using Memtest86+

  • Thread starter Thread starter Synapse Syndrome
  • Start date Start date
S

Synapse Syndrome

How do I use memtest86+ to check what the fastest clock speeds my memory can
reach? I thought there was some mode that checked different timings, but I
cannot find it.

Is there something else I should be using instead?

Cheers

ss.
 
Synapse Syndrome said:
How do I use memtest86+ to check what the fastest clock speeds my memory
can reach? I thought there was some mode that checked different timings,
but I cannot find it.

Set the timings, then run memtest to test for stability.

Jon
 
Jon Danniken said:
Set the timings, then run memtest to test for stability.


Ah, I see. Should I try the highest voltage that the memory is spacced to
handle and then test? ie. Does the voltage used make any difference to the
timings that have been found to be stable? As in that case I could find the
lowest voltage to use *after* I find the lowest timings.

Thanks

ss.
 
Synapse said:
Ah, I see. Should I try the highest voltage that the memory is spacced to
handle and then test? ie. Does the voltage used make any difference to the
timings that have been found to be stable? As in that case I could find the
lowest voltage to use *after* I find the lowest timings.

Thanks

ss.
In general stability increases with increasing voltage. However, power
consumption also goes up so heat may become a problem.
 
Synapse said:
Ah, I see. Should I try the highest voltage that the memory is spacced to
handle and then test? ie. Does the voltage used make any difference to the
timings that have been found to be stable? As in that case I could find the
lowest voltage to use *after* I find the lowest timings.

Thanks

ss.

Only use as much voltage as it takes to be error free. And
stay within the limits provided by the memory maker. Excess
voltage means excess heat.

Paul
 
Synapse Syndrome said:
How do I use memtest86+ to check what the fastest clock speeds my memory
can reach? I thought there was some mode that checked different timings,
but I cannot find it.

Is there something else I should be using instead?

Just run it at default and save yourself the trouble.
 
Michael said:
Just run it at default and save yourself the trouble.

Seconded. Speeding up the memory will hardly be noticeable. Running it
at default eliminates potential headaches.
 
Seconded. Speeding up the memory will hardly be noticeable. Running it
at default eliminates potential headaches.


Ok, but on the other hand, knowing which timings are stable
or what voltage is required per, can help get a system
stable that wasn't, OR help determine how high a system can
o'c the memory bus speed which may be required to raise the
FSB speed on some boards, which can be quite a noticable
difference.
 
Synapse said:
How do I use memtest86+ to check what the fastest clock speeds my memory
can
reach? I thought there was some mode that checked different timings, but
I cannot find it.

Is there something else I should be using instead?


Lots of linux livecds have memtest and other testers.
They boot and run off the CD and don't need to be installed
on hard disk to get their functionality going.

http://www.livecdlist.com

e.g. System rescue cd, DSL, Knoppix, etc..

(Search for diagnostic CDs for a full list)
 
7 said:
Lots of linux livecds have memtest and other testers.
They boot and run off the CD and don't need to be i

Memtest86+ boots up from floppy, it's totally independent on installed OS.
 
Memtest 86, as several have correctly stated, is used to test the stability
of RAM (after you have set the fsb and timings). You can run it from a
bootable floppy or CD. Memtest 86 is generally perceived as the best
program out there for testing RAM. If fact, some motherboard manufacturers
value the program so highly that the code for Memtest is written into the
bios.

However, my reason for posting to this thread is to address a couple of
idiotic statements from those who would tell you to set your memory at
default and leave it alone. That type of response is way off base and has
no constructive value in a forum like this. Let me give you a few reasons
for testing memory:

1. Some memory, even high end memory, will not run stably at default
settings. If you don't care what caused your computer to "blue screen" or
freeze up, then just set you memory to the default settings and go. The
only way to know if your memory is performing as it should is to test it.
Just last week, I RMA'd two sticks of high end dual channel memory because
they would not run stable even with the fsb lowered and the timings
loosened.

2. Not all RAM has the SPD programmed for maximum performance. Example:
The Corsair XMS memory in my computer, advertised and sold as 2-3-3-6 memory
at 200 fsb, is programmed by the SPD to run 3-3-3-8 timings at 200 fsb. In
actuality, my memory runs happily at 2.5-3-3-6 timings at 220 fsb with no
increase in voltage (or heat). I acknowlege that these differences are not
noticeable with most software, but why not make the equipment perform as it
is supposed to?

3. If you overclock to any extent, then testing your memory is extremely
important for ending up with a stable system. I'm not a proponent of
extreme overclocking, but I am a tweaker. And I don't normally mess with
voltages (with increased heat) just to get a little more performance out of
a piece of hardware. However, it is very satisfying to make a $200
component perform like a $300 component with no other sacrifice whatsoever.
There is so much headroom in today's cpus and ram, that the added
performance is there free for the taking.

Thanks for listening to my rant.

Steve
 
Steve said:
Memtest 86, as several have correctly stated, is used to test the stability
of RAM (after you have set the fsb and timings). You can run it from a
bootable floppy or CD. Memtest 86 is generally perceived as the best
program out there for testing RAM.

It's not though. Memtest86 is designed to pick up cell faults, not
timing issues. Test 5 will pick up some timing problems, but it's not
particularly quick or thorough. Prime95 large FFT is faster and picks up
more issues. I believe hardcore overclockers use Orthos, which is
essentially a wrapper for Prime95.

Even Prime95 isn't perfect. I've seen memory timing problems on more
than one machine that only showed up on some 3d apps, so I'd suggest
running a couple of versions of 3dmark as well.

As you say, many memory configurations are unstable at default settings,
and this disturbingly common. I'd strongly advise all homebuilders to
perform all of the above tests on any new PC or after a memory upgrade,
regardless of whether they're using high-end memory or aiming for
optimum performance.
 
Memtest 86, as several have correctly stated, is used to test the stability
of RAM (after you have set the fsb and timings). You can run it from a
bootable floppy or CD. Memtest 86 is generally perceived as the best
program out there for testing RAM. If fact, some motherboard manufacturers
value the program so highly that the code for Memtest is written into the
bios.

However, my reason for posting to this thread is to address a couple of
idiotic statements from those who would tell you to set your memory at
default and leave it alone. That type of response is way off base and has
no constructive value in a forum like this.

On the contrary, someone who doesn't know what/how to set
their memory to anything besides the default, should defer
to the combination of memory manufacturer's SPD programmed
timings, as read (and adjusted per the motherboard
designer's desires) and used by the bios.

It would be ill-advised to not use the default bios setting
until there is a specific reason to change it.

Let me give you a few reasons
for testing memory:

1. Some memory, even high end memory, will not run stably at default
settings. If you don't care what caused your computer to "blue screen" or
freeze up, then just set you memory to the default settings and go. The
only way to know if your memory is performing as it should is to test it.
Just last week, I RMA'd two sticks of high end dual channel memory because
they would not run stable even with the fsb lowered and the timings
loosened.


You are missing the point. If a premium was paid for this
"high end" memory, it should run at the default timings
which are those programmed by the memory manufacturer. If
those aren't stable it is beside the point whether lower bus
or timings are stable, because the premium price paid was
specificially to not have to use the lower settings/slower
performance. There is no reason to test at lower speed
unless still trying to isolate which part is the problem,
when the opportunity to RMA the modules is present.

2. Not all RAM has the SPD programmed for maximum performance. Example:
The Corsair XMS memory in my computer, advertised and sold as 2-3-3-6 memory
at 200 fsb, is programmed by the SPD to run 3-3-3-8 timings at 200 fsb. In
actuality, my memory runs happily at 2.5-3-3-6 timings at 220 fsb with no
increase in voltage (or heat). I acknowlege that these differences are not
noticeable with most software, but why not make the equipment perform as it
is supposed to?

This is a good point you make, but that is, unlike the
generic statement to use the defaults, a special case where
you know a specific reason to change from the defaults.


3. If you overclock to any extent, then testing your memory is extremely
important for ending up with a stable system. I'm not a proponent of
extreme overclocking, but I am a tweaker. And I don't normally mess with
voltages (with increased heat) just to get a little more performance out of
a piece of hardware. However, it is very satisfying to make a $200
component perform like a $300 component with no other sacrifice whatsoever.
There is so much headroom in today's cpus and ram, that the added
performance is there free for the taking.

Thanks for listening to my rant.


You are entitled to your opinion, but it seems a bit more
subjective about specific situations. In other situations
it is expected the memory timings are read correctly and
applied correctly already, it would only be a matter of
changing them after some problem or desire to overclock.
 
It's not though. Memtest86 is designed to pick up cell faults, not
timing issues.


Memtest is designed to find either fault type, a timing
issue will manifest itself too, but merely in different
addresses instead of the same continually.

Test 5 will pick up some timing problems, but it's not
particularly quick or thorough. Prime95 large FFT is faster and picks up
more issues. I believe hardcore overclockers use Orthos, which is
essentially a wrapper for Prime95.

You wouldn't run Test 5 only one time, if memory problems
are suspected or memory is in an unknown state, the tests
would be left looping for hours.

Prime95 cannot discriminate between CPU and chipset or
memory errors, and far more often generates errors from an
instable CPU than memory.


Even Prime95 isn't perfect. I've seen memory timing problems on more
than one machine that only showed up on some 3d apps, so I'd suggest
running a couple of versions of 3dmark as well.

It can help to test at a slightly higher memory bus speed
than the target for regular use. Get it testing stable with
this higher speed and it increases the supposed margin for
stabilty at the lower regular speed.


As you say, many memory configurations are unstable at default settings,
and this disturbingly common. I'd strongly advise all homebuilders to
perform all of the above tests on any new PC or after a memory upgrade,
regardless of whether they're using high-end memory or aiming for
optimum performance.

Absolutely, many people are too quick to assume they are
only tests for overclockers or for problem resolution, but I
feel WinXP shouldn't even be installed on a system /config
that hasn't yet passed memtest86+.
 
kony said:
Memtest is designed to find either fault type, a timing issue will
manifest itself too, but merely in different addresses instead of the
same continually.

Timing issues rarely show up in any of the memtest86 tests except 5, and
that misses many. It's a useful test but not a guarantee of stability on
its own.
Prime95 cannot discriminate between CPU and chipset or memory errors,
and far more often generates errors from an instable CPU than memory.

Memtest Test 5 also doesn't discriminate between CPU cache, chipset and
memory errors. It's usually impossible to hit the memory hard enough to
trigger timing errors without using the cache, so this is unavoidable.

With Prime95 you can at least run the small-FFT test to rule out the
CPU, but then actual CPU errors are very rare on properly cooled systems
run at stock settings.
 
2. Not all RAM has the SPD programmed for maximum performance. Example:
The Corsair XMS memory in my computer, advertised and sold as 2-3-3-6 memory
at 200 fsb, is programmed by the SPD to run 3-3-3-8 timings at 200 fsb. In
actuality, my memory runs happily at 2.5-3-3-6 timings at 220 fsb with no
increase in voltage (or heat). I acknowlege that these differences are not
noticeable with most software, but why not make the equipment perform as it
is supposed to?

Another example: OCZ Gold Enhanced Latency DDRAM is sold as 2.5-3-3-5
(or is it 2.0-3-3-5?) DDR400 or as 3-4-4-8 DDR500. These two types of
DIMM use exactly the same DDRAM chips, and differ only in the defaults
programmed into the SPD ROM chip. In fact, if OCZ had bothered to
implement both settings in the SPD chip (like standard Crucial PC3200,
which defaults to 2.5-3-3-8 at DDR333, and 3-3-3-8 at DDR400), then
the DIMMs would be interchangeable between DDR400 and DDR500 FSB
speeds without the need to fiddle about with the timings in the BIOS.
 
Timing issues rarely show up in any of the memtest86 tests except 5, and
that misses many. It's a useful test but not a guarantee of stability on
its own.

Depends on how bad they are, if they pass the other memtest
tests besides #5, the error rate isn't very bad.
Regardless, if you find #5 finding them best, it can be left
looping that test alone.

Memtest Test 5 also doesn't discriminate between CPU cache, chipset and
memory errors. It's usually impossible to hit the memory hard enough to
trigger timing errors without using the cache, so this is unavoidable.

Nevertheless, Prime95 errors are more often from an instable
CPU, and memtest86+ errors are more often memory-related. I
don't mean to suggest the other subsystems can be ignored if
either generates errors, but playing odds it is usually
quicker to pick the test which tends to find the related
problem more often.



With Prime95 you can at least run the small-FFT test to rule out the
CPU, but then actual CPU errors are very rare on properly cooled systems
run at stock settings.

.... until there is a problem, like improper heatsink
installation or bad capacitors or ...

We can hope these things don't occur very often but it isn't
a consolation to anyone it happens to.
 
Prime95 cannot discriminate between CPU and chipset or
memory errors, and far more often generates errors from an
instable CPU than memory.

Yes it can... by elimination... study its settings far more closely
Remember that today's CPUs have very big internal caches. Some of the
Prime95 documentation dates from a time when CPU caches were small.
You'll figure it out.

Not if you execute sufficient number of simultaneous instances with
the appropriate settings to 100% saturate the CPU core(s) plus ALL of
the available RAM.....

The best test of the lot.... run multiple instances, fully saturating
your CPU(s) and memory for at least 12 hours. Must have ZERO errors
on ALL running instances.

Mandatory if you are experimenting with any CPU or memory overclocking
or tighter than SPD memory settings and want to have impeccable
system-stability.

Not necessary, if you use Prime 95 correctly.

John Lewis
 
John said:
Not if you execute sufficient number of simultaneous instances with
the appropriate settings to 100% saturate the CPU core(s) plus ALL of
the available RAM.....

I don't buy it. Prime95 was written to generate primes. It certainly
wasn't purpose-built to check all possible memory access patterns,
whether at cell or timing level. On a multi-core system it might achieve
that (for timing faults) through interleaved accesses, but on a single
core, it's just following the same patterns. It's implausible that it'd
be perfect, unless you define perfect as "passes Prime95 for 16 hours".

Secondly, if the requirement is to saturate all RAM with Prime95 data
(which is certainly true if you're looking for cell faults), that's
simply impossible in Windows, due to the large chunk of locked kernel
memory. I have access to one old K7 machines that would fail memtest86
test 5 only on the first meg, probably because of different cache flags.
 
Back
Top