New build was fine then disk mark dropped 40%

[pazsint]

Hi...
A new member here, but I have used the Passmark Test for over 6 years now for all my upgrades and new build computers. Usually I use the old V 3.5 or 5.0 because of all the data I have on file. I use the test to compare relative improvements of new parts versus old, or a new system versus a higher tech new system.
I have often noted that I almost never get a disk mark that compares to those in your bench marks for similiar units. Even with all unecessary processes off, defragging, updated drivers, 80 wired IDEs, and tweaking the configs& bios, I have never been able to improve a low disk mark. Even the two I use for office and work station get low disk marks all the time. Even so, the computers I deliver seem to run well for the new owners.
The current new system WAS an exception... for a few days. It is nothing special: AMD Athlon 64 3200+, K8M800 board, 512 PC3200 DDR, one CD-RW, the PATA Western Digital 7200rpm 80Gb HDD, NO video card, on new install of Win XP Pro. In V3.5 this gave a 300 overall rating and a diskmark of 271-281 and consistenly for 5-6 days. I buttoned the case up and was removing progs and tracks prior to delivery when I tried Passmark one more time...and bam the overall rating now is about 270 with a low disk mark of about 165 and as low as 140, a 40% drop! With all the checks, re-tweaks, etc I can't get it back up. All the other marks for math, nmemory, 2D, 3D, and MMX as usual stayed the same, and are good.
Can you offer any proven suggestions for this phenomenom, or at least shed some light on the low disk marks we see out here in Passmark, and what that means to the practical world.
Thanks, Dan

[Ian (PassMark)]

Please confirm that the issue is that the overall disk test result dropped from around 280 to around 150 using V3.5 of PerformanceTest and that
you did not change any performance related system settings, such as the Advanced performance settings, update device drives, or add/remove hardware such as RAM?

Maybe the disk become fragmented over the initial 6 days - does the disk degragmenter indicate it needs defragmenting?

Did you test with other versions of PerformacneTest and if so did you find similar results?

Do you have the breakdown of the disk sub-test results?

Regards,
Ian (PassMark)

[pazsint]

Disk mark drop...

IAN...I have saved the data as HTML from this AMs run. But can't see a way to show it to your forum...how do you want it?
Again nothing was done to config, etc, score just dropped on the 6th day of trials. Since then I have tried many resettings and driver updates, but have not improved score on disk mark at all. I did manage a better 2 D and 3 D by resetting bios but that's all...

[Ian (PassMark)]

Please confirm that the issue seen is when using V3.5 of PerformanceTest?

Does the disk degragmenter indicate the disk needs defragmenting?

Did you test with other versions of PerformanceTest and if so did you find similar results?

You can send the break down of disk results by just typing the 3 numbers in for each subtest. Specifically I am interested in the MB/s score before and after for each of the disk sub tests (and especially the sequential read)?

FYI: The latest version of PerformanceTest (6.1) includes a text export for pasting to forums etc.

Regards,
Ian

[pazsint]

IAN...Once more this new computer has only seen the V3.5 and yes it was defrag both before and after the big disk mark drop.
Here are all the nos and again these are V3.5 after the drop, as I do not have the individual scores before the degradation, just the disk mark of 271.6 to 287.7, and and overall score was 297-300. As you see I found out how to display the html saved:
Benchmark results

Test NameThis ComputerMaths - Addition540.98Maths - Subtraction517.38Maths - Multiplication554.66Maths - Division49.79Maths - Floating Point Addition500.38Maths - Floating Point Subtraction504.17Maths - Floating Point Multiplication504.15Maths - Floating Point Division125.85Maths - Maximum MegaFLOPS1011.95Graphics 2D - Lines40.25Graphics 2D - Bitmaps72.40Graphics 2D - Shapes17.16Graphics 3D - Many Worlds416.89Memory - Allocate small block678.32Memory - Read cached182.93Memory - Read uncached180.61Memory - Write223.70Disk - Sequential Read24.90Disk - Sequential Write30.71Disk - Random Seek + RW3.28MMX - Addition777.10MMX - Subtraction777.00MMX - Multiply774.33MMX - SSE/3DNow!1548.44Maths mark427.582D Graphics mark173.09Memory Mark304.69Disk mark155.453D Graphics mark283.49MMX Mark493.33PassMark Rating271.44
System infomation - This Computer

CPU Manufacturer: AuthenticAMD
Number of CPU: 1
CPU Type: AMD Athlon(tm) 64 Processor 3200+
CPU Speed: 2198.8 MHz
Cache size: 512KB
O/S: Windows XP
Total RAM: 501661696 Bytes
Available RAM: 131747840 Bytes
Video settings: 1024x768x16
Video driver:
DESCRIPTION:
VIA/S3G UniChrome Pro IGP
MANUFACTURER:
VIA/S3G UniChrome Pro IGP
BIOS:
98.F0.00.62=
DATE:
05/16/06
Drive Letter: C
Total Disk Space: 21.0 GBytes
Cluster Size: 4 KBytes
File system: NTFS

[Ian (PassMark)]

I suspect that there is something incorrect with the device driver behaviour (but I can't prove this) such that the Disk read test is cached in the Windows cache, rather than being unbuffered (not cached), as is explicitly used by PerformanceTest for the disk test. In this case the sequential read results would be artifically too high as only the first read of the disk test file would be from the physical disk and subsequent reads (in the same test run) would be from the Windows disk cashe. Without the before reults I can't be as sure.

We changed the disk test bevahiour in V6.1 because we found a small number of disk drives ignoring the unbuffered read reqests and actually performed buffered read requests - leading to the problem described above.

If it is an option I would suggest you use V6.1.

Regards,
Ian

[pazsint]

Ian..
Thanks...another test at this point would not help me much on this build nor others from the past, because of the relative scores I've kept. Just this weekend I was called to work on a computer I built last August. I took along the 3.5V because I had data for it and it is similiar to the one I'm doing now. The test showed a significant drop in the disk mark score, much like I saw with this new one. Even after I worked on the unit, restored proper settings and defraged it, the diskmark and the overall score were the only ones noticeably lower than I recorded for it when it was newly built.
So, are there any sets or resets, or driver change that you would suggest to help sort out this phenomenon ?

[Ian (PassMark)]

I suspect that maybe a device driver has been updated with a Windows update (don't know if this is turned on), and this corrected the bug in the Hard disk device driver. (By bug, I am referring to the suspicion that the HDD device driver was ignoring the unbuffered request, and did buffered/cached reads instead). You could check whether any Windows patchs included a HDD or controller device driver updates, and if you really felt it necessary you could back it out - however I suspect that maybe the system has not actually changed its performance, but the initial result was not comparable as the disk was using cached reads (when it should have been using uncached reads).

If the system really is slower, you might also check the HDD controller is still set to DMA and not PIO.

Ian (PassMark)

[pazsint]

IAN...
THanks once again. I checked the drivers...the HDD has not been updated, nor has the primary IDE contoller that it is on. The master IDE controller has been updated, but unsure if that one would have made any difference here. It is possible that an update changed my scores, I'll agree, since this has happened on several new computers, after they have been used,and updated most likely. But on this system I can't be sure which update, as I thought all updates had been done earlier in my testwork, ie, before the diskmark score dropped.
I need to ship out this new build, but would you post a normal expected test range of disk scores using V5.0 for such a system as this, please?
Dan

[Ian (PassMark)]

At least for the moment you can still download V5 baselines from:
http://www.passmark.com/baselines/indexv5.php

Typical disk scores for a system with V5 of PerformanceTest would be around:
Read:20
Write: 15-20
Seek: 3

The method used to rate the disk performance changed quite a bit in V6.

Regards,
Ian (PassMark)

[mwd]

Recently Microsoft released 2 updates:

VIA Primary IDE Channel and Micro Innovations Mouse.

Showed up on 4 of my machines. The VIA Primary IDE update knocked out 4 for 4 of my machines. Our server lost the IDE channel and the other 3 wouldn't boot and had to be restored.

I don't have the KB # but if either of your IDE channels have had this update it would pay to do a rollback.

I would try rolling back the drivers on your Primary IDE and re-test. If there were no updates done it would tell you and if there is no success you can always update to the latest driver - on the fly.