( ESNUG 477 Item 2 ) -------------------------------------------- [11/20/08]
Subject: 7 users benchmark PrimeTime 2007.12 versus PrimeTime 2007.06
> Synopsys PrimeTime also does STA, SSTA, noise, crosstalk and power. This
> DAC's claim is PT is "up to 4X faster and multi-core enabled". (booth
> 1349) Ask for Ken Rousseau. Freebie: Starbucks card
>
> - from http://www.deepchip.com/gadfly/gad060608.html
From: [ The Mouse That Roared ]
Hi John
Our PrimeTime 2007.12-SP3 (STA only) run time dropped from 15+ hours to
10 hours offering a 30% reduction vs. the 2007.06-SP3 release. Memory was
also down about 30% on this design which was over 30 M instances.
Users should really check out the latest release.
Please keep me anon, thanks.
- [ The Mouse That Roared ]
---- ---- ---- ---- ---- ---- ----
From: Edwin Yew <edwin.yew=user domain=3dlabs not calm>
Hi John,
FYI, this is a small test case I have done to compare the runtime and
quality between PT2007.06 and PT2007.12-SP1 a while back.
PE Block PT Performance Evaluation - Delay Calculation
PT 2007.06 PT 2007.12 SP1
Peak Memory 121 MB 102 MB
Run Time
Read Parasitics 24 sec 17 sec
Timing Step 87 sec 37 sec
No. of RC 004 Warnings 12888 12888
No. of RC 009 Warnings 12 12
No. of Cell Delays Arcs 801387 801387
Cell Delay Comparison:
0 pS Difference: 801371
1 pS Difference: 11
2 pS Difference: 6
No. of Net Delays Arcs 182908 182908
Net Delay Comparison:
0 pS difference: 182908
PE Block PT Performance Evaluation - STA with SI
PT 2007.06 PT 2007.12 SP1
Peak Memory 201 MB 146 MB
Run Time
Read Parasitics 34 sec 26 sec
Update Timing 507 sec 179 sec
Reports 8 sec 7 sec
Save Session 15 sec 13 sec
It's a very small block (~38 K cells) but I reckon the improvement percents
should be linearly reflected in larger blocks.
- Edwin Yew
3D Labs, Inc. Egham, Surrey, UK
---- ---- ---- ---- ---- ---- ----
From: Nidhi Varshneya <nidhi.Varshneya=user domain=csr not calm>
Hello John,
My feedback regarding PrimeTime 2007.12.
- Runtime decreased from 1:25 to around 40 minutes for a single run
which is quite significant.
- With "set timing_crpr_enable_adaptive_engine true" the time taken for
update_timing improved when SI was on and the CPU memory dropped by 20%
(However it required us to use "set_clock_sense" for the clocks to
control only positive unit arcs, which earlier was controlled by case
analysis. It appeared PT wasn't supporting case analysis for the clock
propagation. So no more case analysis required "set_clock_sense" solves
the problem.
- "check_noise" helped in spotting quite a few problems in our CCS noise
models and is very helpful indeed.
Indeed 2007.12 versions are better than their older versions.
- Nidhi Varshneya
CSR Cambridge, UK
---- ---- ---- ---- ---- ---- ----
From: [ The Man In The Iron Mask ]
Hi John,
Please keep me anonymous.
We just taped-out using PrimeTime 2007.12 and it was really fast. We saw
an improvement of 2.5X (from 5 hours to 2 hours) on our design that is
about 18 million gates.
Let me know if you have any further questions in this regard.
- [ The Man In The Iron Mask ]
---- ---- ---- ---- ---- ---- ----
From: Jacob Xiao <jxiao=user domain=qualcomm not calm>
Hi, John,
This version is at least 2 to 3 times faster than version 2007.06. It has
some infrastructure improvements so that performance is much improved. The
runtime improvements apply in PT-SI as well as PT-PX with the accuracy
remaining the same.
I've observed the improved runtime using the same script as in previous PT
releases.
It's "report_timing -exceptions" can show the constraint source file and the
line number when exceptions exist on the reported timing path.
- Jacob Xiao
Qualcomm, Inc. San Diego, CA
---- ---- ---- ---- ---- ---- ----
From: [ Life is Good ]
Hi, John,
My design is not very big as about 2 million gates. When I run it with the
version of 2007.12, it takes 1.7X faster than 2006.12-SP3 version. Maybe I
can guess that it will be about 2X faster than the old PT versions if the
design is a lot bigger.
Please keep my name anonymous.
- [ Life is Good ]
---- ---- ---- ---- ---- ---- ----
From: [ The Horse With No Name ]
Hi John,
Please keep me anonymous.
We recently switched to PrimeTime 2007.12 for our static timing analysis.
I have found on a design of 1 M gates that my STA run time decreased 1/2
from 22 mins to 12 mins over using the 2007.06 release.
With the 2007.12 release we are also consistently seeing drops in memory
usage of 30%.
If just considering the time it took to load netlist/SPEF/SDF without
generating any output reports, runtime was actually reduced from 12 mins
to less than 4 mins and memory saving was 35% from 2007.12-SP3 compared
to 2007.06-SP3.
- [ The Horse With No Name ]
Index
Next->Item
|
|
