UDMA tranfer rate problem

[pehache grmpf]

If not, I'd try a completey clean install of XP and see
what happens with the HDTach charts with that.

This gave me the idea to test it under an unused W98 that I still have on a
small partition.

I had to get HDTach2.7, and to run the test several times to have a full
report without any crash, but it was worth the effort! For the internal
drive:
Average transfer rate 42MB/S
Burst Rate 60MB/s
Access time 11ms
http://cjoint.com/?enuxykaEnk

(only the first 8GB of the disk are tested, but I guess that this is "by
design")

Well, it seems that the hardware is OK and that the problem is somewhere in
WinXP...

(I didn't test the external drive; is W98 supposed to manage USB2 ?)

 

[Folkert Rienstra]

This gave me the idea to test it under an unused W98 that I still have on
a small partition.

I had to get HDTach2.7, and to run the test several times to have a full
report without any crash, but it was worth the effort! For the internal
drive:
Average transfer rate 42MB/s
Burst Rate 60MB/s
Access time 11ms
http://cjoint.com/?enuxykaEnk

(only the first 8GB of the disk are tested,

but I guess that this is "by design")

Nope, just from complete stupidity by the programmers.
They fixed a problem rather stupidly in 2.6 by adding
a checkable option.
Apparently the option was confusing users so they deci-
ded to remove the option and leave the problem as is.

Since the option in 2.61 enabled the correct code it's completely in-
comprehensible that they didn't include that code in 2.70 permanently.

Well, it seems that the hardware is OK

You think so? :
"and to run the test several times to have a full report without any crash"

 

[pehache grmpf]

Yep.

Totally crazy day...

Just after my lastest post, I saw something strange when rebooting the PC:
the frequency had dropped from 1533Mhz to 1150Mhz. Last time I had that, it
was because of flashing the BIOS. But these last days I didn't flash
anything...

Indeed, for some reasons, the MoBo BIOS had reverted from AMI 1.9 to AMI
1.8, with all the settings reset !!! I noticed also a few problems, like a
CRC error during the first boot this evening (don't remember the exact
message, but it was before the BIOS setup screen), and a wrong detection of
"CPU Fan isn't running" (it does run).

Anyway, I then decided to run again HDTach under XP, with all the 1.8 BIOS
settings reset, just to see... And I had slightly better results than
yesterday, with the 1.9 BIOS: average 14MB/s instead of 9 (but a poorer
burst).

Then, I decided to flash the MoBo with the latest BIOS release from MSI
(1.C). Then I ran HDTach, and again the results were better: average 25MB/s.

Last, I went to the BIOS setup, and saw that there was an option "Load high
performance defaults". I selected it, and ran again HDTach... and the
results are almost perfect !!! Average=37.5MB/s, Burst=53.1MB/s.
http://cjoint.com/?enwJkh1w3Z (my disk in red, reference bench for this disk
in blue)

That's not the end: in the BIOS, I then reverted to the default settings
(not high perf) with the option "Load setup defaults"... without any impact
on the HDTach test: still optimum perf for the drive!


I have now good performances, but I'm completely unable to understand and
explain why. Question: is computer science really a science ? Computer art
would be more appropriate...

(I couldn't test the external drive, as I don't have it at home today).


Many thanks anyway for those who helped me !

 

[Rod Speed]

Rod Speed wrote

Totally crazy day...

Yeah, the brown stuff can hit the fan some days.

Just after my lastest post, I saw something strange when rebooting the
PC: the frequency had dropped from 1533Mhz to 1150Mhz. Last time I had
that, it was because of flashing the BIOS. But these last days I didn't
flash anything...

Indeed, for some reasons, the MoBo BIOS had reverted from AMI 1.9 to AMI
1.8,

Thats even odder.

with all the settings reset !!! I noticed also a few problems,
like a CRC error during the first boot this evening (don't remember
the exact message, but it was before the BIOS setup screen),

Thats usually due the cmos battery needing to be replaced.

and a wrong detection of "CPU Fan isn't running" (it does run).

Anyway, I then decided to run again HDTach under XP, with all the 1.8
BIOS settings reset, just to see... And I had slightly better results
than yesterday, with the 1.9 BIOS: average 14MB/s instead of 9 (but a
poorer burst).

Then, I decided to flash the MoBo with the latest BIOS release from MSI
(1.C). Then I ran HDTach, and again the results were better: average
25MB/s.

Last, I went to the BIOS setup, and saw that there was an option
"Load high performance defaults". I selected it, and ran again
HDTach... and the results are almost perfect !!! Average=37.5MB/s,
Burst=53.1MB/s. http://cjoint.com/?enwJkh1w3Z (my disk in red,
reference bench for this disk in blue)

Yeah, that's clearly fixed the problem.

That's not the end: in the BIOS, I then reverted to the default
settings (not high perf) with the option "Load setup defaults"...
without any impact on the HDTach test: still optimum perf for the drive!

I have now good performances, but I'm completely unable to understand and
explain why.

Looks like the cmos got scrambled and thats
what produced the lousy results. The cmos almost
certainly got scrambled by a bad cmos battery.

Question: is computer science really a science ?

Yep. It can appear to be more like black magic at times, but when
you do take the trouble to diagnose a problem properly, it always
turns out to be something quite basic in the end, even if its a fault.

Computer art would be more appropriate...

Nar, it only looks that way at times.

(I couldn't test the external drive, as I don't have it at home today).

OK, bet its fixed that too.

Many thanks anyway for those who helped me !

Thanks for the feedback, much too rare in my opinion with
problems that arent completely straighforward and easy.

Replace the cmos battery to prevent the problem recurring.

 

[Rod Speed]

OK, I think I can now explain that: in the first test with the 1.C
BIOS, the settings were not truly defaults. I selected at least "CPU
Halt detection" and "S.M.A.R.T. for hard drives". After that, when
toggling between high perf defaults and std defaults, those settings
were always reset and disabled.

And it seems that these settings are among the responsible ones for
the HD perfs. Here are the average transfer rates at the beginning of
the disk (all in MB/s):

*Fully default BIOS settings (means FSB=100, and not "high perf"):
HDTach~42 HDTune~42

* +CPU Halt:
HDTach~17 HDTune~42

* +CPU Halt +FSB=133
HDTach~22 HDTune~16

* +SMART +FSB=133
HDTach~42 HDTune~42

* +CPU Halt +SMART +FSB=133
HDTach~22 HDTune~42

What is surprising is again the discrepancy between HDTach and HDTune.

It seems that the "CPU Halt detection" feature is harmful for the
transfer rate, at least when measured with HDTach (not always when
measured when HDTune). In the past, I used to enable it, inorder to
be able reduce the CPU fan speed a bit (less noise). Maybe I should
change that...

Yeah, it certainly appears to have some undesirable side
effects, but that may not happen with real world work, it
may well just be some artifact with something like HDTach.

Finally, the HD perf problem result from an interaction between the OS,
the BIOS version, the BIOS settings, ... and the soft used for the
measures.

Yeah, tho that variation with the software isnt all that surprising,
you can obviously measure the average thruput different ways.

It would be interesting to see the full HDTach chart with just
CPU Halt enabled.

I wouldnt expect SMART to have any effect
and it doesnt appear to from the numbers above.

 

[pehache grmpf]

Then, I decided to flash the MoBo with the latest BIOS release from
MSI (1.C). Then I ran HDTach, and again the results were better:
average 25MB/s.
Last, I went to the BIOS setup, and saw that there was an option
"Load high performance defaults". I selected it, and ran again
HDTach... and the results are almost perfect !!! Average=37.5MB/s,
Burst=53.1MB/s. http://cjoint.com/?enwJkh1w3Z (my disk in red,
reference bench for this disk in blue)

That's not the end: in the BIOS, I then reverted to the default
settings (not high perf) with the option "Load setup defaults"...
without any impact on the HDTach test: still optimum perf for the
drive!

OK, I think I can now explain that: in the first test with the 1.C BIOS, the
settings were not truly defaults. I selected at least "CPU Halt detection"
and "S.M.A.R.T. for hard drives". After that, when toggling between high
perf defaults and std defaults, those settings were always reset and
disabled.

And it seems that these settings are among the responsible ones for the HD
perfs. Here are the average transfer rates at the beginning of the disk (all
in MB/s):

*Fully default BIOS settings (means FSB=100, and not "high perf"):
HDTach~42 HDTune~42

* +CPU Halt:
HDTach~17 HDTune~42

* +CPU Halt +FSB=133
HDTach~22 HDTune~16

* +SMART +FSB=133
HDTach~42 HDTune~42

* +CPU Halt +SMART +FSB=133
HDTach~22 HDTune~42


What is surprising is again the discrepancy between HDTach and HDTune.

It seems that the "CPU Halt detection" feature is harmful for the transfer
rate, at least when measured with HDTach (not always when measured when
HDTune). In the past, I used to enable it, inorder to be able reduce the CPU
fan speed a bit (less noise). Maybe I should change that...



Finally, the HD perf problem result from an interaction between the OS, the
BIOS version, the BIOS settings, ... and the soft used for the measures.

 

[pehache grmpf]

Thats usually due the cmos battery needing to be replaced.
...
Replace the cmos battery to prevent the problem recurring.

Good point !

I even did not remember that there was a battery on the MoBo :-( !

Very likely it is tired...

 

[pehache grmpf]

You think so? :
"and to run the test several times to have a full report without any
crash"

Well, it's W98

(OK, there still may be a hardware problem...)

 

[pehache grmpf]

Yeah, it certainly appears to have some undesirable side
effects, but that may not happen with real world work, it
may well just be some artifact with something like HDTach.

Yeah, tho that variation with the software isnt all that surprising,
you can obviously measure the average thruput different ways.

Here is what the real world says: I tried to read/write a big image file
(600MB) from an image editing software. The observed average transfer rate
is (actually there are significant variations between the trials; also, I
can't garantee that there's no fragmentation, which slows down the
transfer):

Without CPU Halt: read~35MB/s, write~31MB/s
With CPU Halt: read~33MB/s, write~15MB/s

The CPU Halt has the most impact on the write transfer rate.

It would be interesting to see the full HDTach chart with just
CPU Halt enabled.

I wouldnt expect SMART to have any effect
and it doesnt appear to from the numbers above.

If you look carefully at the results, in some situations, SMART has some
effect on the HDTune measure... BTW, what is SMART ? Also, should I
enable "32 bits transfer mode" for the drives in the BIOS (I did try
previously, without any impact on the rates) ?

 

[Folkert Rienstra]

OK, I think I can now explain that: in the first test with the 1.C BIOS, the
settings were not truly defaults. I selected at least "CPU Halt detection"
and "S.M.A.R.T. for hard drives". After that, when toggling between high
perf defaults and std defaults, those settings were always reset and disabled.

And it seems that these settings are among the responsible ones for the HD
perfs. Here are the average transfer rates at the beginning of the disk (all
in MB/s):

*Fully default BIOS settings (means FSB=100, and not "high perf"):
HDTach~42 HDTune~42

* +CPU Halt:
HDTach~17 HDTune~42

* +CPU Halt +FSB=133
HDTach~22 HDTune~16

* +SMART +FSB=133
HDTach~42 HDTune~42

* +CPU Halt +SMART +FSB=133
HDTach~22 HDTune~42

What is surprising is again the discrepancy between HDTach and HDTune.

Yech. HD Tach may not be all that undisputed then, afterall.

It seems that the "CPU Halt detection" feature is harmful for the transfer
rate,

at least when measured with HDTach

HD Tach v2 used to be using a mix of BIOS and driver calls. E.G., if you
deselected a drive from BIOS support, HD Tach would not even see that drive.
So yes, BIOS features that have an effect on BIOS calls may influence the result.
Don't know about v3.

(not always when measured when HDTune).
In the past, I used to enable it, inorder to be able reduce the CPU
fan speed a bit (less noise).

Maybe I should change that...

Or not if it only influences HD Tach and not true Windows apps.