How long to run Memtest86?

[larc-news]

While I realize there are better memory test programs than Memtest86,
it's probably the best free one. So, any suggestions on how long or
for how many passes I should run it?

Thanks for your help.
Larc



§§§ - Change planet to earth to reply by email - §§§

 

[Guest]

While I realize there are better memory test programs than Memtest86,
it's probably the best free one. So, any suggestions on how long or
for how many passes I should run it?

Thanks for your help.
Larc

Several hours at a minimum, some people run it for several days.
Personally I don't waste more than an hour or two on it. If I had
a dollar for every memory subsystem that passed Memtest with
flying colors, then proceeded to crash and be unstable when
running 3D games, I'd be very very wealthy.

 

[philo]

Several hours at a minimum, some people run it for several days.
Personally I don't waste more than an hour or two on it. If I had
a dollar for every memory subsystem that passed Memtest with
flying colors, then proceeded to crash and be unstable when
running 3D games, I'd be very very wealthy.

Right...
if no problems turn up after a few hours...the memory is probably good.

Only once did memtest pass some memory that really did not cut it...
and I've used memtest many time

 

[RobV]

While I realize there are better memory test programs than Memtest86,
it's probably the best free one. So, any suggestions on how long or
for how many passes I should run it?

Thanks for your help.
Larc

Yes, Memtest86+ (also free). Seriously, it recognizes more CPUs and
seems to be just as good, if not better than, Memtest86.
http://www.memtest.org/

As mentioned, run it for at least a few full passes, which shouldn't be
more than a few hours, at most.

 

[Larc]

Thanks very much for the help. I've already downloaded and installed
Memtest86+ to a floppy and will try it later.

BTW, I was using Memtest86 v3.0 but upgraded to the latest v3.4. I
couldn't get it to work correctly on either computer I tried it on
(both CD and floppy). All the tests would appear to run in about four
seconds and then a summary of extensive errors in every test would
come up. Had to drop back to v3.3. It works with no problems.
Larc



§§§ - Change planet to earth to reply by email - §§§

 

[Howard Goldstein]

: While I realize there are better memory test programs than Memtest86,
: it's probably the best free one. So, any suggestions on how long or
: for how many passes I should run it?

When I'm happy with the settings and plan to keep them, overnight.
Then a solid 24 hours of prime95. Then I'll start using it on it.
But this is on a combined production & fun machine, YMMV

 

[Paul]

Thanks very much for the help. I've already downloaded and installed
Memtest86+ to a floppy and will try it later.

BTW, I was using Memtest86 v3.0 but upgraded to the latest v3.4. I
couldn't get it to work correctly on either computer I tried it on
(both CD and floppy). All the tests would appear to run in about four
seconds and then a summary of extensive errors in every test would
come up. Had to drop back to v3.3. It works with no problems.
Larc

Why not try the latest one from here ?

http://www.memtest.org/

There are two streams of memtest, and the one at
memtest.org picked up, after updates for the other
version stopped for a while. So you have the
choice of two different web sites.

The memory tester respects the memory map defined
by the BIOS. If the BIOS says there are several
"do not touch" areas, then memtest works around them.
(It wouldn't be very healthy, if the BIOS was writing
in one of those zones, when memtest came barging in,
and vice versa. Memtest would conclude the memory was
bad.)

A possible reason your test results returned so fast,
is the test program has got confused about what is
reserved and what is not. It could be that by
misreading a reserved segment description, the
majority of the memory got reserved by accident.
It could be a programming bug.

Paul

 

[yaugin]

While I realize there are better memory test programs than Memtest86,
it's probably the best free one. So, any suggestions on how long or
for how many passes I should run it?

Thanks for your help.
Larc

§§§ - Change planet to earth to reply by email - §§§

Time spent is completely irrelevant as the amount of testing done
depends on system speed and the size of your memory (half the memory =
twice as much testing per hour). The more memory you need to test, the
more time you have to spend. Thus reliability should be measured
solely in the number of passes, and how many to run depends on what
margin of error you're willing to live with. Fortunately, problem
areas aren't entirely random -- bad parts tend to produce errors
quickly -- so you don't need to perform 10000 tests to have 99.99%
reliability. A bad part will likely show up within the first few
passes. I think most people do about 6 passes if they are waiting for
the results. At most I would do 10 if you are under time constraints.
Otherwise, assuming this is for your home PC, I recommend you just
leave it running overnight and/or during a workday while you are away.
This should be enough to give you a solid margin of error and does not
impede on your schedule.

 

[Ian D]

Yes, Memtest86+ (also free). Seriously, it recognizes more CPUs and seems
to be just as good, if not better than, Memtest86.
http://www.memtest.org/

As mentioned, run it for at least a few full passes, which shouldn't be
more than a few hours, at most.

The latest Memtest86+ is 1.70. It has no problems with the latest
multicore CPUs and motherboards.

 

[DevilsPGD]

In message <[email protected]> "Ian D"

The latest Memtest86+ is 1.70. It has no problems with the latest
multicore CPUs and motherboards.

Still x86 though, no x64 support, correct?

 

[Larc]

Why not try the latest one from here ?

http://www.memtest.org/

Thanks. RobV had suggested that and I've already downloaded and made
a floppy for later use.

Larc



§§§ - Change planet to earth to reply by email - §§§

 

[Ian D]

In message <[email protected]> "Ian D"


Still x86 though, no x64 support, correct?

The OS is irrelevant as it runs at machine language level and
loads off a diskette with Memtest86+'s own bootloader.

 

[John Doe]

Time spent is completely irrelevant

Unless your time is valuable.

as the amount of testing done
depends on system speed and the size of your memory (half the
memory twice as much testing per hour). Thus reliability should be
measured solely in the number of passes,

Or how long it takes to test the memory, times the number of passes.

/

 

[RobV]

The latest Memtest86+ is 1.70. It has no problems with the latest
multicore CPUs and motherboards.

"it recognizes more CPUs". I believe I said that...

 

[Larc]

Time spent is completely irrelevant as the amount of testing done
depends on system speed and the size of your memory (half the memory =
twice as much testing per hour). The more memory you need to test, the
more time you have to spend. Thus reliability should be measured
solely in the number of passes, and how many to run depends on what
margin of error you're willing to live with. Fortunately, problem
areas aren't entirely random -- bad parts tend to produce errors
quickly -- so you don't need to perform 10000 tests to have 99.99%
reliability. A bad part will likely show up within the first few
passes. I think most people do about 6 passes if they are waiting for
the results. At most I would do 10 if you are under time constraints.
Otherwise, assuming this is for your home PC, I recommend you just
leave it running overnight and/or during a workday while you are away.
This should be enough to give you a solid margin of error and does not
impede on your schedule.

Overnight seems to be the consensus. I ran v3.0 for seven passes and
no errors, but I'll let a newer version go all night to see if results
with more passes are still good.

Memory is 2MB (4x512K) of Corsair XMS PC3200 running at CAS 2.

Larc



§§§ - Change planet to earth to reply by email - §§§

 

[Paul]

Thanks. RobV had suggested that and I've already downloaded and made
a floppy for later use.

Larc

I'm sorry if I repeated what RobV suggested. The problem is, my USENET
server won't carry RobV's posts. Apparently there is a UDP for
roadrunner's USENET service, which means people using roadrunner for
USENET posts, won't necessarily have world-wide distribution.

My server (aioe) now has pretty heavy filtering, and if I really want to
follow a thread, the only way for me to do it, is to read the
newsgroups via Google interface. Which is something I hate.
The end result is, I won't know what RobV has said. Or any other
rr.com users.

Paul

 

[RobV]

Paul wrote:

[snip]

I'm sorry if I repeated what RobV suggested. The problem is, my USENET
server won't carry RobV's posts. Apparently there is a UDP for
roadrunner's USENET service, which means people using roadrunner for
USENET posts, won't necessarily have world-wide distribution.

My server (aioe) now has pretty heavy filtering, and if I really want
to follow a thread, the only way for me to do it, is to read the
newsgroups via Google interface. Which is something I hate.
The end result is, I won't know what RobV has said. Or any other
rr.com users.

Paul

Never knew such a thing could happen. I was better off with Adelphia
and GigaNews (or, whatever it's called). I agree with you about Google;
don't blame you not wanting to go there. I wouldn't go there to read
what I say, either. ;-)

I can see your posts and will keep in mind you can't see mine.

If someone could respond to this so Paul will see it, it would be much
appreciated.

RobV

 

[DevilsPGD]

In message <[email protected]> "Ian D"

The OS is irrelevant as it runs at machine language level and
loads off a diskette with Memtest86+'s own bootloader.

A swing and a miss. I am speaking to the instruction set memtest86
itself uses, if it's x86 then it cannot address or test physical RAM
which overlaps address space allocated to hardware.

If you run a x86 OS, that's not a problem, the OS can't use the RAM
either. However, if you run a x64 OS, you may end up using RAM that a
x86 tester cannot test, and therefore the tester cannot give reliable
results.

 

[DevilsPGD]

In message <[email protected]> John Doe

Unless your time is valuable.

If you sit there and watch it, yes. Most of us who value are time would
not do that.

 

[Larc]

I'm sorry if I repeated what RobV suggested. The problem is, my USENET
server won't carry RobV's posts. Apparently there is a UDP for
roadrunner's USENET service, which means people using roadrunner for
USENET posts, won't necessarily have world-wide distribution.

My server (aioe) now has pretty heavy filtering, and if I really want to
follow a thread, the only way for me to do it, is to read the
newsgroups via Google interface. Which is something I hate.
The end result is, I won't know what RobV has said. Or any other
rr.com users.

A second recommendation strengthens confidence for me. I still
appreciated it.

RobV asks that his reply to your message be passed on to you:

Never knew such a thing could happen. I was better off with Adelphia
and GigaNews (or, whatever it's called). I agree with you about
Google;
don't blame you not wanting to go there. I wouldn't go there to read
what I say, either. ;-)

I can see your posts and will keep in mind you can't see mine.

If someone could respond to this so Paul will see it, it would be much
appreciated.

RobV

Larc



§§§ - Change planet to earth to reply by email - §§§