Slashdot Mirror
Serial ATA and AGP 8X motherboards
bjschrock writes "Tech-Junkie reports that Asus is rolling out new motherboards with the new Serial ATA interface, along with AGP 8X support. Serial ATA will soon become pretty popular with the release of new hardware like the Seagate Baracudda ATA V hard drive, that sports a 8MB cache. The main advantage of Serial ATA, besides a slight speed increase, is the much smaller cable and the ability to hot-swap."
You've got that right about "hot-swap". Those hard drives become pretty hot in there.
With all the improvements happening in IDE world, along with USB 2, Firewire etc.. whats happening with SCSI ?
I'm probably not aware of anything past SCSI 3, since I can't afford it.. but what kind of improvements are in the pipeline ?.
the right place is to point to ST3120023AS and not ST3120023A
I can see no reason for 10,000RPM and 15,000RPM drives to be SCSI-only anymore. consumer technologies like ATA133 or SerialATA are giving consumer drives bandwidth that they can't hope to consume. Do these 10K and 15K RPM drives really need a SCSI connection? What's the point of pushing faster and faster consumer bus connections if manufacturers are unwilling to take advantage of them with faster drives.
Regards, Guspaz.
Wow--A serial drive! is it true the the project's code name was Commdore 1541? :)
the IBM hot swapping you are talking about is supported mostly by software drivers -- i.e., the hardware does it, but it doesn't break your running software because there is a whole bunch of fancy drivers going on under the covers. i'll have to admit, it WAS neat the first time i hot-swapped a PCI card...
:) in general, serial ATA hot swapping will look a lot like USB.
the new serial ATA standard hot-swapping is also driver-supported, but the primary difference is that the hardware is much simpler, thus it is cheaper to build and design than a big IBM server. also, serial ATA will probably not include power supplies
MORTAR COMBAT!
MORTAR COMBAT!
USB is a crap interface, with all of the transactions going through programmed IO. The reason it is popular is because it is cheap.
Firewire and SerialATA are much smarter and can read/write blocks/to from memory without having to go through the CPU. Thus they are much faster, but a little more expensive to implement.
Remember, You are unique...just like everyone else.
I may be wrong, but wasn't one of the advantages of Serial ATA the fact that each device had a dedicated channel, meaning it got the full 100?MB of bandwidth -- as opposed to the current IDE archetecture where the slave drive gets less bandwidth then the master, and only 1 device per channel can be used at a time.
If you chain the devices together, you're defeating what I understand the whole purpose of the technology is--not only that, but there aren't really enough wires for a second or higher device, are there? I'd think it would run into data transmission problems.
25 June 2002
PC World
Seagate is demonstrating its first Serial ATA hard drive at PC Expo/TechXNY with the help of a prototype Intel motherboard, and promises to be among the first hard drive makers to deliver the new technology, in products this fall.
The technology demonstration comes just one day after Seagate announced another first: 60GB-per-platter hard drive technology. Barracuda ATA V 7200-rpm drives using the new 60GB platters will arrive in retail outlets by August, say company executives.
Slashdot: come for the pedantry, stay for the condescension.
That would make sense. I should have included an IIRC, but I swear in the press release that I read, you could have the extra devices. However that could be wrong, I followed links around the aforementioned article, and the MSI board with the new KT400 chipset has 12 serial ATA connectors. It is possible I misread the press release and thought that you got more devices by chaining them together....oh well
--fetch daddy's blue fright wig, i must be handsome when i release my rage
Yes, Serial ATA has one drive per channel. I think most controllers come with at least 4 channels.
a board with serial ATA....guess it's time to replace the 'ole AMD 450.
a llery2.htm
Here's some close-ups:
http://www.ocworkbench.com/2002/asus/p4s8x/p4s8xg
I read somewhere about a company trying to develop a fiberoptic system to replace the wires. Heat was proving to be much lower & speed was much higher using light waves. Anybody know where that article is? It's been a few months....
Or maybe I read it in Scientific American....can't remember now.
wow, my SCSI raid cards POST in 6 seconds if the drives are spun up.. Longer if it is a cold boot and has to spin up all 12.... but then that's the drives not the raid card... 18gig U160 drives cant be all spun up at once when you have that many.. I dont know about lower quality ATA drives, maybe they draw less power and can spin up that many at once.
Do not look at laser with remaining good eye.
Smaller cable? Pshaw... Sound like Martha Stewart of the Mobo set. Big cables, Baby!
I'm still of the mindset that parallel is better than serial, particularly where high bandwidth is concerned. Probably the _real_ advantage is that they'll be making the mobos for instead of $$$.
Hotswap, now that's a definite advantage, assuming your version of Windows doesn't decide you've suddenly changed the system too much and shuts down until you get Microsoft on the phone and they grant you a new code to allow you to keep running. (A friend replaced the CPU on his mobo and Windows stopped working, until he called Redmond and they gave him a 40-some letter code to continue, very nice of them, I can't imagine how we've done w/o that advantage all these years, but that was another story...)
A feeling of having made the same mistake before: Deja Foobar
USB is a crap interface, with all of the transactions going through programmed IO.
The reason USB can get away with that is that it was intended to replace serial and parallel ports, which have fairly low bandwidth.
"We returned the General to El Salvador, or maybe Guatemala, it's difficult to tell from 10,000 feet"
I don't quite understand how is by nature faster in anyway than parallel. Fundamentally it is the other way around. 2x the parallel wired, 2x the data transfer rate, plus all the handshaking is much easier
Some of us still use SCSI just because of the extremely low CPU overhead it requires. The offboard controller can take care of burning a disc for me in the background while I play a quake 3 engine game, without any fear of buffer underruns. I'd like to look into cheaper hardware and Serial ATA certainly fulfills the speed & hotswap needs I have, but what about keeping overhead low? Anybody have any figures on this?
Even Jesus hates listening to Creed.
Why doesn't anyone make cheap, fast, small (3-6gig) HDs?
There really is ZERO reason for the office folk at my workplace to have the 30gig drives that we are getting these days. And we cant get smaller drives.
So they just wind up only getting a 6 gig partition. Lotta waste.
If you're having issues with a dual 1800+ system not being able to burn CDs on an IDE burner while multitasking - you have a configuration problem!
While I, too, rather liked good quality SCSI CD burners, I've come to the conclusion that they've come a LONG way with EIDE models - to the point where they're every bit as good for 99% of the situations out there.
The main reason I see an EIDE CD burner work poorly is because the user placed it on the same IDE ribbon cable as the device he/she wants to transfer files from, to the burner.
Remember, with IDE, both "master" and "slave" devices on a single ribbon cable are going to be sharing the same IRQ and I/O transfer addresses.
Therefore, if you want to do a lot of copying of existing CD-ROM discs to blank CDs, you *don't* want to put an IDE CD-ROM reader on the same ribbon cable that your IDE writer is on!
By the same token, if you're trying to make a CD (expecially "on-the-fly") from files stored on an IDE hard drive, you'll get better results if your IDE burner isn't sharing the ribbon cable with that IDE hard drive.
IMHO, the most flexible setup is using a SCSI CD-ROM (or DVD-ROM) reader, coupled with an EIDE CD writer, and IDE hard drives. Alternately, if you have a system with EIDE RAID, you can already place your IDE drives on seperate channels from your regular 2 IDE ribbon cables.
I personally hope I never have a 10,000 rpm drive. Rotational speed isn't the only factor, a higher rotational speed gets you more power usage, more heat and more noise. At these speeds you have to consider what stress on the media does to the recording surface, as well. A greater data density, on the other hand, can improve transfer rates while giving you a lower RPM, along with the lower power and noise that go with it. New head technology is promising us much greater data density (remember the recent /. article on terabyte drives?) I would much rather see the manufacturers focus on an approach that continues to improve data density than working on increasing rotational speed.
I'm an American. I love this country and the freedoms that we used to have.
The main thing stopping you sending fast signals down parallel cables is transmission line problems. At high speeds, any wire effectively becomes a transmission line.
It's much MUCH easier to get just one wire right for super high speed data than it is to get 8/16/32 wires right. There's also the issue of ensuring that all the signals arrive at the destination at the same time.
So, technically, parallel is faster, but serial is much easier to get going real fast.
Can someone explain to me the advantages of Serial/ATA over FireWire?
FireWire currently does all these things that Serial/ATA is promising, and there's even speed increases in the works. It would be really nice if PC motherboards started shipping with internal and external firewire ports as standard, and it would mean we'd start seeing native firewire external HDDs a lot sooner.
Do we really need ANOTHER standard ?
http://www.wdc.com or http://www.wdc.com/products/products.asp?DriveID=2 7
Those round IDE cables are just regular ribbon cables shoved through a plastic tube. They do take up less space and are more convenient but the cable sometimes tends to twist around inside the tube, and the bunching appears to put stress on the wires at either end of the connector. Also the connector itself is still the large 40-pin type, whereas the SATA cables look and work more like LVD, which is a much easier cable to deal with. Also much thinner than even the round IDE cables.
I wonder why everyone overlooks the fact that Serial ATA will remove the physical cable length restrioction of the traditional ATA interfaces.
IMHO its one of SCSI's major selling points that cables of LVD-SCSI can be > 5 m without problem.
On the Serial ATA website they claimed that they lifted that restriction, but now, how long can they be ?
(and for cable length, thy building a big tower, with a HD near the top. standard IDE cables don't cut it.)
The move from the current standard to serial ATA is as much to allow for backwards compatibility as increased speed & features. There are plans for adaptors for current drives for use with serial ATA controllers. FireWire interfaces would require all new drive hardware in addition to the motherboard. Simple truth is, forklift upgrading scares folks. They -much- prefer to upgrade piecemeal.
-={(Astynax)}=-
"Darkness beyond Twilight"
Yes and no. Serial cables do have the thickness advantage, but an equivalently shielded serial cable has to be 1/N where N is the number of bits as long as a parallel cable to achieve the same maximum data transfer rate. This is from a basic EM point of view, that the electrical length of a wire is doubles as the frequency doubles.
Any shielding you can perform on a serial cable can be performed on a parallel cable as well. Differential pairs, like the ones used in USB reduce the capacitance the signals see and maintain the waveform shapes, but this is because for data transfer rates of 12 Mb/s or whetever USB is now days, the wires need a bandwidth of 24 MHz. An equivalent Parallel implementation would require a wire bandwidth of 1.5 MHZ which is pretty easy to design for.
There are of course many advantage to serial communication that go beyond these electrical considerations. It is so much more practical and easy to modulate a serial system than a parallel one which is why all communication (cellular, ethernet..etc) uses a serial baseband. Otherwise we would need as many carriers as there were parallel bits. Which is not really inpractical, as the fastest way to increase your data transmission rate on a cellular system is to use N phones in parallel with different ESNs.
So the moral of the story is any serial system can be made faster with a parallel equivalent. Certain systems are naturally more easily and efficiently designed serially, especially those requiring communication over a shared medium.
So why would I want to pay extra for the SCSI one to do the same thing?
In case you haven't been keeping up with IDE lately, you could rewrite that sentence with "IDE" instead of "SCSI" and it'd still be accurate. Modern IDE controllers (with UDMA) will not use more than 1-3% of the CPU (which is close enough to "extremely low CPU overhead" for me). And any modern IDE CD burner will have BurnProof, which will mean that even in the unlikely event of a buffer underrun, you won't burn a coaster -- the drive can just pause burning, and restart when the buffer is filled again.
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
It's all about synchronization. As you increase the speed and/or the length of a parallel bus it gets harder and harder to keep all of those parallel signal paths syncronized. Eventially, a point of dimishing returns is hit, where the problems associated with driving a high speed serial bus are cheaper to overcome than the problems of a high speed parallel bus.
At some point in the future, someone may figure out a clever way of keeping a THz-level parallel bus in sync inexpensively. Until then, serial seems to be the way to go. Even the successor to PCI might be a serial bus.
Since most truly heavy-duty servers will use RAID anyway, individual drive reliability isn't really a huge issue -- if one dies, you just hot-swap it with another one, and haven't lost any data. Thus, high size*speed per unit price is really what you care about for a large server. You care about the performance of the overall RAID array; the individual disks themselves are just easily replaceable components.
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
The "serial" idea has already been done, it's called FireWire. I have trouble believing that implementing FireWire more expensive, in large quantities, as the chipset-makers churn out.
I've noticed Intel (and Maxtor and Seagate) are members of the 'Serial ATA' group. Is this being done, instead of just using FireWire, because of Intel not wanting to put FireWire support into their chipsets?
Has anyone seen a site with a nonbiased side-by-side comparison of the various technologies (USB2 / Firewire / SATA / FC [?] / U160~320 ) and what drives they're available on, and what the perfomacnce is?
o/~ Join us now and share the software
The initial Serial ATA will run at 150MB/s (which is faster than the current ATA/133 @ 133MB/s).
However, with the exception of Seagate, all the other Serial ATA drives (from Maxtor, WD, Samsung, etc.) are "donglized" drive, meaning there's a (Marvell) "Parallel ATA-to-Serial ATA" converter chip sitting between the drive and SATA controller. So essentially these are still ATA/133 or ATA/100 drives, and their top Burst Read speed is still bound by either 133 or 100 MB/s. Seagate will be the only "native" Serial ATA drive capable to hitting 150 MB/s. The best I've seen is around 112 MB/s Burst Read.
Also, the initial Serial ATA controllers will only have 2 channels (meaning two drives), but later versions should have 4 or more channels.
Yes, SATA will support drives larger than 128GB. I don't recall exactly how big a drive it will support but it's probably 2TB or some other ridiculously large size.
I remember on the Amiga, I once got the error: "Please insert disk into drive DH0:". Ah those were the days. :)
Please consider making an automatic monthly recurring donation to the EFF
Can someone tell me the advantage of FireWire over Ethernet? FireWire is 200-800 Mbps (soon to be 1.2 Gbps) and up to 64 devices connected. Ethernet is 10Mbps to 10Gbps and supports over 4 billion devices connected with much longer cables. And there's no Apple licensing fees.
Got friends?
Any time an interface-changeover occurs, it's important to look at what else is on the horizon at the same time. Will the first 5% of drives with this new interface be the only ones without build in Digital 'Rights Management' (DRM) features?
I see this as a great opportunity for the DRM advocates to obsolete all older drives ("sorry, your old drive won't plug into the new motherboards") and force a change-over to the new drives with DRM in their firmware.
Just a point to ponder.
Reading that I thought I was reading a SCSI specification! Seeing phrashes like commmand queuing, out-of-order execution/delivery, and scatter/gather! Topology support for multi-inititaor networks sounds a lot like what SCSI can do, with having two SCSI cards on a bus accessing the same drives. Cool
Got friends?
- Support for the SCSI protocol
- Support for tagged queueing, allowing the drive to multitask. The standard ATA and SATA protocols do not support this yet.
- A single port can connect to multiple drives through an expander (similar to a switch). Currently, SATA is a strict point to point connection.
- Multiple adapters can talk to the same drives.
- Backward compatible support for SATA drives using a tunneled protocol that even allows multiple adapters to talk to the same SATA drive.
- Initial speeds of 1.5 Gb/s and 3Gb/s per port, compared to SATA's 1.5Gb/s per port
Expect Serial Attached SCSI to be targeted at the server market. SATA will be targeted more at the desktop and low end servers where performance and reliability aren't as critical, but cost is.Actually, I would say it isn't weirder that they have to stop the PCI card, just because it is in a different case doesn't mean it should be treated much differently than PC Cards.... of course most of the time you can just eject the card and the driver is good enough to recover, unless of course you are doing some sort of asyncronous IO on a drive attached to that interface... I've always viewed the "stop hardware" requirement to be more making sure the user finds out about processes with the device locked before ejecting more than anything else. Granted it actually does make it inaccessible, but it's far more important to have it catch potentially dangerous conditions before bad stuff happens.
XML is like violence. If it doesn't solve the problem, use more.
People are getting tired of their computers sounding like jet engines.
What is it with people complaining about their computers making noise anyway? I actually like my computers to sound like they're on...the lack of noise makes me nervous (see: Dead Silence, aka Power Outage). I have a computer by my bedside and the noise helps me sleep...in fact, I have a very hard time sleeping without that white noise.
Computers make noise, just like refrigerators make noise, washing machines make noise, and cars make noise. It's not like it's constant beeping, either, folks. Get ovah it.
Blog,Twitter
I hope there's somebody with some engineering background out there that can clear something up for me. Back in the day, we were taught from experience that serial (like an RS232) is slow and parallel (like the centronics printer interface) is fast. Yet, lately technology is turning back to serial encoding for high-speed performance interfaces like USB, S-ATA, FireWire, etc.
Is there a particular reason why parallel is being abandoned for new technology? Is it just too complicated to be efficient at high speeds or what?
What did I miss?
"Lawyers are for sucks."
- Doug McKenzie
Serial ATA always struck me as redundant, since we already have IEEE1394. Most 1394 cards already have an internal connector, and it couldn't be that hard to manufacture internal 1394 drives.
So what's the reason to use serial ATA instead of IEEE1394? Cost? Is it easier to implement in the drives?
-- 2 + 2 = 5, for very large values of 2
That, and the fact that it was mainly backed by Intel. For some reason, the people at that company seem to like things that load down the main CPU. It's weird. But, actually, I'm not complaining all too much; I think USB's fairly neat. And there's a "movement" to develop host-less USB technology, which will very much mimic FireWire and allow e.g. a camcorder to talk to a VCR without the involvement of a "smart" host such as a PC. Can't seem to remember the exact name for this new standard at the moment, though.
main(O){10<putchar(4^--O?77-(15&5128 >>4*O):10)&&main(2+O);}
Now don't get me wrong, there will never be a time when 100% of the population using computers is up to speed on stuff like that (at least not for the forseeable future) but to the people it matters to, the word is getting out. My father, a complete computer idiot, called me the other day and talked to me about some of the issues coming up. He's seen some of the Windows Media Player security creeping up on him and he doesn't like it. I never once mentioned it to him, but as more people get informed, they tell others about it. I do not ever expect to hear that stuff from my grandmother since she will probably never download an mp3 or movie file from the internet, but like I said... to the people it matters to the word is spreading.
I'll use the oft-cited reference of Divx. People found out that they would basically have to rent the movie every time they chose to watch it, which pissed off just about everyone. What was the response? Noone bought the technology. I have very little fear about hardware DRM creeping up in all technology (but maybe a few devices which people will choose not to buy). The market will dictate what is successful and what is not, so if hard drives start coming out with DRM in them I can see a huge disaster waiting to happen. Entire stockpiles of these devices will sit unsold until finally the maker takes them back and re-tools them to be non-DRM.
Hell, think back to the whole Intel processor serial number fiasco. It took Intel how long to give people an option to turn it off? Like 2 months I believe. Have faith in the population, people won't just lay over and accept stuff like that.
"Well kids, you tried your best, and you failed. The lesson is, never try." -Homer Simpson
Oops! I guess you're right. It's a funny 80 pin connector called "SCA-2", no power connector like you'd expect, the power seems to be fed to the drive via the one massive connector... and I was also wrong about the 10K rpm too. They're 15K rpm drives :-)
Oh how I love hot swappable drive arrays. My boss just about crapped a cinder block when I yanked one drive out of the array while we were running the payroll job on the machine... Popped the drive back into the array and it took all of 10 minutes to re-syncronize the contents onto it.
No, actually, them's megabytes per second. That's the speed the bus can handle, which is way faster than the drive can read, but any cached reads (from the drive's buffer) are capable of going 150 MB/s. Most drives today can only sustain about 40-50 MB/s, but they can do a cached burst up to the limit of the bus speed.
Carpe Cerevisi - Seize the Beer
Actually, no, it really is MB/s (megabytes per second). No drive can actually sustain that speed, of course, but the buffer can use it. To use WD's famed drive with an 8MB buffer as an example, the buffer completely empties in about 60 milliseconds. To think... just a little bigger, and you could load ALL of Doom 1 in less than 1/10 of a second (if it were already in the read-ahead buffer, of course).
Range Voting: preference intensity matters
Of course, I could just be paranoid. :)
o/~ Join us now and share the software
64bit, that is approx 16 billion gigabytes (assuming one byte per address). Although that seems unlikely now, let's do some approximate math. Disks now are up to 160GB, and in 1990 they were up to 20MB. That's a factor of 8000 in 12 years, that's approximately a factor or 8000^(1/12)=2.115 per year (more than Moore!). If that continues, then we'll be at 16 billion gigabytes in less than 25 years (2.115^25 > 100 million).
;-)
Is it feasible? How many atoms are there on a 3.5" platter? Or in a cubic inch of crystal? For the latter I'm gessing more than 1e21 atoms in a cubic inch of crystal (any physicists out there?), so that's approx 60 atoms per byte in a cubic inch of crystal. That's not a lot, but just might be enough. But there is not a lot of slack, if the LBA addresses blocks of, say, 512 bytes, then suddenly each atom would need to carry 70 bits of entropy. Of course, then with a gallon of crystal material, that would be back to almost 70 atoms per byte.
So.... I'm not convinced yet that 64bits is enough
--- Hindsight is 20/20, but walking backwards is not the answer.
They don't really say that it doesn't work, just that isn't trivial and costs a lot for R. But so does trying to make it spin faster... Eventually the cost for one will be high enough that it'll make it worth pursuing the other.
That page mentions RAIDing the drives, but unless you RAID 1 the drives there aren't multiple heads seeking for the same information so it still takes as long, and up to twice as long as a normal seek, to read data off an array. (The more spindles, the more likely one is just past the data, requiring a full rotation to read it.)
You can fix this by upping the stripe size on a striped array but this decreases performance in the common small reads, right down to the single-drive level.
There are ways around this, some proprietary RAID levels are like 5, with much more redundancy. With results like 5+1. But the results are that you only need x, of y drives, where y is usually two or three times x, to recover the data. Not only are these very very safe, but with a smart controller and well optimized strip sizes you can fulfill read requests very quickly by using the 'x' drives closest to the data. This actually approaches the results of having multiple heads in a single spindle.
But, on the subject of wishful thinking... Why don't they make 5.25" disks anymore? They're about twice as tall as a 3.5 and have about 2.5 times the surface area. A 160mb drive could be 800mb in 5.25 form factor. Sure, the rotational speed would be low and the seeks would be comparatively terrible, but wow, what a ton of storage. It'd be great for long-term archival that didn't actually warrant burning or tape.
I've got 240GB in my computer now and I find it limiting. I wish I could simple throw a fast 80 in as my OS/temp drive and put everything else on a "slow" 800, moving it around when I needed to work on it.
Remember the Quantum Bigfoot?
with more delicate electronics and finer distances, consumers seem to just thrash things that have any kind of inertia or movement. bleh.
oh well.
EOM
What about my previous post was "wrong"?
I have no doubt you're having success burning discs with your IDE writer attached to the same ribbon cable as your CD-ROM drive. The point is, it works for you because you have a newer writer with the "burn-proof" type technology in it. That's really just a band-aid for the buffer-underrun problem. It works fine, yes - but my suggestions were aimed at IDE CD-burner users in general. Using your setup with an older IDE burner that lacks buffer-underrun protection will likely result in "coasters".