6_6_6

NCQ: Best Upgrade For a Power User!

Recommended Posts

I somehow cannot edit my post... Well, in this example:

Average for 1st instance for 1 min run is: 76 mbps

Average for 2nd instance for 1 min run is: 29 mbps

But as i noted, i have emule reading/writing to my disk at the moment... On an empty drive, the averages are close to each other (was 48 mbps for each instance before as far as i recall).

Therefore, NCQ must be working for you then. Post the results (i believe in your previous test, NCQ was not on).

Western Digital 640GB WD6400AAKS

nVidia Geforce7050/nForce710i chipset w/AHCI enabled in BIOS, nVidia SATA-IDE v10.3.0.21 driver with NCQ enabled:

One instance: 112MB/s

Two instances (0%, 50%, 256K block)

1st instance: 112MB/s peak

2nd instance: 93MB/s peak

Graph line hops up and down like the screenshots you just posted. Average jumps around all over the place so taking the number that's displayed when 1 minute is up isn't so useful (it changes signficiantly depending on whether that particular instance is currently at peak while the other is not). Would probably have to run the benchmark for much longer for the averages to smooth out. But I'm just going to assume that with 2 instances, the average is roughly half of peak for each instance.

So it looks like we can add the WD6400AAKS to the list of drives with proper NCQ implementation.

PS My first test run was in a different system with nForce4 chipset, which apparently does not support NCQ even if it is enabled in the driver.

Share this post


Link to post
Share on other sites

Just get 1 minute averages and post their total. That is why it is average... it changes surely but it will give you the median for 1 minute. If 1 min is not satisfactory, run both instances for 10 mins and see if the averages differ than 1 min run.

Apparently this drive supports NCQ too. That is a good think. Don't you notice a difference? For example, when you do a copy, virus scan or zip, how does the system respond to you?

I somehow cannot edit my post... Well, in this example:

Average for 1st instance for 1 min run is: 76 mbps

Average for 2nd instance for 1 min run is: 29 mbps

But as i noted, i have emule reading/writing to my disk at the moment... On an empty drive, the averages are close to each other (was 48 mbps for each instance before as far as i recall).

Therefore, NCQ must be working for you then. Post the results (i believe in your previous test, NCQ was not on).

Western Digital 640GB WD6400AAKS

nVidia Geforce7050/nForce710i chipset w/AHCI enabled in BIOS, nVidia SATA-IDE v10.3.0.21 driver with NCQ enabled:

One instance: 112MB/s

Two instances (0%, 50%, 256K block)

1st instance: 112MB/s peak

2nd instance: 93MB/s peak

Graph line hops up and down like the screenshots you just posted. Average jumps around all over the place so taking the number that's displayed when 1 minute is up isn't so useful (it changes signficiantly depending on whether that particular instance is currently at peak while the other is not). Would probably have to run the benchmark for much longer for the averages to smooth out. But I'm just going to assume that with 2 instances, the average is roughly half of peak for each instance.

So it looks like we can add the WD6400AAKS to the list of drives with proper NCQ implementation.

PS My first test run was in a different system with nForce4 chipset, which apparently does not support NCQ even if it is enabled in the driver.

Share this post


Link to post
Share on other sites
One instance: 112MB/s

Two instances (0%, 50%, 256K block)

1st instance: 112MB/s peak

2nd instance: 93MB/s peak

Graph line hops up and down like the screenshots you just posted. Average jumps around all over the place so taking the number that's displayed when 1 minute is up isn't so useful (it changes signficiantly depending on whether that particular instance is currently at peak while the other is not). Would probably have to run the benchmark for much longer for the averages to smooth out. But I'm just going to assume that with 2 instances, the average is roughly half of peak for each instance.

So it looks like we can add the WD6400AAKS to the list of drives with proper NCQ implementation.

kreativ, could you please run 9 instances of the program with (5% position differences) and post the results. Samsung F1 1 TB seem to be working fine with 2 instances but not performing well with 9 instances (whereas my Seagate 7200.11 does).

Edited by 6_6_6

Share this post


Link to post
Share on other sites
One instance: 112MB/s

Two instances (0%, 50%, 256K block)

1st instance: 112MB/s peak

2nd instance: 93MB/s peak

Graph line hops up and down like the screenshots you just posted. Average jumps around all over the place so taking the number that's displayed when 1 minute is up isn't so useful (it changes signficiantly depending on whether that particular instance is currently at peak while the other is not). Would probably have to run the benchmark for much longer for the averages to smooth out. But I'm just going to assume that with 2 instances, the average is roughly half of peak for each instance.

So it looks like we can add the WD6400AAKS to the list of drives with proper NCQ implementation.

kreativ, could you please run 9 instances of the program with (5% position differences) and post the results. Samsung F1 1 TB seem to be working fine with 2 instances but not performing well with 9 instances (whereas my Seagate 7200.11 does).

kreativ - I could also greatly benefit from this test as I am attempting to ascertian which is the best one to buy. If possible please consider doing the test with these paramaters.

It would be great to know if the WD does implement NCQ well.

Thanks

LittleJhon.

Share this post


Link to post
Share on other sites

4 instances is the limit on my WD6400AAKS and nVidia nForce610i chipset in AHCI mode.

Anything past that and some of the instances will remain at near-zero instead of alternating between peak and zero.

Share this post


Link to post
Share on other sites
4 instances is the limit on my WD6400AAKS and nVidia nForce610i chipset in AHCI mode.

Anything past that and some of the instances will remain at near-zero instead of alternating between peak and zero.

Please run 9 instances and see the total throughput at:

TASK MANAGER > PERFORMANCE > RESOURCE MONITOR > DISK

Share this post


Link to post
Share on other sites
4 instances is the limit on my WD6400AAKS and nVidia nForce610i chipset in AHCI mode.

Anything past that and some of the instances will remain at near-zero instead of alternating between peak and zero.

Please run 9 instances and see the total throughput at:

TASK MANAGER > PERFORMANCE > RESOURCE MONITOR > DISK

Is the 9 instances possible in the way that 666 has described?

Share this post


Link to post
Share on other sites

I agree many drives do not implement NCQ well at all, it seems the new samsungs are very bad for it (making a new post on that one) and some drives that advertise it have clearly not had proper testing.

I think the problem with reviewers in regards to NCQ is they just testing things like raw speed and access time, game loading time stuff like that, NCQ really benefits when your computer is doing multiple things at once so lets say half a dozen apps open, you load a new app whilst outlook is downloading email and the disk is also paging. That kind of thing NCQ should handle better, whilst basic video streaming or simply copying a file I wouldnt expect much of an improvement if any. Because of the compatability issues with NCQ I would only enable it on the OS drive and use legacy ide on all other drives but of course it depends on the scenario, and everyone using raid will likely be using NCQ as well anyway since raid drivers typically enable NCQ as well.

Share this post


Link to post
Share on other sites
4 instances is the limit on my WD6400AAKS and nVidia nForce610i chipset in AHCI mode.

Anything past that and some of the instances will remain at near-zero instead of alternating between peak and zero.

Please run 9 instances and see the total throughput at:

TASK MANAGER > PERFORMANCE > RESOURCE MONITOR > DISK

Is the 9 instances possible in the way that 666 has described?

Hi kreativ, I was wondering if this is possible in the way that 666 described? The only reason I'm asking is that I really would like to know!!!!

Thanks kreativ

LittleJhon.

Share this post


Link to post
Share on other sites

Hi!

I posted my problems in Slower with NCQ ?, Concurent access and 6_6_6 responded, suggesting to do a test with hd_speed.

So here they are:

(test as recommended, with 256KB blocks, position...)

First , a test on my laptop :

HW: HP nw8440 laptop

OS: WinXP pro SP2

driver : Intel iastor.sys version 7.8

BIOS setting : "SATA Native Mode" : "Enabled" (AKA ahci mode)

Disk : ST980825AS

1 instance : 42 MB/s

2 instances (0%,50%) : 11 MB/s

9 instances (++5%) : 15 MB/s

(all results from Performance monitor , "PhysicalDisk" / "Disk Bytes/sec", unless noted otherwise)

Bad or no NCQ, I guess. (I also tried in legacy mode, with very similar results)

Then,my main system:

HW: Asus P5K-E WiFi

OS: Win2008 Datacenter SP1

driver : MS pciide

BIOS setting : "IDE"

Disk 0 : Western Digital WD7500AAKS

1 instance : 98 MB/s

2 instances : 14 MB/s

9 instances : 21 MB/s

Disk 1 : Samsung HD642JJ

1 instance : 116 MB/s

2 instances : 65 MB/s

9 instances : 21 MB/s

Not good, lets see AHCI mode (well RAID*).

driver : Intel iaStorV.sys version 6.2

BIOS setting : "RAID" , 2 individual disks

Disk 0 : Western Digital WD7500AAKS

1 instance : 100 MB/s

2 instances : 95 MB/s (010101 mode ...**)

9 instances : 27 MB/s

Disk 1 : Samsung HD642JJ

1 instance : 115 MB/s

2 instances : 15 MB/s

9 instances : 25 MB/s

Not much better.

**-that is : one instance gets full speed, while the other waits. Then switch around. Then switch... again ...

Let's also check a RAID setup while we're at it:

driver : Intel iaStorV.sys version 6.2

BIOS setting : "RAID" , 2 disks in striped mode (aka RAID-0)

Disk 0 : Western Digital WD7500AAKS

Disk 1 : Samsung HD642JJ

1 instance : 200 MB/s

2 instances : 165 MB/s (one at zero, other at full speed, see Image1)

9 instances : 30 MB/s (but system is not "frozen", all tasks get a share of the disk)

* - switching BIOS to AHCI mode makes Windows not boot, I will look into this later

PS: Does anyone know how to run HD_Speed on Linux ? I managed to start it under Wine, but it gives errors when reading the disk.

Image1 : image12instraid0vf5.png

Share this post


Link to post
Share on other sites

And the data for AHCI mode:

driver : MS AHCI / BIOS setting : "AHCI"

Western Digital WD7500AAKS

1 instance : 99 MB/s

2 instances : 90 MB/s (flip-flop mode, PerfMon froze for 5 seconds)

9 instances : 26 MB/s

Samsung HD642JJ

1 instance : 116 MB/s

2 instances : 14 MB/s

9 instances : 25 MB/s

Same but with Intel driver:

driver : Intel iaStor.sys version 8.2 / BIOS setting : "AHCI"

Western Digital WD7500AAKS

1 instance : 99 MB/s

2 instances : 90 MB/s (flip-flop mode)

9 instances : 26 MB/s

Samsung HD642JJ

1 instance : 116 MB/s

2 instances : 14 MB/s

9 instances : 25 MB/s

Edited by xerces8

Share this post


Link to post
Share on other sites

xerces8, thanks for the detailed post.

Samsung HD642JJ does not support NCQ at all as seen by the results.

Western Digital WD7500AAKS is similiar to Samsung F1 1TB... that is, it is alright at 2 instances... but it does not work at 9 instances. My Seagate 7200.11 750GB total throughput is 80 mbps at 9 instances (czr had his seagate 1TB at 90 mbps) .

If you look at my 2 instances graphs on the previous page, none of the instances block the other. They alternate (as can be seen bottom and peak spikes). So each process gets its share of the hard disk. The graph you posted below behaves similiar to non-NCQ (one instance gets it all while the other blocks).

My conclusion would be: Seagate 7200.11 is still the only drive so far to implement NCQ properly.

Share this post


Link to post
Share on other sites

I am so stupefied so many review sites mentioning that NCQ degrades performance on single-user computers and it must be turned off in workstations and kept at server use. These people apparently did not get a taste of NCQ or did not have a properly working drive and misled all of us.

Share this post


Link to post
Share on other sites
These people apparently did not get a taste of NCQ or did not have a properly working drive and misled all of us.

Well, if there is only one good NCQ drive on the market, then this is no surprise. ;)

PS: So your SCSI drive also sucks at CQ (if I read your first post, maybe in other thread, correctly) ?

Share this post


Link to post
Share on other sites
These people apparently did not get a taste of NCQ or did not have a properly working drive and misled all of us.

Well, if there is only one good NCQ drive on the market, then this is no surprise. ;)

PS: So your SCSI drive also sucks at CQ (if I read your first post, maybe in other thread, correctly) ?

Yes, that is accurate. Not mine only, but another user tested his SCSIs and same conclusion. But i must admit, my drive (Seagate 36 gb Cheetah 15K.3), albeit unused, was old. Ditto with the interface (U320) and controller. It would be good if someone having newest SCSIs with newest controllers post their results.

Share this post


Link to post
Share on other sites

What are you confused about?

It is very simple. Only 7200.11 works properly in windows. And those who do not have NCQ working properly, is missing quite a lot.

Wow, I'm so confused.
Edited by 6_6_6

Share this post


Link to post
Share on other sites

I was just copying files from a secondary drive i attached (WD 320GB SATA) just now. I was wondering why the movie would not start since a minute. Then realized the movie was on the same drive the copy process was initiated from. Cancelled the copy, the movie started. Yeah... my pre-NCQ days re-lived all over again.

In comparison... I have the OS running on my 750GB 7200.11. I do initiate a copy on the same drive from a partition to another partition in the same drive. I run a a winzip... And the same movie starts in a nanosecond.

EDIT: Oh, even emule is constantly writing to the same drive with downloads at 2 mbps.

Thank you NCQ. Thank you again.

Edited by 6_6_6

Share this post


Link to post
Share on other sites

I really think that this is a topic that storage review should do a whole article on, and also get a response from the manufacturers as to why this seems to be happening.

If most drives claim NCQ as a purcahsing feature, then why do we see such discrepencies.

I think this might expose some interesting information about this part of the market.

Why in Lunix does some drives work reasonablly well, but not on the Os of the majority of customers, enterprise or home.

Also as 666 says - if you are using the pc and get above 2gb mem in order to multitask better, whats up with a drive that keeps all the best of components waiting, and basically slows down the users session???/

Could storage review, get a response from segate, WD etc.?????

LittleJhon

Share this post


Link to post
Share on other sites

And from Intel.

We already know that the Intel ICHx/R chips cause more trouble than other chips.

During the next days I'll get a Dell Perc 5/i (similar to the LSI MegaRAID SAS 8408E). I'm really curious how it'll cope with my WD drives...

- A mate of mine uses 12 (!) WDs in a RAID 6 on a 3WARE 9650SE-12ML and has no trouble at all running multiple instances of HD_Speed at the same time. So I hope that the Perc will solve my problems.

Edited by FAT_Punisher

Share this post


Link to post
Share on other sites

Listen dude, you (not you personally) are wasting your time with benchmarks and all.

How does anyone get a taste of anything by doing benchmarks?

I told what needs to be done. Initiate a copy or a zip on a drive and try loading a program from the disk. Or try doing some other disk operation. Your system will block... and you will wait until the first disk operation ends. Why would i care if i get 500 mbps on throughput if it will keep me blocking until it finishes? Sure it will take 5 mins instead of 25 on my 100 mpbs drive.. but i will wait this 5 mins with a nonresponsive system -- unable to do absolutely nothing. I would rather let it do that in 25 mins at the background if it will let me continue use my computer meanwhile. Well, this was an extreme exaggeration... But even so, if i am given the choice, I would choose 500 mbps Non-NCQ drive against 100 mbps NCQ drive any time of the day.

Get a drive. And experience it. You won't regret. Read my topics, my complaints... and see how i went "WOW".

If you anyone would like a good grounding in this NCQ feature thing, and even those who are simply curious about the performance of some ncq capable drives have a looky here:

http://expertester.wordpress.com/2008/07/2...mark-advantage/

LittleJhon

Share this post


Link to post
Share on other sites

I have tested my two WD2500YDs and two WD6400AAKSs with the Dell Perc 5/i (LSI MegaRAID SAS 8408E).

NCQ is not working with all drives (terrible throughput with two concurrent disc accesses).

But at least the behaviour seems to be a bit better than with the ICH9R, the system does not seem to block as often anymore.

Still, the situation is very unsatisfactory. Even a quite expensive SAS/SATA RAID controller couldn't improve much. Which means I burnt a lot of money for nothing... :(

Share this post


Link to post
Share on other sites

Next, I'll try two Seagate 7200.11 640 GB. Let's hope they're really as good as you state. ;)

I'm at the point where I don't care what it costs, as long as I get the performance I want. :D

The parts I don't need will be sold at eBay... <_<

Share this post


Link to post
Share on other sites
Next, I'll try two Seagate 7200.11 640 GB. Let's hope they're really as good as you state. ;)

I'm at the point where I don't care what it costs, as long as I get the performance I want. :D

The parts I don't need will be sold at eBay... <_<

Hi Punisher, I posted this PM to you on the Hexus forum,

I read your thread regarding the 6400AAks, about AHCI and the NCQ performance. You mention that you have had some emails/correspondence from western Digital and with intel. Could you possibly post these on the forum, as I feel I have been ripped off by the manufacturer. I brought the WD on condition that it had good NCQ and AHCI performance.

You have mentioned that intel say that no one else has contacted them, and that WD have simply passed off the problem. They should be finding thier own solution if necessary, because they have our money. I would like to get some more people to write to them with this problem, so please provide the contact details, as I want the same department to start recieving these problem emails to put a little pressure on them to find a solution.

Thanks Punisher

If there is anyone here, who would like to join me in contacting both westerndigital and also intel about this, let me know, and we shall send in a combined message to make them find a soultion, or resolve this issue either way.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now