Serial ATA arrives...finally!

[Jason Dunn]

<a href="http://www.pcworld.com/news/article/0,aid,103268,tk,dn073002X,00.asp">http://www.pcworld.com/news/article/0,aid,103268,tk,dn073002X,00.asp</a><br /><br />I've been waiting for Serial ATA for quite some time, so this is an exciting announcement! What's still missing, however, is the increase in data throughput from the hard drives themselves. We now have no bottleneck from the interface, but we need hard drives capable of spitting up more data. Between this announcement and the forthcoming processors from AMD (Clawhammer, Opteron) and Intel (3 Ghz before Christmas), this will be a good year to upgrade. Start saving that loose change!<br /><br />"Enter Serial ATA, the successor technology we examined on a prototype hard drive from Seagate. Debuting with a speed of 150 MBps and having the potential to scale up to 600 MBps, Serial ATA provides the headroom needed for faster drives in the future. And because it uses 3-volt (instead of 5-volt) signals, it will work better with the low-power motherboard chip sets and CPUs coming soon. Unlike parallel ATA drives, the serial model we installed in our Windows 2000 test system required no master or slave designation. We simply plugged one end of the 0.25-inch cable into the drive and the other into a supplied Silicon Image Serial ATA card that we plugged into a PCI slot. (The thin cable also permits better airflow and more-compact system designs.) <br /><br />Seagate will offer its Serial ATA drive this fall. Maxtor and Western Digital promise to provide the interface on some drives by year's end--about when Via expects to integrate it into PC chip sets. Intel plans to follow suit in 2003. Users also have new capacities to look forward to, thanks to an increase in drive-platter areal density from 40GB to 60GB. With two platters, Seagate's Barracuda ATA V will hold 120GB. Western Digital's new Caviar model should reach 200GB. Later this fall, look for both Maxtor and IBM to launch their own ultrahigh-capacity products as well."

[klinux]

CNET just had an article in the last few days about HD being a commodity item. Their study showed that there is no real world difference between ATA/133 and ATA/100 or HD with a 8mb cache vs 2mb cache. [I will try to find a link for it.]

I love all hardware improvements but I just hope this translates to real world performance!

[Jason Dunn]

Their study showed that there is no real world difference between ATA/133 and ATA/100 or HD with a 8mb cache vs 2mb cache

I would believe the ATA133 vs. ATA100, but the 8 mb cache makes a HUGE difference in what I do - I noticed an immediate performance gain. I find that very, very hard to believe...

[Will T Smith]

ATA100 is great. ATA133 is even better ... Right? Wrong!!!!

In fact, ATA133 exceeds the bandwidth capacity of PCI bus. In addition, HD bandwidth must contend for bandwidth with every other component that uses the PCI bus. This further degrades the performance of ALL components.

Hope is on the horizon. AMD and Nvidia currently are shipping products with HyperTransport bus. Not only is it WAY, WAY faster ... it also includes mechanisms for priority so that high latency I/O can get priority over say, your Video card.

The nForce boards have the capacity to utilize ATA133 at full bandwidth. Unfortunately, the new nForce chipset will NOT ship with integrated Serial ATA. One cannot get the bandwidth advantage by plugging an adapter into a PCI expansion slot. If one had a PCI-X bus, then you could crank up Serial ATA to it's full potential.

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

Beyond that, I hate configuring IDE devices. It is the BIGGEST pain in the ass. An even BIGGER pain is dealing with that stupid ribbon cable and trying to Origami a secondary drive onto it.

IDE had it's uses. I certainly would have rather seen SCSI become dominant. That wasn't in the cards though. One thing is for sure though, IDE will NOT be missed.

[sundown]

I heard a rumor that these serial ATA drives will at first be dual format, meaning you could use them even if you have a board that doesn't yet suport serial ATA by usually your old IDE cables and then upgrade when the boards come out. Anyone know if that's true?

[blade_of_narsil]

From somebody that knows, and isn't a AMD fanboy who like to bask Intel, does pci express hold any advantage over hypertransport. And what devices are going to use a 2.5Gb per channel bandwith?
:?

[klinux]

I am glad the 8MB cache version of HD showed real improvement for you, Jason but here is the "Behind the numbers" article fromCNET (http://www.cnet.com/techtrends/0-6014-8-20069949-1.html) where they said the 8mb version has ""no discernible boost over similar 2MB drives in desktop applications."

Their summary, forget specs and try to get more gb for your buck. I would append that by checking consumer experience who found that IBM "Deathstar" HDs are less reliable than others.

[Rob Alexander]

I am glad the 8MB cache version of HD showed real improvement for you, Jason but here is the "Behind the numbers" article fromCNET (http://www.cnet.com/techtrends/0-6014-8-20069949-1.html) where they said the 8mb version has ""no discernible boost over similar 2MB drives in desktop applications."

Their summary, forget specs and try to get more gb for your buck. I would append that by checking consumer experience who found that IBM "Deathstar" HDs are less reliable than others.

Their article is probably good advice for most people. Most users don't stress their systems and would probably see pretty much the same results as CNet. These folks should probably spend their money on size rather than performance. And that's what makes it good advice. If you don't know why their analysis is flawed, then you should probably follow their advice. If your needs exceed the average, you will probably immediately recognize why this doesn't apply to you and will know to use your own experience (instead of theirs) to guide you. You'll know to ignore them because you'll already understand that the potential impact of these various performance parameters are completely dependent on the use to which you put your computer.

Look, it's not unlike how these types of articles they typically tell you to buy 128-256 MB of RAM, exaplaining how more RAM than that won't speed up your system. Well with their business benchmarks, it won't because their benchmarks simply don't use more than 256MB of RAM, but when I have half a dozen 40MB TIFF images open in Photoshop at once, having 512MB in the system most certainly DOES make a difference. You can point me to all the articles you want, but it won't change the fact that MY system runs faster with 512MB than with 256MB because of what I do with it.

The HDD cache issue is similar. Clearly the CNet 'business' tests did not ask the hard drive to reuse more than 2MB of the same information during the test. If repetatively-requested data during the test is less than 2MB in size, then of course they won't see any speed difference between a 2MB and an 8MB cache. But run a large database application that pulls a large amount (>2MB) of the same information from the disk over and over and it simply must make a difference. Jason is probably seeing exactly this when he notes the significant speed increases he sees with the larger caches.

It's the same with CPU caching. Business benchmarks have always shown only a small difference between a Pentium &lt;x> and its corresponding Celeron with less cache. But for the types of mathemtical models we run here, the difference is significant (as in a model running in 1/3 the time with the same clock speed, but a larger cache). That's simply because the nature of the modelling we do is particularly senstitive to the kind of help a cache can give. The same information is swapped in and out of RAM over and over to build large matrices, which are then submitted to nonlinear solution algorithms, which themselves perform the same operations over and over until an optimum is found. It's exactly what caches are for.

The article was also disappointing in missing the whole point of when you would prefer SCSI to IDE. Their entire discussion of SCSI vs IDE was based on single drives. Anyone who's ever paid the extra money for SCSI knows they didn't do it because they thought a single drive was quicker, but because a series of SCSI drives is collectively quicker than any combination of IDE drives. They only get close to accurate when they say "SCSI drives outperform EIDE drives because the interface handles simultaneous requests from multiple users more efficiently". Well, yes, but not with a single SCSI drive (who's heads can still only be in one place at a time), and it's not just about multiple users, but also about multiple tasks from a single user.

The bottom line is that, if you're running mostly Word and Excel, you're probably not going to notice any difference in one drive versus another, and CNEt's advice is good. But if you stress your HDD subsystem in large database applications, digital video work, some sound recording/mixing work, and perhaps certain games, then you start to hit up against these performance barriers and you do see a difference.

For me, I feel these limits when I capture and render digital video, despite having the fastest IDE technology currently available. If serial ATA will give me higher sustained throughput, then bring it on!

[klinux]

Hey Rob:

I am not dismissing your or Jason's real world experience with your HD perfomance.

And if one is a gamer, a graphics professional, or someone who has specific needs, you already know what your needs are and you do not need an article from CNET to tell you that. As for running database application, I am a datawarehousing professional working with multiple databases with each having multiple million row fact tables. It goes without saying we do not use EIDE drives, 8MB cache or not! :D

I thinking ultimately we do agree on the same thing. If your work/needs are limited by HD performance, you already know where the problem is and what the soluton may be. But for 99.9% of the computer users out there, their bottleneck is not their HD and thus the CNET article holds true.

[st63z]

I'm going to have to brush up on Serial ATA again. If I recall correctly, doesn't it natively support hot-plugging? Personally for me, this will be a great boon, not necessarily the faster interface speed (of which current HDDs for the foreseeable future can't take advantage of anyways)...

Now I'll bug Addonics to see what SATA solutions they'll be bringing out :D

P.S. Will got me thinking, with nVidia's performance-minded designs and existing dual-channel DDR, why didn't they come out with dual PCI busses w/ a bridge. Even standard 32-bit/33MHz PCI busses would be OK, not 64-bit or 66MHz or PCI-X or even 3GIO...