best computer for large spice sims
Paul Spitalny
It's time for me to contemplate a new, better, faster computer. I run a
lot of spice simulations with fast clock rates, and slow time constants.
Anyone who has done any switching power supply work or sigma delta work
will know what I am talking about!
Anyhow, there are several factors in my choice
1) processor - what type, what speed (I am thinking Pentium 4, 1.8GHz)
2) Ram - I am thinking of 768MEg of Rambus DRAM
3) Disk - 7200rpm ultra ATA drive
What do you think is the biggest speed factor? It has been my belief
that having a lot of RAM is good because it prevents the need to read
and write from disk during the simulation run. I don't think SCSI will
give me a good read/write speed increase for the money since the ultra
ata drives are pretty fast these days. One question I have is whether
Rambus DRAM makes sense, or should I go for 133MHZ DDR Ram (effectively
this would be 266MHz).
Well, what do you think?
--
To respond to this posting, remove -nospam- from my email address.
Sorry for the inconvenience
Jim Thompson
Mark Chun (Reson) and I have run a series of PSpice benchmarks. The
net result is that P4's SUCK! Following is the recommendation from
Mark for fastest/best system...
Mother board: Asus A7A266. Cost $130 at the computer show, $146 at
www.memman.com. Sound on-board is available. Reviews on this board are
very good.
CPU: AMD Athlon 1.2GHz w/ 266 FSB. The Athlon is available at 1.4GHz.
Be sure to get the 266 MHz front side bus (FSB), retail (boxed)
version. The boxed version has a 3 year warrantee and comes with
heatsink and fan, the OEM version has a 60 day warrantee. Retail
version is only $15 more. Athlon's run HOT.
Memory: DDR (PC2100). Micron is all I see out there right now. For
your system, you should have a minimum of 512MB of RAM. It's pretty
cheap, $44 per 256MB module from www.memman.com ($36 at the computer
show). The A7A266 mother board only has 2 RAM sockets.
Hard drive: We have had good luck with IBM and Maxtor. Avoid Western
Digital. IBM's 60GXP <http://www.storage.ibm.com/hdd/desk/ds60gxp.htm>
is pretty good. 40GB drives go for $120, 60GB for $170 at
www.memman.com.
Heatsink: Athlon may benefit from a better heatsink. I'm going to try
an ORB heatsink, made by Thermaltake. Presently, the silicon
temperature is running at 55 to 57 deg C.
Video board: If you want dual screen, best I have found is the Matrox
G450. Sucks at 3D gaming, but is very good with 2D graphics. It's the
only dual monitor solution that works well. Matrox doesn't support
VESA mode graphics in Win2k. VESA mode used in old DOS programs with
graphics.
Monitor: I am very pleased with the Samsung 950P. $250 at www.buy.com.
(If replying by E-mail please observe obscure method of anti-spam.)
...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona Voice:(480)460-2350 | |
| Jim-T@analog_innovations.com Fax:(480)460-2142 | Brass Rat |
| http://www.analog-innovations.com | 1962 |
For proper E-mail replies SWAP "-" and "_".
Darryl Okahata
> CPU: AMD Athlon 1.2GHz w/ 266 FSB. The Athlon is available at 1.4GHz.
> Be sure to get the 266 MHz front side bus (FSB), retail (boxed)
> version. The boxed version has a 3 year warrantee and comes with
> heatsink and fan, the OEM version has a 60 day warrantee. Retail
> version is only $15 more.
If you go the Athlon route, make sure that you're using an
AMD-approved power supply (AND one that's approved for the *particular*
speed processor that you're using). Often, the #1 and #2 problems that
people have with Athlon-based systems are the power supply and memory
(however, Athlon motherboards can have particular quirks, so be sure to
be aware of any). In particular, don't even think about reusing an old
power supply to use with the Athlon; while there's a chance that it may
work, there's a bigger chance that using it will only result in nasty,
intermittent crashes and reboots. Non-AMD-approved power supplies may
work, but, *if* you have any problems, they'll be harder to fix because
of the unknown power supply (unless you happen to like swapping out
power supplies). Also, make sure that you're using high-quality memory;
if you scan the various motherboard newsgroups, you'll find that a
number of people solved their intermittent crash/reboot problems by
replacing their RAM.
> Athlon's run HOT.
Yup. Make sure that you have good cooling, all around.
--
Darryl Okahata
dar...@soco.agilent.com
DISCLAIMER: this message is the author's personal opinion and does not
constitute the support, opinion, or policy of Agilent Technologies, or
of the little green men that have been following him all day.
qrk
<pauls_...@pauls.seanet.com> wrote:
>Hi Folks,
>It's time for me to contemplate a new, better, faster computer. I run a
>lot of spice simulations with fast clock rates, and slow time constants.
>Anyone who has done any switching power supply work or sigma delta work
>will know what I am talking about!
>Anyhow, there are several factors in my choice
>
>1) processor - what type, what speed (I am thinking Pentium 4, 1.8GHz)
>2) Ram - I am thinking of 768MEg of Rambus DRAM
>3) Disk - 7200rpm ultra ATA drive
>
>What do you think is the biggest speed factor? It has been my belief
>that having a lot of RAM is good because it prevents the need to read
>and write from disk during the simulation run. I don't think SCSI will
>give me a good read/write speed increase for the money since the ultra
>ata drives are pretty fast these days. One question I have is whether
>Rambus DRAM makes sense, or should I go for 133MHZ DDR Ram (effectively
>this would be 266MHz).
>
Pentium 4 is rather disappointing if your using programs which do not
use the spiffy P4 specific instruction set. As Jim Thompson said, we
have benchmarked a few machines with the AMD Athlon Thunderbird coming
in first. If you can hold out, AMD might be releasing the Palomino
processor soon (mid Oct by one account), but don't hold your breath.
The Palomino includes the full SSE instruction set which is very
helpful if you use programs that take advantage of the SSE instruction
set (Photoshop and some 3D rendering programs). If your programs use
SSE, then you might consider using the Intel processors (P3 or P4) or
wait for the Palomino.
On tests I have done with a few engineering programs (simulation and
CAD), memory speed was not an issue. I have tried using PC133 or DDR
(PC2100) on the same mother board (Asus A7A266 can use DDR or PC133)
with no difference in job time. On programs doing image processing,
memory speed is very apparent. Memory is cheap. If your using
Windoze2k or NT, go with 512MB minimum.
If you have access to other machines, you should benchmark the
programs you use to see what makes sense. Contacting others in this
news group is also helpful if you can supply a job file to benchmark.
Hard drive, ATA66 or 100 is good. On modern motherboards I'm seeing 80
MB/sec burst speed on ATA100 drives and 66 MB/sec on ATA66. Hard
drive sustained transfer rate is important especially if your using
programs that write huge data files. Once you get over the 20 MB/sec
rate, you should be OK. Any modern 7200 rpm drive will do 35 to 40
MB/sec sustained transfer rates on the outer tracks.
================
Mark Chun
Santa Barbara, CA
Darryl Okahata
> On tests I have done with a few engineering programs (simulation and
> CAD), memory speed was not an issue. I have tried using PC133 or DDR
On the topic in question (large spice sims), faster RAM will result
in faster simulations, although perhaps only by a few percent. Whether
the faster RAM is worth the extra cost, is a decision left up to each
individual buyer.
> Memory is cheap. If your using
> Windoze2k or NT, go with 512MB minimum.
Yup. At today's prices, there's no reason to get any less, as long
as your motherboard supports it (and most should).
Paul Spitalny
Thanks for all the help guys!
Paul
Paul Spitalny
Hi Jim,
Could you quantify a little bit your statement that P-4's "suck." I see some
1.8GHZ P-4's available but the Athalons only go up to 1.3GHz or so. Are you
saying that these will be faster at running SPICE. How much faster than the
P-4's ??
JD
"Paul Spitalny" <pauls_...@pauls.seanet.com> wrote in message
news:3BC1C502...@pauls.seanet.com...
>
> Hi Jim,
> Could you quantify a little bit your statement that P-4's "suck." I see some
> 1.8GHZ P-4's available but the Athalons only go up to 1.3GHz or so. Are you
> saying that these will be faster at running SPICE. How much faster than the
> P-4's ??
>
code is optimized for P4s, the total performance isn't all that bad, but few
compilers do a good job of optimizing for the P4.
For software developers who are optimizing for P4s, they are okay (IMO.) I'd
wait until higher performance versions of P4s become available as an end
user. I don't believe that they 'suck', but they aren't as good as other processors
when running code not specially built for them. Most code isnt' optimized
for P4's, and they do especially poorly (relatively-so) with the stack-based
FP codes.
John
Jim Thompson
<pauls_...@pauls.seanet.com> wrote:
Paul, Here is a table of the comparisons made on various systems....
PSpice Sim Sim time
Computer (seconds) at 1 GHz Mflop MHz/Mflop
============ ========== ======== ====== =========
P2, 266 MHz 1800 479 107.7 2.47 (Mark/Win95)
P2, 440 MHz 1082 476 163.7 2.68 (Jim/NT4.0)
P3, 800 MHz 510 408 347.2 2.30 (Jim/Win2K)
P3, 850 MHz n/a n/a 348.0 2.44 (Kirk/Win95)
P3, 1000 MHz 420 420 411.0 2.43 (Mark/Win2k)
P4, 1500 MHz 413 620 413.0 2.70 (Roy/Win2k)
ATH 1200 MHz 270 324 495.0 2.42 (Malcolm/WinME)
ATH 1200 MHz 244 293 500.0 2.40 (Mark/Win2k)
All simulations were from the same netlist.
S Ross
> Darryl Okahata wrote:
>
> > ma...@NO.reson.SPAM.com (qrk) writes:
> >
> > > On tests I have done with a few engineering programs (simulation and
> > > CAD), memory speed was not an issue. I have tried using PC133 or DDR
> >
> > On the topic in question (large spice sims), faster RAM will result
> > in faster simulations, although perhaps only by a few percent. Whether
> > the faster RAM is worth the extra cost, is a decision left up to each
> > individual buyer.
> >
> > > Memory is cheap. If your using
> > > Windoze2k or NT, go with 512MB minimum.
> >
> > Yup. At today's prices, there's no reason to get any less, as long
> > as your motherboard supports it (and most should).
I'm using a very similar machine at the moment, an Athlon 1.1GHz and even
using this old cpu I can run it at 1.3 without any trouble. The newer ones
can often run at 1.6GHz without crashing.
I would recommend a decent heatsink and not a Thermaltake orb, I have a
superorb sitting here and it is only really suitable for the lower end AMD
cpu's like the Duron. At the moment I am using a Taisol 742 and my
temperature never goes above 46 C and then only hits this if it has been
running a sim for a while or the 3d card is also being used heavily.
I might be tempted to wait a little for the new Athlons and some of the
motherboards that are starting to appear and use 166MHz DDR memory. With
the price of DDR now there is no reason to go with a board that only uses
PC133 mem. The possibilty of running dual Athlons is also inviting.
As far as a power supply you won't pay much more for a 300W + AMD approved
one.
Paul Spitalny
Hi Jim,
I found this data to be astounding. The 1GHz P3 is only slightly slower than the
1.5GHz P4!. Considering the price differences the P3 the better bargain in the
Intel family. Currently you can get 1.8GHz P4's, but what I might do is wait for
the next AMD processor and go with DDR since the Rambus memory (needed for P4's)
is more expensive than the 266MHz DDR.
Thanks for all the info Jim (and everyone else too).
Jim Thompson
<pauls_...@pauls.seanet.com> wrote:
|Jim Thompson wrote:
|
[snip]
|> |Hi Jim,
|> |Could you quantify a little bit your statement that P-4's "suck." I see some
|> |1.8GHZ P-4's available but the Athalons only go up to 1.3GHz or so. Are you
|> |saying that these will be faster at running SPICE. How much faster than the
|> |P-4's ??
|>
|> Paul, Here is a table of the comparisons made on various systems....
|>
|> PSpice Sim Sim time
|> Computer (seconds) at 1 GHz Mflop MHz/Mflop
|> ============ ========== ======== ====== =========
|> P2, 266 MHz 1800 479 107.7 2.47 (Mark/Win95)
|> P2, 440 MHz 1082 476 163.7 2.68 (Jim/NT4.0)
|> P3, 800 MHz 510 408 347.2 2.30 (Jim/Win2K)
|> P3, 850 MHz n/a n/a 348.0 2.44 (Kirk/Win95)
|> P3, 1000 MHz 420 420 411.0 2.43 (Mark/Win2k)
|> P4, 1500 MHz 413 620 413.0 2.70 (Roy/Win2k)
|> ATH 1200 MHz 270 324 495.0 2.42 (Malcolm/WinME)
|> ATH 1200 MHz 244 293 500.0 2.40 (Mark/Win2k)
|>
|> All simulations were from the same netlist.
|>
|> (If replying by E-mail please observe obscure method of anti-spam.)
|>
|> ...Jim Thompson
[snip]
|
|Hi Jim,
|I found this data to be astounding. The 1GHz P3 is only slightly slower than the
|1.5GHz P4!. Considering the price differences the P3 the better bargain in the
|Intel family. Currently you can get 1.8GHz P4's, but what I might do is wait for
|the next AMD processor and go with DDR since the Rambus memory (needed for P4's)
|is more expensive than the 266MHz DDR.
|
|Thanks for all the info Jim (and everyone else too).
Here's the latest info from Mark Chun:
Athlon 1.4GHz test added. I removed the MFLOP because the compiler did
a poor job on code efficiency.
PSpice Sim Sim time
Computer (seconds) at 1 GHz
============ ========== ========
P2, 266 MHz 1800 479 (Mark/Win95)
P2, 440 MHz 1082 476 (Jim/NT4.0)
P3, 800 MHz 510 408 (Jim/Win2K)
P3, 850 MHz n/a n/a (Kirk/Win95)
P3, 1000 MHz 420 420 (Mark/Win2k)
P4, 1500 MHz 413 620 (Roy/Win2k)
ATH 1200 MHz 270 324 (Malcolm/WinME old system)
ATH 1200 MHz 244 293 (Mark/Win2k PC2100)
ATH 1200 MHz 244 293 (Mark/Win2k PC133)
ATH 1400 MHz 210 294 (Bud/Win2k PC2100)
Mark also says there's a new AMD processor soon to be released.
"Palomino" I think he said, but, sorry, I didn't write it down.
Paul Spitalny
Hi Jim,
When Roy ran the 1.5GHZ P4, did he use rambus RAM? And, were the 1.2GHz Athlons
running 100MHZ SDRAM, 133MHZ SDRAM or DDR?
Also, I have heard reports of the Athlon t-bird chips locking-up periodically. Have
you experienced any behavior like this?
Darryl Okahata
> Also, I have heard reports of the Athlon t-bird chips locking-up
> periodically. Have
> you experienced any behavior like this?
Lockups are often caused by:
* Using a non-AMD-approved power supply (one not approved for the
*particular* AMD processor that you're using).
* Using cheap/bad RAM (from scanning the newsgoups, I find it amazing
how many people fix their "stability problems" by replacing their RAM
with known, high-quality RAM).
* Bad cooling. The newer/faster processors run hotter than older
processors, and need decent cooling (this is true of both Intel and
AMD, although AMD chips tend to run a bit hotter).
* Using buggy motherboards with certain VIA chipsets (generally causes
IDE data corruption, which can lead to a lockup). There are a few
older motherboards that have problems, and, as long as you avoid them
(or, sometimes, update the BIOS), everything's fine. For more info,
do a google groups search on "via ide data corruption".
Bottom line: as long as you buy a recent, quality system, Athlons are
great.
[ Of course, this can -- and will -- change as Intel, AMD, Via, Sis, and
Nvidia release new chips and change their pricing. ]
qrk
<pauls_...@pauls.seanet.com> wrote:
>Jim Thompson wrote:
>> PSpice Sim Sim time
>> Computer (seconds) at 1 GHz
>> ============ ========== ========
>> P2, 266 MHz 1800 479 (Mark/Win95)
>> P2, 440 MHz 1082 476 (Jim/NT4.0)
>> P3, 800 MHz 510 408 (Jim/Win2K)
>> P3, 1000 MHz 420 420 (Mark/Win2k)
>> P4, 1500 MHz 413 620 (Roy/Win2k)
>> ATH 1200 MHz 270 324 (Malcolm/WinME old system)
>> ATH 1200 MHz 244 293 (Mark/Win2k PC2100)
>> ATH 1200 MHz 244 293 (Mark/Win2k PC133)
>> ATH 1400 MHz 210 294 (Bud/Win2k PC2100)
>Hi Jim,
>When Roy ran the 1.5GHZ P4, did he use rambus RAM? And, were the 1.2GHz Athlons
>running 100MHZ SDRAM, 133MHZ SDRAM or DDR?
>
>Also, I have heard reports of the Athlon t-bird chips locking-up periodically. Have
>you experienced any behavior like this?
Paul, P4 was running RAMBUS. If you look in the list, you will see
that the Athlons were running PC133 and PC2100 (266 MHz DDR). The Asus
A7A266 motherboard allows you to use PC133 or DDR memory. For this
benchmark, the PC133 and DDR made no difference. This probably means
that the whole sim got loaded into L1/L2 cache. I've been running 3
Athlon's over the past month with no problems. I have inadvertantly
followed Darryl Okahata's suggestions, i.e. decent memory and power
supplies rated for the Thunderbirds.
-
Paul Spitalny
Mark,
In the PSPICE benchmark testing between P4's and Athlon chips, did the P4's use the
400MHz Rambus bus? I had heard that when the first P4 machines came out the bus was only
working at 200MHz? But, anyhow, it appears from your testing that RAM speed was not a
relevant factor. Why do you suppose that is?
Also, do you know if there would be any advantage to getting an Athlon XP processor
(their newest) vs. the regular Athlon?
Thanks
Paul
qrk
Paul, The P4 machine uses 800MHz RamBus. BTW, RamBus isn't all that
great. Search the web for articles on the RamBus fiasco. RAM speed in
this simulation is not a factor since the project probably is small
enough to get sucked up into the L1 and L2 CPU cache as noted above.
Don't get to enamored over RAM speed. There are many factors that
affect RAM speed. Here are a few memory benchmarks we have run:
Write Read
(MB/s) (MB/s)
Asus CUV4X-D, PC133: 172 666 (Via chipset)
Asus A7A266, PC133: 324 733 (Ali chipset)
Asus A7A255, PC2100: 371 869
Asus A7M266, PC2100: 533 812 (AMD chipset)
Dell, 800MHz RamBus: 975 1657 (Intel chipset)
PC2100 is DDR memory.
I have built up an Athlon XP (Palomino) machine. See results below.
The XP came in slightly slower. This could be due to the motherboard
or the processor. I don't know. The A7M266 uses the AMD chipset which
is significantly more efficient at RAM memory transfer rate than the
Ali or Via chip sets. For programs that don't use the SSE instruction
set, there isn't much advantage over the Thunderbird or the Palomino
processors. If you use programs that utilize the SSE instruction set,
then you can see a significant speed advantage. One review claimed
that some Photoshop functions ran almost 2 times faster using the XP
core. I also saw a review that you must do a fresh install of your
operating system if you replace your Thunderbird processor with the
Palomino for the SSE instruction set to work properly. The XP has
additional advantages of lower power (amounts to running 2 or 3
degrees C cooler in the machine I built up) and faster clock speeds
(beware of XP labeling, the number on the processor is not the clock
speed!). If your building up a new machine, get the XP core. The cost
differential is pretty small, perhaps $10 or $20.
PSpice Sim Sim time
Computer (seconds) at 1 GHz
============ ========== ========
P2, 266 MHz 1800 479 (Mark/Win95)
P2, 440 MHz 1082 476 (Jim/NT4.0)
P3, 800 MHz 510 408 (Jim/Win2K)
P3, 1000 MHz 420 420 (Mark/Win2k)
P4, 1500 MHz 413 620 (Roy/Win2k)
ATH TB 1200 MHz 270 324 (Malcolm/WinME old system)
ATH TB 1200 MHz 244 293 (Mark/Win2k A7A266, PC2100)
ATH TB 1200 MHz 244 293 (Mark/Win2k A7A266, PC133)
ATH TB 1400 MHz 210 294 (Bud/Win2k A7A266, PC2100)
ATH XP 1467 MHz 210 308 (Mark/Win2k A7M266, PC2100)
Darryl Okahata
> or the processor. I don't know. The A7M266 uses the AMD chipset which
> is significantly more efficient at RAM memory transfer rate than the
> Ali or Via chip sets. For programs that don't use the SSE instruction
As much as I like the A7M266, it's just about EOL (end-of-life),
because the newer VIA KT266A ("KT266A", not "KT266") and SiS 735
chipsets perform similarly (or are faster), and are cheaper.
qrk
<dar...@soco.agilent.com> wrote:
Darryl, are the KT266A boards out yet? I didn't see any when I was
looking around last month.
Paul Spitalny
Thanks Mark,
This is exactly the type of info I have been looking for. By the way, I do not use Pspice,
but, rather, Smartspice and Tanner T-Spice. But, I assume that all three programs are quite
similar in that they use lots of floating point operations and that the speed differences you
have measured will roughly apply to other SPice simulators as well. Do you agree?
Thanks for your insights
Paul
Paul
qrk
<pa...@cascadelinear.com> wrote:
[snippage]
>Thanks Mark,
>This is exactly the type of info I have been looking for. By the way, I do not use Pspice,
>but, rather, Smartspice and Tanner T-Spice. But, I assume that all three programs are quite
>similar in that they use lots of floating point operations and that the speed differences you
>have measured will roughly apply to other SPice simulators as well. Do you agree?
>
>Thanks for your insights
>
>Paul
Paul,
I would assume that other Spice programs, number crunchers, and
general programs will exhibit the same sort of benefits regarding
processor selection. Reviews and testimonials claim that the Athlon
processor is more efficient than the P3 and P4.
One thing that this benchmark does not test is RAM speed for PSpice.
The P4 RamBus transfer speed is pretty snappy compared with DDR and
PC133 in some memory tests we have run.
Darryl Okahata
> Darryl, are the KT266A boards out yet? I didn't see any when I was
> looking around last month.
Supposedly, yes. Anandtech is recommending two of them (I have no
idea if they really are any good, though):
Shuttle AK31 Revision 3.1
EPoX 8KHA+
Darryl Okahata
> I would assume that other Spice programs, number crunchers, and
> general programs will exhibit the same sort of benefits regarding
> processor selection. Reviews and testimonials claim that the Athlon
> processor is more efficient than the P3 and P4.
>
> One thing that this benchmark does not test is RAM speed for PSpice.
> The P4 RamBus transfer speed is pretty snappy compared with DDR and
> PC133 in some memory tests we have run.
Forget RAM speed benchmarks, which, when they work as advertised,
measure only one part of a system. Today's PCs are horribly complex
systems, and cannot be characterized by trivial benchmarks such as "CPU
speed" or "memory speed".
To really compare systems, you've got to run your own applications
(Spice-like programs, in this case), under a variety of conditions
(e.g., large -- and NOT necessarily long -- spice simulations).
Also note that, a system that is "faster" for one person, may not
be "faster" for another person, because each person may use different
applications. That's just how things are (it's like saying that car XXX
is "The Best").
Mohi
You might find that T-Spice may be a little different. They use a table
look up technique for speeding up simulations. This may result in more
memory usage, table lookups, simple interpolations, and numerical
differentiation, and fewer hardcore math calculations, when calculating
current/charge/conductance/capacitance for each semiconductor device. In
other words, for (very) big IC circuits, memory may play a bigger role
compared to CPU power.
Mohi
Disclaimer: These thoughts came out of MY head. Don't blame anyone else for
them.
"Paul Spitalny" <pa...@cascadelinear.com> wrote in message
news:3BF2AE2B...@cascadelinear.com...