SUMMARY: 100ns SIMM in SPARCstation 2

From: Todd Pfaff (todd@flex.Eng.McMaster.CA)
Date: Mon Apr 27 1992 - 14:13:56 CDT


I asked:

] From: todd@flex.Eng.McMaster.CA (Todd Pfaff)
] To: sun-managers@eecs.nwu.edu
] Subject: 100ns SIMM in SPARCstation 2
]
] Will 100ns SIMMs in a SPARCstation 2 degrade the performance of the system?
] Don't bother telling me it won't work because I've already tried it and it
] does :-)

The overwhelming opinion was that it may work but reliability, not
performance, would be degraded. The SIMMs may not have passed the
manufacturers tests at a higher speed (70 or 80 ns) and so were sold
as 100 ns SIMMs, but in actual use in a SPARCstation 2 they may meet
the necessary timing specifications. The hardware does not detect the
speed of the SIMMs and slow down memory access accordingly (with wait
states). If a SIMM does not meet the necessary timing specs it would
likely cause parity errors (and panics/reboots). So the bottom line
is don't use 100ns SIMMs in SPARCstation 2 because it will lower the
reliability of the system...

But, having said this, I must add that I've had some 100ns SIMMs (mixed
with 80ns SIMMs) in a SPARCstation 2 for several months without any
reliability problems (but I was wondering about its performance, hence
the question). And to add a little more gray to the big picture, I've
had transient parity error problems on a different SPARCstation 2 and
it has only 80ns SIMMs.

Thanks for all the responses.

-------------------------------------------------------------------------------
From: dan@breeze.bellcore.com (Daniel Strick)

It will degrade performance on those occasions where the SS2 is forced
to stop and print messages like "Parity Error" and "pid 2345 killed".

Dan Strick, aka dan@bellcore.com or bellcore!dan, (201)829-4624

-------------------------------------------------------------------------------
From: George A. Planansky <gplan@aer.com>

(How do you know it works?)

I would have thought they wouldn't work reliably, unless your hardware
address logic held the cpu back (wait states) long enough to make sure
your data latched reliably ... . This is based on fun with a motorola
68000 based board a few years back, not on any specific knowlege
nor penetrating understandings. So, I am surprised. Maybe it
works most of the time? If your data were not reliably latched, then,
I wouldn't design space shuttles with it, but it probably wouldn't
degrade sidewalks all that much :-) .

We'll all be interested in your summary on *this* one. I bet
you get an earfull!

cheers,

George Planansky
Atmospheric & Environmental Research
Cambridge, MA 02139
gplanansky@aer.com (617) 547-6207

-------------------------------------------------------------------------------
From: birger@vest.sdata.no ( Birger Wathne)

Sooner or later you will crash with odd parity errors, etc....

-------------------------------------------------------------------------------
From: Daniel@europarc.xerox.com

It should work , but you may experience some strange goings on if you mix it with
80nS Simms for example. Also , how much memory does the machine have? What are
you running on it? It isn't normally recommended to run different speeds of Simm in a
machine, but it can sometimes get you out of trouble. Do you plan to leave it in there
indefinitely?

-------------------------------------------------------------------------------
From: "(Alain Brossard EPFL-SIC/SII)" <brossard@sasun1.epfl.ch>

        It will not work. SS2 needs 80ns SIMMS, it just
happens that your SIMS which are graded 100ns actually work,
for now, at 80ns. You are just looking for trouble if you
proceed this way. Mind you, if it works for now...

-------------------------------------------------------------------------------
From: "Craig Carpenter (E-Mail ccarpent@dw3f.ess.harris.com)" <ccarpent@dw3f.ess.harris.com>

The word we got from our local Sun rep is that the only "supported"
memory level is the 70ns SIMMS.

My guess is that if all your chips are 100ns then all you should loose
is a little performance, particulary during memory swapping.

-------------------------------------------------------------------------------
From: don@mars.dgrc.doc.ca (Donald McLachlan)

Well, it may work, but in the past we have had "80ns" ram (which didn't
meet Sun specs) which worked, but occasionally showed parity errors. If
you can live with that, then go for it!

-------------------------------------------------------------------------------
From: "Jon Mellott" <jon@alpha.ee.ufl.edu>

Well, it's not supposed to work. Those memories did not pass
testing at higher speeds so they were sold at 100 ns, the speed
at which they passed testing. Maybe most of the memory operates in
the 70-80 ns range, however, all of the memory passed only at the
100 ns speed, not the 80ns speed. You will probably experience
greater error rates than you might have otherwise.

And, except for parity error panics, you won't get degraded
performance because the machine has no way of telling what
kind of memory you've put in the SIMM sockets. Parity errors
will indicate whether it is working or not. You could put in
35ns SIMMs and the machine would still run them as if they were
80 ns. Get the picture?

Good luck, and I hope that you aren't counting on the machine
not panicing frequently :)

Jon Mellott
High Speed Digital Architecture Laboratory
University of Florida
jon@alpha.ee.ufl.edu

-------------------------------------------------------------------------------
From: poffen@sj.ate.slb.com (Russ Poffenberger)

It won't degrade it in the sense of running slower. The memory is asynchronous
in that the CPU doesn't wait for the memory to signal it is ready, the
access time is fixed based on the CPU clock cycle and other associated
circuitry.

What you may find are random crashes caused by memory errors. While it may
appear to work at first, be prepared to have flakiness or crashes, especially
during heavy loading, or if the temperature increases.

Russ Poffenberger DOMAIN: poffen@sj.ate.slb.com
Schlumberger Technologies UUCP: {uunet,decwrl,amdahl}!sjsca4!poffen
1601 Technology Drive CIS: 72401,276
San Jose, Ca. 95110 Voice: (408)437-5254 FAX: (408)437-5246

-------------------------------------------------------------------------------
From: Steve Hanson <hanson@pogo.fnal.gov>

Well, depends on what you mean by "works". You're undoubtedly
going to get parity errors from time to time (maybe months apart).
Your system may crash, etc unexpectedly. It won't degrade the
performance of your system in other ways. Even normal 80 ns SIMMS
tend to be flakey in SS-2 machines. I wouldn't want to trust any
real important work to a machine set up that way.

-- 
Steve Hanson - FERMILAB, Batavia, Il.  (708)840-8043
hanson@pogo.fnal.gov or hanson@fnal.fnal.gov

------------------------------------------------------------------------------- From: Bill Bathurst <bathurst@noaapmel.gov>

If you like having an additional wait state, yes it will. If you have 70ns memory you basically have only one wait state. Now you have two (:

------------------------------------------------------------------------------- From: mark@maui.Qualcomm.COM (Mark Erikson)

the 100ns is a minimum performance specification and (more or less) guarantee as slow as 100ns. What you will find is that you will get parity errors every so often. It may take a week, month, year, phase of the moon. We have found that 100ns sims have more problems over a period of time than 80ns.

------------------------------------------------------------------------------- From: etnibsd!vsh@uunet.UU.NET (Steve Harris)

You're just lucky.

-- .. Steve Harris - Eaton Corp. - Beverly, MA - uunet!etnibsd!vsh

------------------------------------------------------------------------------- From: Perry_Hutchison.Portland@xerox.com

I believe it will not degrade the performance, but it will degrade the reliability. The problem is that, while a memory device rated at 100ns will often deliver its data within the SS2's timing requirements, it cannot be depended upon to always do so. You are likely to experience occasional failures, including undetectable (double-bit) failures, especially at temperatures near the operating limits.

------------------------------------------------------------------------------- From: seidc!gwolsk@mips.com (Guntram Wolski)

I will assume you are just lucky. Your 100ns SIMMs may actually be working much faster but because they missed some spec when they were being tested at the manufacturer, they probably downgraded them to 100ns memories. No vendor would have given up the chance to sell 80ns SIMMs at the time when 80ns were a premium. I bet the 100ns SIMMs might also be working because they meet that spec that SUNs need and that is why you need to buy VENDOR approved SIMMs. (I might be able to dig up the old sun-managers mail message describing the actual timing if you want).

--G -- Guntram Wolski gwolsk@sei.com Sequoia Semiconductor, Inc. 408-438-5331 x112 ...!{mips,ames}!seidc!gwolsk

------------------------------------------------------------------------------- From: paulo@dcc.unicamp.br (Paulo L. de Geus)

No, memory circuits don't auto-sense the speed of memory chips, since this varies with temperature, among other factors.

So, if it works, you go at the same speed. If it is too slow for the memory controller, it just won't work. Some computers in the past allowed you to set the number of wait states to insert to CPU cycles in order to match the speed of the memory if fitted in. With massification of the memory production industry, everybody produces the same speeds, at alomos the same prices, so there's no point anymore.

-- Paulo L de Geus INTERNET:paulo@dcc.unicamp.br Depto de Ciencia da Computacao DCC - IMECC - UNICAMP caixa postal: 6065 13081 Campinas SP Brazil

------------------------------------------------------------------------------- From: dd@mips.com (David DiGiacomo)

The CPU has no way of knowing what speed SIMMs you've installed. It just sends an address and expects to see data ~80 ns later (which requires 70 ns SIMMs for other reasons). So, it won't work reliably, but if it does work, it won't affect system performance.

------------------------------------------------------------------------------- From: cae780!miles!bill@apple.com (Bill McDonald)

Yes, they will work. However, the memory managemnet chip will slow all traffic down to the slowest simm. This is the only way it knows how to avoid timing errors. Its also a cheap and easy algorythim {sic} that works.

--bill mcdonald --comdisco systems, inc --foster city, ca --voice (415) 358-3665

------------------------------------------------------------------------------- From: Terry Rosenbaum <tlr@toy.rad.msu.edu>

Just a wild guess:

Perhaps the '100ns SIMMs' you tried just happened to work with the timing for 70ns SIMMS that the systems expects. That may have been luck alone. You may have gotten a batch of SIMMs that were not qualified as 70ns SIMMs, but worked as such by LUCK alone. The situation where chips work at faster clock speeds than rated for is common. However, next time you may not be so lucky ?

------------------------------------------------------------------------------- From: jjmhome!jjm@transfer.stratus.com (Jim Murray)

Only in the sense that it may lower the reliability. The machine is going to access them at the faster speed. The set of ram chips you have are runing at the faster speed. You can probably get away with this these days since almost all ram chips manufactured today fall into the 60 to 80 nano second range anyway.

What happens with ram chips is that they are all the speeds are manufactured on the same line and then they are tested and labeled for the speed that they passed. What usually happens is this. Most of the parts coming off the line will pass at 60 or 70 nsec so when that the company has orders for 100nsec parts they will not have enough and will either label parts that have passed for a faster speed and ship them more likely they will just test the next next batch for 100nsec and ship them. Of course most of these parts would have passed as faster parts.

Bottom line. It is almost impossible to find a true 100nsec ram chip today no matter what they are labeled. On the otherhand if you had some 100nsec ram chips of 3 or 4 years ago they would probably not work in your machine.

Jim --- Jim Murray INET jjm%jjmhome@stratus.com 2 Mohawk Circle SMART UUCP jjm@jjmhome.uucp Westboro Mass 01581 uunet!lectroid!jjmhome!jjm USA voice (508) 366-2813

------------------------------------------------------------------------------- From: "Anthony A. Datri" <datri@concave.convex.com>

"work" and "work reliably" are two very different things. 100/120ns chips are typically 70/80 ns chips that didn't pass margin testing at 70/80 ns, but did at 100ns. You're taking something of a risk by putting 100ns-rated chips in a 4/75.

------------------------------------------------------------------------------- From: kalli!kevin@fourx.Aus.Sun.COM (Kevin Sheehan {Consulting Poster Child})

For the moment - you will have problems, trust me.

l & h, kev

-------------------------------------------------------------------------------

-- Todd Pfaff \ Internet: todd@flex.eng.mcmaster.ca Dept. of Mechanical Engineering \ Voice: (416) 525-9140 x2902 McMaster University \ FAX: (416) 572-7944 Hamilton, Ontario, CANADA L8S 4L7 \



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:41 CDT