WD Raptor - Real SATA?

John Gardner · Oct 6, 2003

I was reading an article in Lost Circuits regarding the AMD 64 CPU, and to my surprise they referred to the Western Digital Raptor (10K rpm SATA) as a "bridge" solution and not a native SATA implementation. For the review they instead used a Seagate? 7001 or some such.

I couldn't believe what I read. Is the Raptor not a native SATA implementation? I find it hard to believe. Since I plan on buying a pair, or more, I would really like to hear from someone who know whether these are true native SATA implementations.

Thanks!
John G.

Wayne Youngman · Oct 7, 2003

"John Gardner" wrote

I was reading an article in Lost Circuits regarding the AMD 64 CPU, and to my surprise they referred to the Western Digital Raptor (10K rpm SATA) as

a

"bridge" solution and not a native SATA implementation. For the review they instead used a Seagate? 7001 or some such.

I couldn't believe what I read. Is the Raptor not a native SATA implementation? I find it hard to believe. Since I plan on buying a pair, or more, I would really like to hear from someone who know whether these are true native SATA implementations.

Hi,

for a better newsgroup to discuss this try: comp.sys.ibm.pc.hardware.storage

As far as I know the Raptors use a *Bridge* chip to convert PATA signals to SATA, therefore they are *not* true native SATA disks. . . however they still work very well. I believe WD will release a newer 72GB raptor very soon. . . .

Richard Hopkins · Oct 7, 2003

"John Gardner" wrote:

I was reading an article in Lost Circuits regarding the AMD 64 CPU, and to my surprise they referred to the Western Digital Raptor (10K rpm SATA) as a "bridge" solution and not a native SATA implementation.

Why are you surprised? Virtually all the S-ATA hard disks available at the moment are "bridged" versions of parallel ATA designs.

For the review they instead used a Seagate? 7001 or some such.

Possibly because Seagate are the only manufacturer (at time of writing) selling hard disks with native Serial-ATA controller electronics. Everyone else's are bridged.

Is the Raptor not a native SATA implementation?

Correct. It is a parallel ATA drive with a bridge chip to convert to/from serial. Like all current drives bar Seagate's, it's best to think of them as parallel ATA drives with built-in Serillel converters.

Since I plan on buying a pair, or more, I would really like to hear from someone who know whether these are true native SATA implementations.

I'm not sure what you're asking here. Are you just having a random bitch because the drives aren't what you thought they were, or are you asking whether you should buy them?

If you liked the Raptor's combination of strengths enough to want to buy some in the first place, knowing about their bridged nature should not be enough to put you off. As the Seagate drives show, "native" SATA operation doesn't confer significant performance benefits over bridged or parallel operation.

Speaking personally though, I don't find the Raptor's high price per gigabyte and small capacity an acceptable trade-off for its supposed speed benefits. It is matched in some benchmarks by the fastest 7200rpm drives, which are much cheaper and available in much higher capacities.

John Lewis · Oct 7, 2003

On Tue, 7 Oct 2003 12:34:18 +0100, "Richard Hopkins"
wrote:

"John Gardner" wrote:
I was reading an article in Lost Circuits regarding the AMD 64 CPU,
and to my surprise they referred to the Western Digital Raptor (10K
rpm SATA) as a "bridge" solution and not a native SATA
implementation.

Why are you surprised? Virtually all the S-ATA hard disks available at the moment are "bridged" versions of parallel ATA designs.

For the review they instead used a Seagate? 7001 or some such.

Possibly because Seagate are the only manufacturer (at time of writing) ...

Yes, indeed !!

Thanks, Richard for clearly pointing this out. Also there are few applications in a single-processor PC, HT or not that won't throttle data-flow for reasons other than drives. The latest 7200/8Meg are a VERY economic and nicely-tuned match to a current high-performance PC and for those who need the extra oomph........there is always RAID0. DV- Video sure does not need Raptors -- 50Mbits/sec (7 Mbytes/sec)streaming for DV50; normal DV is 25 Mbits/sec. And uni- or dual processor PCs are hopeless for any processing of full-bandwidth component video ( 270 Mbit/sec for SD,
1.5 Gbits/sec for HD )

John Lewis

Richard Hopkins
Cardiff, Wales, United Kingdom
(replace .nospam with .com in reply address)

The UK's leading technology reseller www.dabs.com
Out with the old & in with the new at www.dabsxchange.com Get the most out of your digital photos www.dabsxpose.com

Greg Jones · Oct 7, 2003

Speaking personally though, I don't find the Raptor's high price per
gigabyte and small capacity an acceptable trade-off for its supposed speed benefits. It is matched in some benchmarks by the fastest 7200rpm drives,
which are much cheaper and available in much higher capacities.
--

Richard Hopkins
Cardiff, Wales, United Kingdom
(replace .nospam with .com in reply address)

The UK's leading technology reseller www.dabs.com
Out with the old & in with the new at www.dabsxchange.com
Get the most out of your digital photos www.dabsxpose.com

Not to pick a fight or anything, but I've built a system with Dual Raptors using RAID 0 and it screams. The otherday I was running some MD5 checks over large files and was getting 110 MB/sec read operations. Considering the PCI is limited to 132 MB/sec (theoritical) these numbers are blazing...

I can't argue that the best value is with the 8MB cache 7200rpm drives. The Raptors are expensive about 3.25x the cost per Gigabyte and for many folks this is just not worth the extra cost. However, folks shell out lots of extra dollars to squeeze more out of there memory, or to get a few extra Mhz out of the CPU. Just notice how often when you are waiting on your computer that the disk light is on. I belive disk performance has been a limiting factor in system performance for far too long. The market has been focused on large and cheep drives and performance has taken a back-seat, except for the server market, where SCSI or Fibre Channel has a strong hold, but talk about expensive

But I've long wanted SCSI drives but didn't want to shell out the bucks for the drives and controllers. Now the raptor provides a good middle ground. The drives are just a tiny bit cheeper, but I don't have to buy the SCSI controller.

Just to share some real data I collected because I did want to see if my investiment was paying off. Here are some WinBench 99 DiskMark scores I measured from my P4 2.4 Ghz system. Granted these are disk oriented benchmarks, but it gives you a good feel for the relative disk I/O performance of various configuration taken using the exact same system. Sorry I don't have dual WD 80 GB 7200 (8 MB) drives, I'd loved to have been able to measure them in a RAID 0 configuration too.

WinBench 99 Configuration Business (KB/s) High-End(KB/s) Avg Access Time (ms) Raptors - RAID 0 (32K Stripe) 16700 54533 8.98 ms Raptor - Single Drive 15400 39500 8.71 ms Western Digital 80 GB 7200 (8MB) 9050 32200
12.3

I'm happy and overjoyed that the IDE market has finally started to get some high-performance disks. Hopefully the success of the Raptor will bring in more vendors and we'll see a drop in the cost of these 10K RPM drives and some increase in capacity. I've heard that WD is going to release a 2nd generation of the drive that will be at about 70 GB. So you may want to wait for that version, perhaps it will be a little cheeper.

As for the bridge, from what I've read the overhead is minimal and that shouldn't be a driving consideration for your decision.

jpsga · Oct 7, 2003

Glad to hear from Greg.

My raptors with a 16 K stripe perform very well.

PCMark 2002 showed and uncached read of 133.3 MB/s.

That is consistent with the modtown article of 18 September.

We still do not have a good speed test for RAID configurations.

JPS

"Greg Jones" wrote:

Speaking personally though, I don't find the Raptor's high price per
gigabyte and small capacity an acceptable trade-off for its supposed speed
benefits. It is matched in some benchmarks by the fastest 7200rpm drives, which are much cheaper and available in much higher capacities.
--

Richard Hopkins
Cardiff, Wales, United Kingdom
(replace .nospam with .com in reply address)

The UK's leading technology reseller www.dabs.com
Out with the old & in with the new at www.dabsxchange.com
Get the most out of your digital photos www.dabsxpose.com

... WinBench 99
Configuration Business (KB/s) High-End(KB/s) Avg Access Time (ms) Raptors - RAID 0 (32K Stripe) 16700 54533 8.98 ms Raptor - Single Drive 15400 39500 8.71 ms Western Digital 80 GB 7200 (8MB) 9050 32200
12.3

I'm happy and overjoyed that the IDE market has finally started to get some high-performance disks. Hopefully the success of the Raptor will ... 2nd generation of the drive that will be at about 70 GB. So you may want to wait for that version, perhaps it will be a little cheeper.

As for the bridge, from what I've read the overhead is minimal and that shouldn't be a driving consideration for your decision.

John Lewis · Oct 8, 2003

On Wed, 08 Oct 2003 02:50:10 GMT, "Greg Jones"
wrote:

Speaking personally though, I don't find the Raptor's high price per
gigabyte and small capacity an acceptable trade-off for its supposed speed
benefits. It is matched in some benchmarks by the fastest 7200rpm drives,
which are much cheaper and available in much higher capacities.
--

Richard Hopkins
Cardiff, Wales, United Kingdom
(replace .nospam with .com in reply address)

...

I think that you missed the point. What is the use of all of this speed in a uni-processor PC, when doing real work, not just running benchmarks for bragging rights............. I am very surprised that you are not running an overclocked 3.0 or 3.2 GHz in your system to at least give you some extra benefit from your disk speed.

I do a lot of video editing of DV material and run large 7200/8Meg disks on a P4-3.06 HT. When I have some more$$$ to spend, I will upgrade the processor to the P4 3.2 Extreme edition and may add a few Gbyes of main memory, as my performance tooling shows the bottleneck to be in rendering --- by far. The disk performance does not feature AT ALL in the length of time to complete a job. This might be an extreme case, but I challenge you to take a look at the applications that you commonly run and measure the REAL benefit either in time saved or real-time performance in useful applications of your fast, small and expensive disks. You might be more than a little disappointed.

I am very familiar indeed with the design of large video servers. Having the fastest disks in this application is of paramount importance. SCSI is the norm. Striping is up to 8-way for both bandwidth and for overlaps to avoid interrupts due to thermal compensation. Parallel-processing using arrays of DSP chips and dedicated hardware is the norm, usually under the control of a PC core to provide control- and user-interfaces, diagnostics and network I/O.

John Lewis

I can't argue that the best value is with the 8MB cache 7200rpm drives. The Raptors are expensive about 3.25x the cost per Gigabyte and for many folks this is just not worth the extra cost. However, folks shell out lots of extra dollars to squeeze more out of there memory, or to get a few extra Mhz out of the CPU. Just notice how often when you are waiting on your computer that the disk light is on. I belive disk performance has been a limiting factor in system performance for far too long. The market has been focused on large and cheep drives and performance has taken a back-seat, except for the server market, where SCSI or Fibre Channel has a strong hold, but talk about expensive

...

TomG · Oct 8, 2003

Diskspeed32 does a good job of testing raid 0 configurations, personal opinion. I carry it on my server under /utilities.

Greg Jones · Oct 8, 2003

"John Lewis" wrote:

On Wed, 08 Oct 2003 02:50:10 GMT, "Greg Jones"
wrote:

Speaking personally though, I don't find the Raptor's high price per
gigabyte and small capacity an acceptable trade-off for its supposed
speed
benefits. It is matched in some benchmarks by the fastest 7200rpm drives, which are much cheaper and available in much higher capacities.
--

Richard Hopkins

Not to pick a fight or anything, but I've built a system with Dual Raptors using RAID 0 and it screams. The otherday I was running some MD5
checks over large files and was getting 110 MB/sec read operations.
Considering the PCI is limited to 132 MB/sec (theoritical) these
numbers are blazing...

Greg Jones

I think that you missed the point. What is the use of all of this speed in a uni-processor PC, when doing real work, not just running benchmarks for bragging rights............. I am very surprised that you are not running an overclocked 3.0 or 3.2 GHz in your ...

I've been overclocking the system to 3 Ghz and with
the new heatsink that just arrived, I'm hoping to get 3.2, so yes I plan on squeezing more out of the system. But for benchmarking the drives I stuck with the stock 2.4. I don't like to change too many things when doing testing.

I didn't miss the point, I was just saying as an owner of a dual raptor system I'm happy. But as you point out the value of faster I/O depends on your workload.

In your case the the DV processing is limited by rendering time so CPU is paramount. For lots of things like DV or Audio MP3 compression the CPU makes the biggest difference. So if you have $200 extra bucks, putting it in the CPU is the best choice.

For 3D games Graphics Cards make the biggest difference. So for the gamer the $200 is best spend on a high-end card. Although there is some hefty I/O when loading a game.

For me I do software development which involves compiles of 1000's of small files, so I tend to be more I/O bound.

Really depends on what you do with the box.
The original poster didn't say what he does with the box.

According to Storage Review
http://www.storagereview.com/articles/200003/20000313OSandBM_3.html the WinBench 99 DiskMark benchmark is a simulation of I/O done by office applications, so it does attempt to measure impact of disk performance on real-world applications. The Business benchmark uses Corel WordPerfect Suite 8, Lotus SmartSuite, and MS Office as the application base. The High-End tests uses AVS/Express, FrontPage, Microstation SE, Photoshop 4.0, Premiere 4.2, Sound Forge 4.0, and Visual C++. So I chose this benchmark, over Sisoft and IOMeter, because I belive these numbers do represent value to the typical PC user.

So no I'm not disappointed, I'm very please with
the dual raptors performance. The OS and Applications
load very very fast. Given how much the cost of
building a new high-end PC has dropped over the
past few years, I didn't mind putting my extra $200
into the raptors.

But I do agree for most users the 3.25x cost isn't
justified. As Richard said, you could buy 2
regular drives and use Raid 0 and that would
provide much of the same performance for
only 2x the cost -- atleast it would give you
similar bandwidth, but you'll still be stuck
at a 12+ ms Access time while the 10K
rpm of the raptor keeps that at 8-9ms.
If you're loading tons of small files that
access time is critical.

One thing I think we can all agree on is that its nice
that we have more choices today in disks than
ever before. For me I'm finally glad that someone
is looking at bringing a high performance disk
in the PC market place. For your editing, I'm
sure you are happy that we are looking at 200+ Gig
drives at less that 1$ per Gig....

Wayne Youngman · Oct 9, 2003

"TomG" wrote

Diskspeed32 does a good job of testing raid 0 configurations, personal opinion. I carry it on my server under /utilities.

Hi Tom,

downloading *DiskSpeed(New Version)32setup.zip* as I type. . .

Will test my new RAID-0 set-up, as according to Sandra my results are way low. .

cheers,

Courseyauto · Oct 9, 2003

Hi Tom,

downloading *DiskSpeed(New Version)32setup.zip* as I type. . .

Will test my new RAID-0 set-up, as according to Sandra my results are way low. .

Where can i download diskspeed?

TomG · Oct 9, 2003

they have their own web site out there or you can get it from my FTP server at the address below under /utilities.

Richard Hopkins · Oct 10, 2003

"Wayne Youngman" wrote:

downloading *DiskSpeed(New Version)32setup.zip* as I type. . .

Will test my new RAID-0 set-up, as according to Sandra my results are way low. .

You've got to be a bit careful when interpreting Sandra results Wayne, and especially when comparing them with other benchmark figures.

Sandra's file system test is a high level benchmark that writes and reads via the file system and partition table in use. This means that the results will be skewed by the file system and cluster size in use, as well as other parameters like the way the drive is partitioned and so-on.

What this means in practice is that if you benchmarked a disc that was nearly empty, you'd get better results than you would if the disc was nearly full. The reason being that the test data would be written at the beginning of the drive (which is usually faster) than at the end (which is usually the "slowest" bit). As an illustration of this, I benchmarked a freshly formatted 120GB RAID 0 array with Sandra earlier, and got a score in excess of 80MB/sec. I then put 110GB of data on it, defragged and benchmarked again, to find the score had dropped to 55MB/sec

Diskspeed 32 is a low level benchmark that reads directly off the cylinders. This gives it a degree of independence from file system related issues, although precisely because of this you could say it has less real-world relevance.

In practice we can learn a lot through both types of benchmark. Low level tests can show up differences in different drives, driver/BIOS revisions and RAID stripe sizes. High level marks show all of those *plus* the effects of different file system and disc usage configurations. For example Sandra is much better than Diskspeed32 for illustrating the performance differential between FAT32 and NTFS - DS32 wouldn't show any difference between the two, as it reads "underneath" the file system.

Also don't forget the effects that other system processes can have on benchmarks. If you've got a lot of bits 'n' pieces running in the background while benching, your figures won't necessarily be accurate.

Wayne Youngman · Oct 12, 2003

"Richard Hopkins" wrote
<snip>

You've got to be a bit careful when interpreting Sandra results Wayne, and especially when comparing them with other benchmark figures.

Also don't forget the effects that other system processes can have on benchmarks. If you've got a lot of bits 'n' pieces running in the background while benching, your figures won't necessarily be accurate.

Hi Richard,

thanks for information, good to know. In no way did I mean to imply I wasn't happy with my new SATA RAID-0 set-up, its ridiculously fast in real world use and has given me 0% problems to date (well maybe waiting for 240GB's to format, even in RAID 0 still took a while

). . .

I got that DiskSpeed32(new version) from TomG's FTP downloaded to my desktop since Friday, but I just got myself a new girlfriend so my PC time has been cut by 75%. I would be interested in testing out this DiskSpeed app and comparing its results with your RAID-0 set-up, I have a sneaky feeling that an INTEL set-up would perform RAID-0 better than the AMD Platform (nForce2). .

You or anyone else up for a little RAID-0 comparison

TomG · Oct 12, 2003

run Diskspeed32 and tell me what you got from it. Diskspeed32 will post some fabulous numbers at the beginning and trend downward as the test goes on and that is normal due to the decrease in read speed from the outside to the inside of the platters. even still, by the time it gets to the end, I was still impressed and it was better than the numbers that I used to get from my Highpoint raid 0 array.

pay attention to the numbers it posts initially as well and see what those are...

Wayne Youngman · Oct 12, 2003

"TomG" wrote

run Diskspeed32 and tell me what you got from it. Diskspeed32 will post some fabulous numbers at the beginning and trend downward as the test goes on and that is normal due to the decrease in read speed from the outside to the inside of the platters. even still, by the time it gets to the end, I was still impressed and it was better than the numbers that I used to get from my Highpoint raid 0 array.

pay attention to the numbers it posts initially as well and see what those are...

Hi,

ok thanks Tom, just installed it and am now running the full test, looks like it will take a while!. The machine I am running has been on for days and has been running loads of programs, although most are closed now, is there an ideal testing method with this? like shutdown or restart the PC before running?

Thanks,

Wayne Youngman · Oct 13, 2003

"TomG" wrote

run Diskspeed32 and tell me what you got from it. Diskspeed32 will post some fabulous numbers at the beginning and trend downward as the test goes on and that is normal due to the decrease in read speed from the outside to the inside of the platters. even still, by the time it gets to the end, I was still impressed and it was better than the numbers that I used to get from my Highpoint raid 0 array.

pay attention to the numbers it posts initially as well and see what those are...

Hi again,

well finished running Diskspeed32, but I'm not sure how to interpret the results. I took a screenshot of the program window once it was finished:

www.waynes.spamtrap.btinternet.co.uk/My_NF7S_RAID0_WDSE240GB.jpg

and here is the results file:
www.waynes.spamtrap.btinternet.co.uk/My_NF7S_RAID0_WDSE240GB.dat

I see one figure that means something, *burst-speed* = 108.2mbps ?? can that be right?

Rod Speed · Oct 13, 2003

Wayne Youngman wrote
in message

TomG wrote

run Diskspeed32 and tell me what you got from it. Diskspeed32 will
post some fabulous numbers at the beginning and trend downward
as the test goes on and that is normal due to the decrease in read
speed from the outside to the inside of the platters. even still, by
the time it gets to the end, I was still impressed and it was better
than the numbers that I used to get from my Highpoint raid 0 array.

pay attention to the numbers it posts
initially as well and see what those are...

well finished running Diskspeed32, but I'm not sure how to interpret the results. I took a screenshot of the program window once it was finished:

www.waynes.spamtrap.btinternet.co.uk/My_NF7S_RAID0_WDSE240GB.jpg

I'd toss them in the bin myself.

and here is the results file:
www.waynes.spamtrap.btinternet.co.uk/My_NF7S_RAID0_WDSE240GB.dat

I see one figure that means something, *burst-speed* = 108.2mbps ?? can that be right?

Nope, pure fiction.

Barton (AQXEA) XP2500+ @ 2.2GHz (10x220) - 1.75vCore
CoolerMaster Aero 7 Lite - 3,200rpm
ABIT NF7-S (v2.0 - BIOS#14)
512MB Dual TwiSTER PC3500 @ DDR440 1:1 (8,3,3,2.0 - 2.8v) Sapphire Atlantis 9800 - 3.3ns Samsung 240GB (2x120GB) WD-SE SATA RAID-0 (16k Stripe) Antec SX630II Mini-Tower Case Inc 300w PSU WinXP-PRO-SP1 Cat 3.7 - DX9.0b

Folkert Rienstra · Oct 14, 2003

Header says that there were 11 (11 !!!) reference posts, none of which show up here. Clicking below links show them like mail. What happened?

References:

"Wayne Youngman" wrote:

"TomG" wrote
run Diskspeed32 and tell me what you got from it. Diskspeed32 will post some fabulous numbers at the beginning and trend downward as the test goes on and that is normal due to the decrease in read speed from the outside to the inside of the platters. even still, by the time it gets to the end, I was still impressed and it was better than the numbers that I used to get from my Highpoint raid 0 array.

pay attention to the numbers it posts initially as well and see what those are...

Hi again,
...

Simon (17) · Oct 14, 2003

Folkert Rienstra wrote:

Header says that there were 11 (11 !!!) reference posts, none of which show up here. Clicking below links show them like mail. What happened?

It was a thread in alt.comp.periphs.mainboard.abit that had comp.sys.ibm.pc.hardware.storage added very late in the thread.

Even you should have been able to work that one out.

References:

...