Q2 2004 Desktop Hard Drive Comparison: WD Raptor vs the World
by Anand Lal Shimpi on June 7, 2004 12:05 AM EST- Posted in
- Storage
The biggest argument to why you would ever need a faster CPU is often that everything is disk limited anyways, so getting a faster CPU isn't the solution to your performance woes. While, to an extent, this is true, we have seen over the years that quite a few things are just as CPU bound as they are disk bound.
We've shown in the past that even very disk intensive operations, such as extracting a zip archive, can vary significantly with CPU. We've also shown that disk bound benchmarks, such as Winstone, can significantly reduce the performance benefit seen when upgrading to a faster CPU.
More than anything, what we've learned in the past is that there is no one component that significantly bottlenecks the system; rather, it's a combination of all of your components - your CPU, chipset, video card, memory and hard drive - that determine the performance of your system. While one component (e.g. your video card) may be the major determinant of performance in a particular application (e.g. a game), it's rare that the only applications you run are bound by a single component. To put this into perspective, would you ever not upgrade your CPU for a next generation game just because "everything is GPU bound to begin with"? Of course not. Take a 500MHz Athlon and pair it up with a X800 Pro and you'll realize quickly that this sort of logic won't work. So why, then, apply it to hard drives?
Luckily, the average AnandTech reader is smarter than that, and understands the importance of maintaining a balance of performance within his/her system. But here's where the problem resides: how do we measure hard drive performance?
Hard drives continue to be the only component where performance is measured using purely synthetic benchmarks. Our latest CPU review has no less than 12 real world application benchmarks to showcase the performance of the CPU. Our last GPU review took 13 games, and we benchmarked them to help you decide what video card would run games the fastest. But look around for hard drive reviews and you see a bunch of numbers that are, at best, great hypothetical indicators of performance or over-exaggerations of the impact of a particular hard drive. If we converted all of our CPU and GPU reviews to a similar set of synthetic benchmarks, you would quickly find a replacement site for your information, so why settle for the same treatment with hard drives?
We've tried numerous times in the past to bring hard disk reviews to AnandTech, but the limitations have seemingly been unsurmountable. At first, we couldn't get drives, then we had no good benchmarks; then, we got drives and benchmarks, but had no time to tackle the testing. Finally, we are able to offer a good test suite, review the drives that you want reviewed, and do so on a regular basis.
In order to kick off our new suite of drive benchmarks and our return to hard drive reviews, we figured that we'd focus on one of the hot-topic drives as of late: the 2nd generation Western Digital Raptor.
Facebook
Twitter
RSS
50 Comments
View All Comments
SignalPST - Monday, June 7, 2004 - link
Great review, I must say. But one thing that would certainly interest alot of people including myself would be using RAID. We know that using multiple hard drives in a RAID array is very popular among gamers and almost every motherboard out now supports RAID as well. I'm sure it'll be quite interesting to see 4 of 74GB Raptors in RAID 0 in future reviews! It would also be interesting to see the different effects of stripe sizes configurations.Doormat - Monday, June 7, 2004 - link
What about putting some meta-data in there? Like current street price, length of warranty, etc. Also temperature would be nice.I'd kinda like to see some RAID tests too, I'm looking at RAID 5 for a bunch big drives for a video on demand system.
Speaking of, a big-drive comparison would be cool too. Where's that hitachi 400GB drive they announced a while ago?
Murst - Monday, June 7, 2004 - link
Hmm,well, I'm pretty sure that there should be a significant difference in system performance when your system runs out of RAM. When virtual memory takes over, I have seen the performance of my computer drop significantly. I was hoping that a benchmark could be made showing just how large of a difference could be seen when virtual memory is a significant source of data for program execution. There should be a noticable difference in this performance between different drives.
Also, it would be interesting to know if the file system on a drive makes a difference in performance. I have a feeling that if it does, it would be unnoticable, but nevertheless, unless its tested, we would never know for sure. I don't neccessairly mean the type of file system, either. Just as RAM can have different latency settings, so can a hard drive have different block sizes (and optimal block sizes).
Again, I'm not positive if this would make a difference in performance, but I'm just trying to think of practical tests for hard drive performance.
Anand Lal Shimpi - Monday, June 7, 2004 - link
Murst,Sorry, the last response must've been posted at the same time as yours :)
Both the Winstone and SYSMark tests use multiple applications running at the same time, but I do understand the point you're trying to make.
We do have a synthetic test that shows the benefit of defragging a hard drive, but I have yet to do significant investigation in to how that affects performance between drives other than it reduces it.
Take care,
Anand
Anand Lal Shimpi - Monday, June 7, 2004 - link
Must,The drives being tested are secondary drives only for the game loading tests and the theoretical IPEAK tests. The remaining Winstone and SYSMark tests all use the drive as the only drive in the system.
Take care,
Anand
Murst - Monday, June 7, 2004 - link
Hmm... let me clarify that. I believe that all of your tests were probably ran with no other programs executing. It would be interesting to see the difference in performance when a lot of page swapping is occurring (ie, fill up the page table by executing other programs and then run a benchmark).Oh, and I just thought of another issue... why not have a benchmark which evaluates a drop in performance of a drive with data that is, say, 60% fragmented as compared to mostly unfragmented data.
Murst - Monday, June 7, 2004 - link
Its nice to see a hard drive comparison. I will be building a new comp soon and I always wondered if I'd see a difference between drives.I do, however, have one concern. It seems like the drives you used were secondary drives in the system, with the operating system working off a different drive. I have always assumed that the largest benefit of choosing a very fast drive was to minimize the access and read times of a page fault (as I generally do not spend much time at all waiting for something to load). It would seem that none of your tests take this into consideration.
Thanks
Anand Lal Shimpi - Monday, June 7, 2004 - link
#1 - You're correct, the mentions of command queuing were leftover from some early tests on a new SATA controller with support for the feature. Those tests didn't make it into the article, and I've updated it accordingly.Take care,
Anand
jliechty - Monday, June 7, 2004 - link
#1 - I was under that assumption also. I do recall hearing of a controller chip that supported TCQ being in the works (or perhaps already available), but the question remains whether that chip has been put in any controllers that are on the market at this time?Anyway, I'm glad that my preciousss... er... my Raptor didn't do too badly, though for what I do I probably could have kept my old WD Caviar Special Edition and not noticed much of a difference, except for my wallet being heavier. :-(
RyanVM - Monday, June 7, 2004 - link
Doesn't the WD74GD require a controller which supports command queuing in order for that feature to be of actual use? And I was under the impression that no current SATA controllers support that function.