Help to interpret HD Tune results please

Mike Tomlinson wrote:

En el artículo , Mike Tomlinson
escribió:

For what it's worth, HDTune shows my two SATA3 SSDs and one SATA2 HD as
running in UDMA mode 5 (Ultra ATA/100), but another utility, SIW,
correctly identifies the SSDs as SATA-600 (aka SATA3) and the hard drive
as SATA-300 (aka SATA2). Benchmarking with another utility (ASD SSD
Benchmark) confirms that.

Just to clarify the above (I hope!): HDTach is getting its information
at the software layer (from the device driver), which, because the
controller is set to IDE mode in the BIOS, is operating the disks as if
they were IDE devices, and ATA/133 is the fastest IDE access mode
possible.

SIW looks at the actual hardware and is able to determine that the
physical controller and drive are SATA.

You've now strolled out of my technical comfort zone!

--
Terry, East Grinstead, UK

Franc Zabkar wrote:

On Sun, 20 Jan 2013 12:55:13 +0000, Terry Pinnell
put finger to keyboard and composed:

I have two identical 750 GB Samsung HDs on this Quad Core 2.66 GHz, 4 GB
RAM PC, under Win XP.

C: Benchmark
https://dl.dropbox.com/u/4019461/C-Bench-HDTune.jpg

I: Benchmark for comparison
https://dl.dropbox.com/u/4019461/I-Bench-HDTune.jpg

C: Info
https://dl.dropbox.com/u/4019461/C-Info-HDTune.jpg

C: Health
https://dl.dropbox.com/u/4019461/C-Health-HDTune.jpg

Error Scans
Both C: and I: (and both external drives) gave no errors, solid green
display.

Q1 If C: is indeed considered by the experts to be performing abnormally
badly, what can be the explanation? And what if anything can be done?
(Note: Last night I set chkdsk C: /f, rebooted and went to bed. I did the
test shortly after getting back to my PC this morning.)

Q2 Why do transfer rates drop during the test, especially C:?

Q3 Why do the access times for C: vary so much?

Q4 Are the temperatures OK?

Many thanks Franc.


The dips in the transfer rate may correspond to interference from
background Windows tasks. Otherwise there may be "slow" sectors, ie
ones that require several retries. Each retry would require an
additional revolution of the platters. That would account for the
scatter in the access time graph.

You could run MHDD in DOS mode. MHDD will scan the surface and
identify the slow sectors, ie those that require more than 500ms.

Duly downloaded and may try later.

The temperatures are good. In fact the temperature attributes are
better referred to as "Temperature Difference From 100". Therefore a
normalised value of 71 corresponds to 29C (= 100 - 71).

Thanks, that makes sense, and confirms Jaimie's post earlier. Strange way
of reporting it though, IMO. Equivalent to my boasting that I'm getting 75
mpg from my car instead of the actual 25!

The raw values for the temperature attributes consist of several bytes
corresponding to the current, maximum and minimum temperatures for the
current power cycle. As such, the raw values are best viewed in
hexadecimal mode.

For example, if the raw value of attribute C2 were 555352093, then
this would equate to 0x0000211A001D in hexadecimal. This gives us
three temperatures, namely 0x21 (= 33C), 0x1A (= 26C), and 0x1D (=
29C).

I'll take your word for that!

As for the performance, it appears that the slow drive has three 250GB
platters and 6 heads whereas the fast drive has two 500GB platters and
3 heads. Normally a fully stroked 3.5" drive would have a 2:1 ratio
between the transfer rates at the outermost and innermost zones. In
the case of the slow drive, the max/min rates are about 90/45 MB/s.
However, in the case of the faster drive, the rates would normally be
127MB/s max and 63.5MB/s min.

Now that is particularly interesting. On this OEM PC, a 'MESH GTS Xtreme,
Quad Core Q9450' nearly 5 years old, the two internal drives are supposed
to be IDENTICAL. The order confirmation from MESH UK says "2 x 750 GB
Serial ATA2 HDs with 32 MB buffer (SAMSUNG HD753LJ SCSI)". Can you amplify
on why they are setup differently please? It was an arbitrary choice at
the time to put my OS and data on C: (now the slower one). Had I known
that I: was potentially faster then I'd have chosen that, instead of using
it mainly for backup. (BTW, I do at least have my 3.4 GB pagefile on I:,
which presumably helps performance a bit?)


IME the rule of thumb for transfer rates and data densities appears to
be ...

(data rate A) / (data rate B) = sqrt (density A / density B)

So ...

(127MB/s) / (90MB/s) = 1.411

... and ...

sqrt(500GB / 250GB) = 1.414

The reason that the fast drive has a plateau at the beginning of the
benchmark graph is that it is being throttled by the SATA interface.
Other people have suggested that the SATA controller may be emulating
IDE mode. Could it be that the drives are connected to different SATA
controllers? For example, some motherboards have both Intel and
Marvell controllers.

I wish I could answer your question. Is there some way I can establish
that, bearing in mind my relatively low technical savvy?

--
Terry, East Grinstead, UK

In article , Terry Pinnell
writes

I wish I could answer your question. Is there some way I can establish
that, bearing in mind my relatively low technical savvy?

Open Device Mangler and expand the IDE/ATA/ATAPI controllers entry. It
might also be interesting to expand the Disk Drives tab to see if
slightly differing model numbers are reported for your C: and I: drives,
which would tend to support Franc's suggestion that they are in fact
different.

--
(\_/)
(='.'=)
(")_(")

Mike Tomlinson wrote:

In article , Terry Pinnell
writes

I wish I could answer your question. Is there some way I can establish
that, bearing in mind my relatively low technical savvy?

Open Device Mangler and expand the IDE/ATA/ATAPI controllers entry. It
might also be interesting to expand the Disk Drives tab to see if
slightly differing model numbers are reported for your C: and I: drives,
which would tend to support Franc's suggestion that they are in fact
different.

Here's an extract of what I see from that:

https://dl.dropbox.com/u/4019461/HDs-DeviceMgr.jpg

I wasn't sure what would be most relevant for the Controller stuff so
chose 'Device Instance Id'. Please advise if other entries in that
drop-down would help.

IDE ATA/ATAPI controllers
=========================
NVIDIA nForce Serial ATA Controller #1
--------------------------------------

Properties Details:
PCI\VEN_10DE&DEV_0266&SUBSYS_81BC1043&REV_A1\3&241 1E6FE&0&70

NVIDIA nForce Serial ATA Controller #2
--------------------------------------
Properties Details:
PCI\VEN_10DE&DEV_0267&SUBSYS_81BC1043&REV_A1\3&241 1E6FE&0&78


Disk Drives
===========
SAMSUNG HD753LJ SCSI Disk Device #1
-----------------------------------

Properties General Location:
Bus Number 0, Target ID 0, LUN 0

Properties Details Device Instance Id:
SCSI\DISK&VEN_SAMSUNG&PROD_HD753LJ&REV_1AA0\4&358D CF36&0&000


SAMSUNG HD753LJ SCSI Disk Device #2
-----------------------------------
Properties General Location:
Bus Number 1, Target ID 1, LUN 0

Properties Details Device Instance Id:
SCSI\DISK&VEN_SAMSUNG&PROD_HD753LJ&REV_1AA0\4&358D CF36&0&110

--
Terry, East Grinstead, UK

In article , Terry Pinnell
writes

https://dl.dropbox.com/u/4019461/HDs-DeviceMgr.jpg

You appear to have both PATA (parallel IDE) and SATA ports on the
motherboard, and your two Samsung drives are attached to the SATA ports
provided by the motherboard nForce chipset.

Franc's suggestion of the differing performance of the two drives being
down to them being attached to different controllers therefore doesn't
apply.

Running out of ideas here, but two suggestions:

1) have a look through Event Viewer's System and Application logs, are
any disk I/O errors reported?

2) update your nForce chipset drivers to the latest - you can do this
via nvidia.com/Download drivers.

--
(\_/)
(='.'=)
(")_(")

On Mon, 21 Jan 2013 13:24:51 +0000, Mike Tomlinson
put finger to keyboard and composed:

2) update your nForce chipset drivers to the latest - you can do this
via nvidia.com/Download drivers.

I wonder if HD Tune's Burst Rate result of 140MB/s can be believed. If
so, then this would suggest that the SATA interface is not a
bottleneck.

BTW, doesn't the fact that the drive is reported as a SCSI device
suggest that Microsoft's drivers are being used rather than NVIDIA's?
Or does this just confirm that IDE mode is active rather than AHCI?

- Franc Zabkar
--
Please remove one 'i' from my address when replying by email.

On Mon, 21 Jan 2013 10:22:09 +0000, Terry Pinnell
put finger to keyboard and composed:

The temperatures are good. In fact the temperature attributes are
better referred to as "Temperature Difference From 100". Therefore a
normalised value of 71 corresponds to 29C (= 100 - 71).

Thanks, that makes sense, and confirms Jaimie's post earlier.

I looked, but I missed it. Sorry Jamie.

Strange way of reporting it though, IMO. Equivalent to my boasting that I'm getting 75
mpg from my car instead of the actual 25!

The normalised values are "health" scores. The higher the number, the
healthier the drive. In most cases a figure of 100 reflects a perfect
score. Therefore the HDD manufacturers sometimes fiddle with the
numbers so that they appear logical. In fact WD often uses the
following relationship:

Normalised Value = 150 - Temperature

So a temperature of 50C scores 100 points while a temperature of 29C
scores 121.

See http://en.wikipedia.org/wiki/S.M.A.R.T.

... the raw values are best viewed in hexadecimal mode.

For example, if the raw value of attribute C2 were 555352093, then
this would equate to 0x0000211A001D in hexadecimal. This gives us
three temperatures, namely 0x21 (= 33C), 0x1A (= 26C), and 0x1D (=
29C).

I'll take your word for that!

It's not as difficult as it sounds. You could use Google's calculator
to help you.

http://www.google.com/search?q=555352093+in+hexadecimal
http://www.google.com/search?q=0x1D+in+decimal

As for the performance, it appears that the slow drive has three 250GB
platters and 6 heads whereas the fast drive has two 500GB platters and
3 heads. Normally a fully stroked 3.5" drive would have a 2:1 ratio
between the transfer rates at the outermost and innermost zones. In
the case of the slow drive, the max/min rates are about 90/45 MB/s.
However, in the case of the faster drive, the rates would normally be
127MB/s max and 63.5MB/s min.

Now that is particularly interesting. On this OEM PC, a 'MESH GTS Xtreme,
Quad Core Q9450' nearly 5 years old, the two internal drives are supposed
to be IDENTICAL. The order confirmation from MESH UK says "2 x 750 GB
Serial ATA2 HDs with 32 MB buffer (SAMSUNG HD753LJ SCSI)".

They are SATA drives, not SCSI.

Can you amplify on why they are setup differently please?

HDD manufacturers often play with the internal configuration. For
example, I have seen one case where WD shipped a short stroked 750GB
drive to satisfy a 500GB order. So instead of a single 500GB platter
with 2 heads, WD took a 750GB model which had two 500GB platters and 3
heads, and then reduced its capacity to 500GB by short stroking it.

Another example is Seagate's current ST2000DM001-9YN164 2TB model
which can be configured with two 1TB platters and 4 heads, or three
666GB platters and 5 or 6 heads:

http://www.users.on.net/~fzabkar/HDD/GRCC4CD9.TXT

Samsung's datasheet states that the SpinPoint F1 series can have 1, 2,
or 3 platters, with a maximum formatted capacity of 334GB per platter.

http://www.samsung.com/us/business/s...s/dsF11008.pdf

So it appears that some drives have 250GB platters, while others have
334GB. But your fast drive seems a lot faster than what we would
expect from a 334GB-per-platter model (if we extrapolate the curve,
and if we assume that the drive is fully stroked).

It was an arbitrary choice at
the time to put my OS and data on C: (now the slower one). Had I known
that I: was potentially faster then I'd have chosen that, instead of using
it mainly for backup. (BTW, I do at least have my 3.4 GB pagefile on I:,
which presumably helps performance a bit?)

Yes, I would think so.

- Franc Zabkar
--
Please remove one 'i' from my address when replying by email.

On Tue, 22 Jan 2013 07:34:36 +1100, Franc Zabkar
put finger to keyboard and composed:

I looked, but I missed it. Sorry Jamie.

Sorry, that should have been "Jaimie".

Can you amplify on why they are setup differently please?

HDD manufacturers often play with the internal configuration. For
example, I have seen one case where WD shipped a short stroked 750GB
drive to satisfy a 500GB order. So instead of a single 500GB platter
with 2 heads, WD took a 750GB model which had two 500GB platters and 3
heads, and then reduced its capacity to 500GB by short stroking it.

Another example is Seagate's current ST2000DM001-9YN164 2TB model
which can be configured with two 1TB platters and 4 heads, or three
666GB platters and 5 or 6 heads:

http://www.users.on.net/~fzabkar/HDD/GRCC4CD9.TXT

Samsung's datasheet states that the SpinPoint F1 series can have 1, 2,
or 3 platters, with a maximum formatted capacity of 334GB per platter.

http://www.samsung.com/us/business/s...s/dsF11008.pdf

So it appears that some drives have 250GB platters, while others have
334GB. But your fast drive seems a lot faster than what we would
expect from a 334GB-per-platter model (if we extrapolate the curve,
and if we assume that the drive is fully stroked).

This benchmark is consistent with three 300GB platters and 5 heads:
http://www.pcaxe.com/slike/hardver/h...ng_HD753LJ.PNG

This one looks like your slow drive (three 250GB platters, 6 heads):
http://imageshack.us/scaled/landing/...jhdtuneoi8.jpg

This one looks like it has 2 short stroked 400GB (?) platters:
http://i50.tinypic.com/2i90zr4.png

This one looks like your fast drive, also with a leading 110MB/s
plateau:
http://forums.overclockers.co.uk/sho...&postcount=702

- Franc Zabkar
--
Please remove one 'i' from my address when replying by email.

On Tue, 22 Jan 2013 07:46:59 +1100, Franc Zabkar
put finger to keyboard and composed:

This benchmark is consistent with three 300GB platters and 5 heads:
http://www.pcaxe.com/slike/hardver/h...ng_HD753LJ.PNG

Sigh, I should pay more attention before posting.

The above graph is for a write benchmark, not read. Writes are slower
than reads, so I don't know whether my interpretation makes sense.

- Franc Zabkar
--
Please remove one 'i' from my address when replying by email.

On Mon, 21 Jan 2013 16:25:14 +1100, Franc Zabkar
put finger to keyboard and composed:

On Sun, 20 Jan 2013 12:55:13 +0000, Terry Pinnell
put finger to keyboard and composed:

I have two identical 750 GB Samsung HDs on this Quad Core 2.66 GHz, 4 GB
RAM PC, under Win XP.

C: Benchmark
https://dl.dropbox.com/u/4019461/C-Bench-HDTune.jpg

I: Benchmark for comparison
https://dl.dropbox.com/u/4019461/I-Bench-HDTune.jpg

As for the performance, it appears that the slow drive has three 250GB
platters and 6 heads whereas the fast drive has two 500GB platters and
3 heads. Normally a fully stroked 3.5" drive would have a 2:1 ratio
between the transfer rates at the outermost and innermost zones. In
the case of the slow drive, the max/min rates are about 90/45 MB/s.
However, in the case of the faster drive, the rates would normally be
127MB/s max and 63.5MB/s min.

I have a WILD idea that may explain the results, but it's pure
speculation.

AIUI, 250GB-per-platter data densities result in maximum sustained
transfer rates of 90MB/s at the outermost zones for 7200RPM drives. At
333GB per platter, the figure becomes 105MB/s, and at 500GB per
platter it is of the order of 125-130MB/s.

What if the SpinPoint F1 models were at an end-of-life stage waiting
to be transitioned to 500GB-per-platter technology? The existing
technology accommodated 250GB- and 334GB-per-platter hardware and may
not have been up to the task of 500GB-per-platter data rates.

Let's assume that the existing heads had a frequency response (bits
per second) that limited them to 400GB-per-platter data densities.
Normally (?), if HDD manufacturers wish to reduce the capacity of a
1TB drive to 750GB, they will do so by shortstroking it, ie by
discarding the lesser performing innermost zones.

What if Samsung took two 500GB platters and reduced their capacities
by carving 250GB from the outermost zones? They could do so by
reducing the bits-per-inch to a value that stays within the frequency
response limits of the older heads. That would account for the
plateau, and the burst rate, and the 500GB-per-platter data rate at
the innermost zones.

- Franc Zabkar
--
Please remove one 'i' from my address when replying by email.