speed and accuracy. And a good idea to
generate ECOs by the timer taking into account physical implementation
of the chip circuit.
---- ---- ---- ---- ---- ---- ----
Cadence Tempus >> b'cos of claims its fast & accurate.
Need to test it out!
---- ---- ---- ---- ---- ---- ----
Cadence Tempus is a new tool, and it seems to have been way over-hyped.
It appears to be vaporware at this point, and I could not see anything
they were suggesting that would make it competitive with PrimeTime.
Maybe in a couple of years they will prove me wrong.
It seemed several announcements from Cadence were basically admitting
defeat: Tempus to replace ETS; Spectre-XPS to replace Ultrasim; and
according to rumors, an upcoming replacement for VoltageStorm.
They are making some nice improvements to Virtuoso, at last.
---- ---- ---- ---- ---- ---- ----
Cadence Tempus
But who am I fooling? ETS tried and failed. Magma tried and failed.
Extreme-DA tried and failed. Aart always seems to win whenever it
comes to PrimeTime. I predict Lip-Bu will lose interest after a few
months and Tempus will be sidelined inside CDNS.
---- ---- ---- ---- ---- ---- ----
Cadence seems to be betting big on a comeback in the signoff space with
Tempus and ETS (or whatever the new name is going to be)
---- ---- ---- ---- ---- ---- ----
CDNS Tempus
My company wants it to succeed. We really hope it works. Anything to
break the stranglehold Synopsys has on us.
---- ---- ---- ---- ---- ---- ----
Related Articles
Jasper SEC/Datapath, Atrenta BugScope, Zocalo were #2 at DAC'13
Calypto PowerPro versus Atrenta Spyglass Power was #3 at DAC'13
Berkeley DA ACE, AFS Mega, & Solido variation were #4 at DAC'13
CLK-DA AOCV FX, Excellicon, Nangate, & Dorado were #5 at DAC'13
Join
Index
Next->Item
|
|
