Seagate Barracuda 7200.8: 400GBs with NCQ
by Purav Sanghani on April 20, 2005 4:30 PM EST- Posted in
- Storage
Hard Drive Buffer: Does Size Really Matter?
The answer to the title of this section is simply, "YES!" A larger buffer will definitely prove to be a performance booster no matter what type of application we use most. We mentioned how NCQ helps organize the list of requested data into a logical order relative to their location on the hard disk, and also how the rotational speed of the spindle increases the speed at which data can be read and written by the hard drives' heads. Yet we also mentioned that the drive's on board cache can be the deciding factor on how much better it performs compared to other drives.Why is the buffer so important?
Data flow to and from the system bus is a two-part process on the hard disk side. We are all familiar with solid state memory and how it is much faster than hard disk drives because solid state storage devices do not use moving parts, but instead store information on ICUs. This reduces transfer times because the solid state media does not have to search for data across a platter, in turn eliminating service times, seek times, etc. A hard drive's buffer and system bus act in a similar way through the interface. The buffer is the ICU that holds data until it is needed. The only difference is that the buffer of a hard disk drive is volatile, as it requires power to retain the data by which it is occupied, and solid state media devices are not and they can hold data without the need for a power source.On the other side of the buffer, we have a data flow path from the media to the buffer and vice-versa. This flow of data is much slower than that between the buffer and the interface because of the mechanical characteristics of hard disk media. Since the read heads need to physically move to read and write data, the entire process is almost 33% slower on a SATA drive capable of 150MB/sec transfer rates. Without a buffer, the flow of data would be much slower than what we find on drives today and the hard disk would be an even tighter bottleneck than it currently is.
In our look at the SATA vs. IDE interfaces earlier, we tested two drives that are physically identical in every way except the interface. This is probably the best way possible to show the difference in performance, which is why we have chosen a 2MB PATA version of the Samsung SpinPoint to compare to the 8MB SP1614N PATA unit. Take a look at the performance advantage of an 8MB buffer over the SP1604's 2MB buffer.
2MB vs 8MB Drive Buffer |
|||
SP1614N (8MB) |
SP1604N (2MB) |
8MB Performance Advantage |
|
SYSMark 2004 - Internet Content Creation Performance |
|||
| Overall | 201 |
197 |
2.03% |
| 3D Content Creation | 181 |
179 |
1.12% |
| 2D Content Creation | 253 |
250 |
1.2% |
| Web Publication | 177 |
171 |
3.51% |
SYSMark 2004 - Office Productivity - Communication Performance |
|||
| Overall | 173 |
157 |
10.19% |
| Communication | 199 |
153 |
30.01% |
SYSMark 2004 - Overall System Performance |
|||
| Overall Performance | 186 |
176 |
5.68% |
| Internet Content Creation | 201 |
197 |
2.03% |
| Office Productivity | 173 |
157 |
10.19% |
Winstone 2004 - Overall System Performance |
|||
| Business | 24 |
23.9 |
0.42% |
| Multimedia Content Creation | 32.2 |
31.7 |
1.58% |
| Multitasking | 2.79 |
2.72 |
2.58% |
Pure Hard Disk Performance - IPEAK, Winstone 2004 |
|||
| Business | 544 |
459 |
18.52% |
| Multimedia Content Creation | 357 |
287 |
24.39% |
Real World Performance - File System Tasks (seconds) |
|||
| File Zip (1 300MB File) | 60.321 |
61.519 |
1.99% |
| File Zip (300 1MB Files) | 61.094 |
63.210 |
3.46% |
| File UnZip (1 300MB File) | 13.928 |
14.048 |
0.86% |
| File UnZip (300 1MB Files) | 14.260 |
14.366 |
0.74% |
| Copy Folder (1 300MB File) | 5.271 |
7.853 |
48.99% |
| Copy Folder (300 1MB Files) | 6.400 |
9.909 |
54.83% |
Real World Performance - Application Load Times (seconds) |
|||
| Photoshop CS | 7.311 |
7.596 |
3.9% |
| Office 2003 - Word | 2.040 |
2.466 |
20.88% |
| Office 2003 - Excel | 2.189 |
2.437 |
11.33% |
| Office 2003 - Access | 2.449 |
2.855 |
16.58% |
| Office 2003 - PowerPoint | 2.090 |
2.817 |
34.78% |
Real World Performance - Game Level Loading Times (seconds) |
|||
| Half-Life 2 (d1_canals_01) | 19.033 |
23.533 |
23.64% |
| Doom 3 (caverns1) | 42.567 |
45.8 |
7.6% |
| C&C: Generals (GLA C3S1)* | 33.967 |
34.7 |
2.16% |
Service Time |
|||
| IPEAK Average Read Service Time | 13.53 |
14.18 |
4.8% |
WinBench 99 - Transfer Rate Test |
|||
| Beginning | 60500 |
61100 |
-0.99% |
| End | 36900 |
37600 |
-1.9% |
C&C:Generals playing as GLA (campaign 3, stage 1)
The 8MB model took the win in most of the tests, and in some cases with about a 50% performance increase over the 2MB unit. There is obviously a performance advantage when increasing the drive cache from 2MB to 8MB. A larger cache means more efficient retrieval and organization of data before the hard disk drive sends it off to be processed. More is definitely better.
Facebook
Twitter
RSS
44 Comments
View All Comments
AtaStrumf - Thursday, April 21, 2005 - link
#29 - I found a similar test that includes a WD Caviar drive and from what I can tell it is not exactly lagging.http://www.storagereview.com/articles/200504/20050...
Calin - Thursday, April 21, 2005 - link
In "WinBench99" page, you said "The Disk Transfer Rate test reads from the media in a linear fashion from the beginning (inner tracks) to the end (outer tracks)". It's false, the hard drives have the beginning tracks on the outside (well, exterior) of the platters, and the inner drives in the interior part. The reason is that while stationary, the read heads stay outside of the media, and they will reach the outer tracks sooner. Also, on the outer tracks the data density is increased, so the data read and write speed is increased also.emboss - Thursday, April 21, 2005 - link
I'd say you need to ditch Winbench 99 for transfer tests. It's physically impossible for drives to have the same transfer rate on the inside and outside of the platters. Not to mention that the ONLY drives that showed this behaviour were NCQ drives. I suspect what is happening is that the NCQ reordering is stuffing things up by reading the data out-of-order, and that the reordering process delivers the data in one (or several) burst blocks that do not correspond to the real transfer rate off the platters. Maybe HDTach might return more sensible numbers.Lonyo - Thursday, April 21, 2005 - link
Are you going to do some more HDD/NCQ testing when we get more dual core CPU's to test in multi-taking situations?The recent article on the Pentium D shows the benefits of NCQ combined with a dual core CPU (the single core CPU's didn't really show any improvement), so are you going to go more in depth hopefully soon (after you can publish results of AMD X2 CPU's)?
http://www.anandtech.com/cpuchipsets/showdoc.aspx?...
http://www.anandtech.com/cpuchipsets/showdoc.aspx?...
jm20 - Thursday, April 21, 2005 - link
How is the 7200.7 120Gb drive louder then a Raptor? My 7200.7 120Gb drive is near SILENT, no where loud as a Raptor. I think your measuring device is off forthe Acoustics test.segagenesis - Thursday, April 21, 2005 - link
#20 - Thats easy. Ignoring the Raptor they are lagging behind on the consumer front compared to others. Last I checked they still charge a fair amount extra for a drive with a FDB motor. The performance just hasnt been up to par either. The days when the "Special Edition" drives were great are gone.Palek - Wednesday, April 20, 2005 - link
Purav, you did not explain why you chose to test with an nForce chipset over a chipset from intel.For one thing, nVidia's ATA controllers/drivers have a fairly poor track record. I still remember the multitude of problems that cropped up when people installed nVidia ATA drivers on their nForce2 motherboards. I run my nForce2-based computer with MS ATA drivers because I am too afraid that the nVidia drivers will wreck my system (that, and ExactAudioCopy does not recognize any optical drives with the nVidia drivers installed). Admittedly, these issues were driver-related, but then nVidia's checkered past does not boost my confidence in their ability to provide an nForce4 driver that actually works according to spec. Maybe we're seeing no boost with NCQ because of poor implementation, who knows. Testing with just one platform will not reveal such issues.
Also, among other things intel is known for their rock-solid and impressively fast ATA controllers, so an intel chipset would be the obvious platform of choice for testing such new technologies as NCQ.
erwos - Wednesday, April 20, 2005 - link
"It's mentioned in the article that all of the 7200.8 drives use a 3x133gb platter configuration."This actually isn't true, from what I've read elsewhere. Read the following at StorageReview:
http://www.storagereview.com/articles/200504/20050...
It makes a lot more sense than the "leftover space" theorem.
-Erwos
quorm - Wednesday, April 20, 2005 - link
xsilver, the drive is not "guaranteed reliable." The only warranty is that if it breaks within five years, they will repair/replace it. There is a possibility that data can be lost from any portion of the drive. You have no way of knowing whether this additional space, if accessible, would be any less reliable than the rest of the drive. Yes, modifying the drive would probably void the warranty, but I'm wondering if Seagate is selling software-limited, yet physically identical drives at different prices, much like with ATI's 9500/9700.Zar0n - Wednesday, April 20, 2005 - link
With NCQ on u get worst results than with it off.This may be good at servers, but no good at desktop.
I’ll say its bad implemented but, all drivers seem to suffer.
So no NCQ for me...