Sign in to follow this  
Davin

Western Digital Raptor Preview

Recommended Posts

Few drives have generated as much buzz as Western Digital's upcoming Raptor, the first ATA drive to feature a 10,000 RPM spindle speed. Long-time SR sponsor Hyper Microsystems has made it possible to put an early pre-production Raptor through our standard battery of tests.

When reading the preview and perusing the figures, please remember that these results have been culled from a pre-production drive. Performance figures on the final shipping product may differ substantially. Though SR has traditionally refrained from presenting pre-production figures, HyperMicro's gracious gesture combined with the SR readership's insatiable curiousity has pushed this preview forward.

Western Digital Raptor Preview

Share this post


Link to post
Share on other sites

From the performance leader (supposedly, I say IBM, anyway) we see the first 10,000.00 RPM drive in PATA. What a disappointment! I thought maybe WD could have put a higher areal density. I really believe the if they put a higher areal density it WOULD yield a better all-around score! I hope that WDC can make it better. Though, I think that WDC really CANNOT make it any better. UNLESS THEY TWEAK THEIR FIRMWARE, LIKE THEY ALWAYS DO TO PRODUCE BETTER THAN AVERAGE SCORES! If maxtor or even seagate try to oust WD, they will fail. Maybe HITACHI can do something with its long history of drive mechanical advantage (thanks to IBM).

Thx for the great review, but I really think that when they do finally produce the real unit, any improvement will be the result of FIRMWARE TWEAKING, not real physical change.

I really hope I AM WRONG!!!...

thx mann

Share this post


Link to post
Share on other sites
From the performance leader (supposedly, I say IBM, anyway) we see the first 10,000.00 RPM drive in PATA.  What a disappointment! I thought maybe WD could have put a higher areal density.  I really believe the if they put a higher areal density it WOULD yield a better all-around score! I hope that WDC can make it better.  Though, I think that WDC really CANNOT make it any better.  UNLESS THEY TWEAK THEIR FIRMWARE, LIKE THEY ALWAYS DO TO PRODUCE BETTER THAN AVERAGE SCORES! If maxtor or even seagate try to oust WD, they will fail.  Maybe HITACHI can do something with its long history of drive mechanical advantage (thanks to IBM).

I can't help but think this stands in direct contrast to previous posts where you asserted the supremacy of access times.

I really hope I AM WRONG!!!...

The odds are excellent.

Share this post


Link to post
Share on other sites

Very interesting (albeit somewhat disappointing) results. I suppose the big question is what will the difference be when the final version comes out for review? I would imagine mature code would have a big impact.

Share this post


Link to post
Share on other sites

access time is very important, but remember that areal density does do a lot too. WDC does not understand how to make 10,000 rom drives. They got out of that business a long time ago. I believe that the companies that still produce high rpm drives may make a better after many trial and error models.

And please cleaify what YOU feel I may be wrong about, do tell.

thx, mann

Share this post


Link to post
Share on other sites
but remember that areal density does do a lot too.  WDC does not understand how to make 10,000 rom drives.  They got out of that business a long time ago.  I believe that the companies that still produce high rpm drives may make a better after many trial and error models.

36gb/platters is the norm for the current generation of 10k drives...

Share this post


Link to post
Share on other sites
but remember that areal density does do a lot too.  WDC does not understand how to make 10,000 rom drives.  They got out of that business a long time ago.  I believe that the companies that still produce high rpm drives may make a better after many trial and error models.

36gb/platters is the norm for the current generation of 10k drives...

That's right, but we're up to 83gb per platter on 7200 rpm drives - that damn near alleviates (sp?) the issue as can be clearly seen from the benchmarks.

Obviously the ideal situation is a single 83gb platter (although they need it slightly smaller due to 10,000 rpm, so perhaps 60gb per platter?) @ 10,000 rpm.

I must say the noise (whine) annoys me - I can't STAND that whine! early 7200's had it also. - but the seeks being quieter than the 200gb model is impressive as was the heat.

All in all, it's a disapointment, the least it could have done was beat all other ide drives outright, then it would be a sure fire purchase for at least SOME enthusiasts.

Sorry WD - go back and do one with a much higher arial density (sp?)

I hope this "failure" doesn't turn them off though.

As for people nitpicking my post and saying "hey firmware might improve it" - sure if so, great - however I'm far far more skeptical the firmware can do much for it - it seems to be lacking in pure data xfer speeds - I'd say that's not a firmware issue.

Share this post


Link to post
Share on other sites

The drive is far from a "failure". It is, perhaps, the fastest (overall) ATA drive ever made. It's not quite as fast as current 10K SCSI offerings, but then again, WD has not had the benefit of generations of this model to perfect the firmware. In time, I think the drive will prove to be solid. You know, based on the specs, it seems to be about as fast as my Maxtor Atlas 10K III, which I'm still quite happy with.

Share this post


Link to post
Share on other sites
The drive is far from a "failure".  It is, perhaps, the fastest (overall) ATA drive ever made.  It's not quite as fast as current 10K SCSI offerings, but then again, WD has not had the benefit of generations of this model to perfect the firmware.  In time, I think the drive will prove to be solid.  You know, based on the specs, it seems to be about as fast as my Maxtor Atlas 10K III, which I'm still quite happy with.

I think we don't share the same thoughts on the benchmarks then - I definately feel it's too slow considerign what it could and or should be.

Share this post


Link to post
Share on other sites

Okay. If you were looking for the ultimate desktop drive, then I can see why you'd be disappointed. I think the target market this drive is trying to hit is entry-level servers where cost and reliability are the main decision points. Decent performance is important too, but it's not the highest concern. You don't have to be the fastest drive on the market to make a killing in that arena. As long as WD can offer SCSI-quality drives at ATA prices, they'll sell a ton (just not to the high-performance enthusiast market).

Share this post


Link to post
Share on other sites

What happened here? Ouch. I wasn't exactly expecting Atlas 10K IV performance, but I was expecting more than this. I knew the access time would be a bit off, but what happened to the STR? The top 3 SCSI 10k drives are within 1.9MB/s of each other, how does this drive with the same spinrate and same platter density, get bombed so badly? I guess the track density was increased vs SCSI at the expense of sector density. Is there any technical or cost reason for them to do that? Considering the sizable access time advantage this drive has over other ATA drives, the server benchmark are pretty mediocre as well. Its results still put it closer to current gen ATA than current gen SCSI. About the only things this drive has going for it are the noise and heat measurements, which would make it a better consumer drive than the enterprise market WD is trying to target. I don't think the potential 30% price reduction vs SCSI is big enough to make this drive attractive to people looking for high performance storage options.

Share this post


Link to post
Share on other sites
Obviously the ideal situation is a single 83gb platter (although they need it slightly smaller due to 10,000 rpm, so perhaps 60gb per platter?) @ 10,000 rpm.

obviously there are also technical reasons as to why the areal densities on 10k drives are the way they are.

WD did a darned good job, IMO in terms of areal density on the raptor.

Share this post


Link to post
Share on other sites
access time is very important, but remember that areal density does do a lot too.  WDC does not understand how to make 10,000 rom drives.
And you do?
I really believe the if they put a higher areal density it WOULD yield a better all-around score!
And IBM Storage, the company you have supported beyond all rational -- even during the 75GXP fiasco, has implimented higher density platters? Think before you type. Obviously a denser platter would yield some performance improvement, but platter density has to be sacrificed when you step up the rotational speed. It would not be a good idea for WD to test a new, unconventional, untested platter technology on a drive which is already largely based on new and untested technologies.
They got out of that business a long time ago.  I believe that the companies that still produce high rpm drives may make a better after many trial and error models.
From the performance leader (supposedly, I say IBM, anyway)
While trial and error (and time, and advancing technology, and many other factors) do tend to lead to higher performance, keep in mind that IBM, presumably your favorite company, frequently makes the loudest, slowest, hottest server drives on the market. In any case, Western Digital did make enterprise SCSI drives at one time. I doubt their engineers have died or forgotten all they had learned in. They may have been planning a move like this for some time.

As the preview emphasized numerous times, this is a pre-release drive and the performance results may change. Significantly.

Even if they do not, this is the fastest IDE drive for servers ever made. Not bad for a first generation serial ATA drive, running on a first generation serial ATA controller, with a first generation 10K design from a company that, as you stated, hasn't made server drives for several years.

Share this post


Link to post
Share on other sites

I have to say I'm disappointed by the STR-figures, and well with many other benchmark results, but since this IS a preproduction model and THE WORLD'S FIRST ATA 10kRPM drive, let's give WD a break here! I really hope they will release a 2-platter version.

I'm surprised about the idle whine (which I really can't stand in hard drives) since the drive has so low temperature and sound figures otherwise, and it only has one platter.

What I'd draw from these results is that they have taken their old SCSI WD Enterprise 10k drive and fine-tuned it with a higher-density platter. If we could retest an old WD Enterprise, I'd say its results wouldn't be far off in Storagereview's tests compared with the WD Raptor, I'd guess mechanically they are close.

Well, if they manage to improve STR and areal density as well as double the capacity, then I think we would have a winner here. With this package my feelings are a little mixed.

A great thanks to SR for the early review! I rather see a review sooner than later, so I think it's great you reviewed a beta sample, regardless of how off the figures might be from the final product.

Share this post


Link to post
Share on other sites

My fellow SR members,

You are obviously forgetting about the huge capacity disadvantage the WD Raptor is handicapped with in the desktop testing. SR's Desktop Performance suite tests a fixed amount of sectors, giving drives with larger capacity a decided advantage over smaller ones since the heads do not have to travel over as much of the platter as with smaller drives.

If some of the veterans remeber, I was adamant that capacity could be the "great equalizer" among drives -- so much so that I proposed that a 75GXP short-stroked to 18 GB could take on an 18 GB Cheetah in the benchmarks and possibly win. In some testing Buck did (I believe it was Buck... if not, I am sorry... the MBF trashed many useful threads, including this one), he reported IOMeter results that showed every doubling of capacity (given otherwise identical drives) could yield an almost 10% improvement in performance. So, while I definitely overestimated the impact capacity had on performance, Buck and I gleaned some useful insight on the extent to which capacity affects performance -- and 10% is a significant amount, especially when you are dealing with such a disparity in the capacity of tested drives.

So, if you take that 10% rule and apply it to the desktop performance suite results, you would see that...

If the Raptor were 146 GB, it would have 20% higher performance

If the Raptor were 200 GB, it would have 28% higher performance

And the SR Desktop Performance results might look like this:

raptor_scaled.gif

Now this 10% rule will probably receive a lot of scrutiny -- and rightfully so -- but focus not on the accuracy of the 10%... maybe it's closer to 7%... whatever. My contention is that if the Raptor had capacity anywhere close to that of its competitors in a capacity dependent benchmark, it would had put forth a very competitive showing. Pound for pound, gigabyte for gigabyte -- WD should be proud of their midget on the desktop.

Okay, why am I hammering this point home? Because I bet most of you can put your OS and all the apps you use 80% of the time into the first 10 GB of your drives. And I bet most of you who use SCSI are running 18 or 36 GB boot/main drives. That puts the 36 GB Raptor into direct competition with what you are buying. The performance of 146 GB SCSI drives are NOT representative of what you are buying. Most of you do not buy those massive drives. What if I scaled the performance of the Atlas 10k IV and Cheetah 10k6 back to that of a 36 GB capacity model? They'd fare no better than the WD Raptor. If you take this into consideration, the Raptor should be strongly considered as one of your possible alternatives on the desktop.

----------------

Now, as for the lackluster showing in the server benchmarks... as I understand it, IOMeter results are tested on full unpartitioned drives (thereby rendering capacity insignificant), so the Raptor's poor showing there cannot be excused.

Share this post


Link to post
Share on other sites

A very interesting drive. But I wonder if the pricing will be in line with 7200 rpm drives. The idea that it is best suited for use in servers leads me to think there may not be much, if any, of a cost saving over scsi with this drive. If so I see no reason to drop scsi just yet.

Share this post


Link to post
Share on other sites
Okay, why am I hammering this point home? Because I bet most of you can put your OS and all the apps you use 80% of the time into the first 10 GB of your drives. And I bet most of you who use SCSI are running 18 or 36 GB boot/main drives. That puts the 36 GB Raptor into direct competition with what you are buying. The performance of 146 GB SCSI drives are NOT representative of what you are buying. Most of you do not buy those massive drives. What if I scaled the performance of the Atlas 10k IV and Cheetah 10k6 back to that of a 36 GB capacity model? They'd fare no better than the WD Raptor. If you take this into consideration, the Raptor should be strongly considered as one of your possible alternatives on the desktop.

Maybe SR should test this; compare all different capacities of, let's say Seagate's 10k.6?

Jan

Share this post


Link to post
Share on other sites

What I'm surprised is that the Raptor shows very poor performance in the High-end test, which is one of the tests I look at the most. Even if you take into account the capacity difference, the Raptor is beaten in this respective benchmark.

Share this post


Link to post
Share on other sites

In light of this "challenge", Jan, I should bring up a factor that may affect such testing:

Extra platters and heads make the head stack and actuator heavier, sometimes penalizing larger capacity drives with slower seek performance (and giving smaller drives with fewer platters/heads an edge). So, it is not as clear cut as I originally made it out to be -- this 10% rule only holds if seek performance does not vary by platter count in the same drive family.

Share this post


Link to post
Share on other sites
What I'm surprised is that the Raptor shows very poor performance in the High-end test, which is one of the tests I look at the most. Even if you take into account the capacity difference, the Raptor is beaten in this respective benchmark.

High-end test scenarios usually involve a lot of large sequential reads. Is this the type of usage you have planned for the Raptor? Do you do a lot of A/V editing? If so, you would be better off with striped 7200 rpm IDE drives. In any event, the Raptor's high-end result is hampered by its below-average STR (since the test is done at the beginning of the disk, it's beginning STR that counts). I expect WD to improve the firmware in the next month before release, so if you take capacity into account, it may still not be what you expected, but I think it will at least be in the ballpark.

Share this post


Link to post
Share on other sites
In light of this "challenge", Jan, I should bring up a factor that may affect such testing:

Extra platters and heads make the head stack and actuator heavier, sometimes penalizing larger capacity drives with slower seek performance (and giving smaller drives with fewer platters/heads an edge). So, it is not as clear cut as I originally made it out to be -- this 10% rule only holds if seek performance does not vary by platter count in the same drive family.

Um, then test matching capacities?

Jan

Share this post


Link to post
Share on other sites
Now this 10% rule will probably receive a lot of scrutiny -- and rightfully so -- but focus not on the accuracy of the 10%... maybe it's closer to 7%... whatever. My contention is that if the Raptor had capacity anywhere close to that of its competitors in a capacity dependent benchmark, it would had put forth a very competitive showing. Pound for pound, gigabyte for gigabyte -- WD should be proud of their midget on the desktop.

A couple of comments about this "10% rule" you have developed. Have you thought about the fact that since different tasks emphasize different aspects of drive performance, any "10%" rule cannot be universally applied across applications or benchmarks?

Also, by your logic, a WD800JB should have benches more than 12.5% worse than the WD2000JB -- due to both increased platter density (resulting in higher STR) as well as a larger size. However, if you look at SR's database, it is clear that the difference is rarely more than 12.5%, and often much less. For example, the difference in HE Diskmark is negligible, despite the lower STR of the 800JB. In addition, the ServerSuite benchmarks clearly display parity between the drives (though admittedly, the 2000JB has a slower seek).

I'm curious about something; seeing as the difference between the two WDs is mostly within 15% or so, do you think that size increases in a drive actually have significantly more to do with the increase in overall drive performance than any change in "metrics?" It would be interesting to benchmark an 800JB with identical platter size to the 2000JB to find a better answer.

Share this post


Link to post
Share on other sites
I don't think the potential 30% price reduction vs SCSI is big enough to make this drive attractive to people looking for high performance storage options.

Remember the cost of a SCSI controller...

Share this post


Link to post
Share on other sites
In some testing Buck did (I believe it was Buck... if not, I am sorry... the MBF trashed many useful threads, including this one), he reported IOMeter results that showed every doubling of capacity (given otherwise identical drives) could yield an almost 10% improvement in performance. So, while I definitely overestimated the impact capacity had on performance, Buck and I gleaned some useful insight on the extent to which capacity affects performance -- and 10% is a significant amount, especially when you are dealing with such a disparity in the capacity of tested drives.

Its somewhat difficult to perform short-stroke IOMeter tests on a drive without a recording-playback tool like IPEAK SPT. Without it, one must resort to partitioning and using formatted drives, a procedure which intel itself says yields erratic results. Apologies to Buck if he did indeed perform such tests... I did too though, you can find the results here:

http://www.storagereview.com/articles/2001...ort_stroke.html

So, if you take that 10% rule and apply it to the desktop performance suite results, you would see that...

I'm not 100% sure that it applies to desktop performance. I haven't run any formal tests, but some informal experiments between differing capacities within the same family hint at smaller (though still distinctly measurable) differences. This is likely due to the higher preponderance of short-distance seeks that occur with localized single-user patterns.

Now, as for the lackluster showing in the server benchmarks... as I understand it, IOMeter results are tested on full unpartitioned drives (thereby rendering capacity insignificant), so the Raptor's poor showing there cannot be excused.

Yep, for those subsequent posters who questioned this, running IOMeter natively without any IPEAK intervention negates any capacity advantages. In fact, one of the great internal debates we had in the deployment of TB3 was whether to run IOMeter straight or to run captured traces... the former would not take capacity into account while the latter would.

If anything, a smaller drive should theoretically enjoy a slight advantage on native IOMeter tests due to its lighter actuator assembly. At any rate, WD indicated that, at least with lower I/O depths, that the Raptor should be quite competitive with current 10k RPM drives... which is why we're holding out hope for substantial improvements.

Share this post


Link to post
Share on other sites
SR's Desktop Performance suite tests a fixed amount of sectors, giving drives with larger capacity a decided advantage over smaller ones since the heads do not have to travel over as much of the platter as with smaller drives.

Your 10% theory does not hold water if you look at the results of the DM+9. The 60GB/platter version and the 68GB/platter version use the same number of surfaces but the 68GB version is short stroked a bit more than the 60GB version to compensate for the differences and achieve the same 160GB capacity. By your theory the 68GB drive should win every desktop benchmark since it has a shorter platter surface to cover, yet if you look at the benchmarks, the 60GB wins all 4.

The current SR testbed does not have any drive with multiple capacities, but the old one had a 45GB and 75GB 75GXP. If you look at the I/O meter results, the 45GB version swept the 3 indicies. So adding platters does not increase the server performance which the drive is supposed to be targetted at.

Your 10% theory doesn't explain what happened with the STR of this drive either.

The Raptor is just not what most of us expected. We can devise all the theories we want as to why, but theories don't improve benchmarks, so I'll stick with SCSI for my OS/apps drive for now.

Remember the cost of a SCSI controller...

Most people don't have a SATA controller now either. And for those of us using SCSI now, it's an expense that has already been covered. If someone released an ATA drive that could compete with SCSI, I would consider dumping SCSI, but this drive certainly isn't it.

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
Sign in to follow this