AnandTech

Needed? by Spivonious, 256 days ago

The current SATA spec has a theoretical max of 375MB/s. With even the fastest drives today barely breaking 250MB/s and then only under heavily cached reads, what is the need for upping the limit to 750MB/s?

Reply

RE: Needed? by Veerappan, 256 days ago

Due to the way that SATA traffic is encoded, 2 bits of every 10 is lost, and not usable for data transfer (8B/10B encoding). Because of this, the max transfer rate of SATA is 300MB/s, not 375MB/s. As mentioned, some drives are hitting ~290MB/s from cached reads, which is essentially saturating the bus (300MB/s is the theoretical max, not necessarily the usable max).

I'd say for people who are in this situation could use 6Gb/s SATA right now (or should switch to a different underlying technology).

Reply

RE: Needed? by JarredWalton, 256 days ago

I wouldn't be surprised if there's also the old MB vs. MiB issue cropping up. Most software applications still measure megabytes as 1024^2, but hard drives measure it as 1000000. So 288MiB/s would in fact be 301,989,888 bytes per second, which fully saturates a 3Gbps SATA link. Likewise, 589MiB/s would be 617,611,264 bytes per second, so the buffer of a drive can already saturate the 6Gbps link. Of course, as someone else points out below, reads from the buffer are really a marketing gimmick. SSDs on the other had will likely top out on 3Gbps in the next year, making 6Gbps necessary for further speed improvements.

Reply

RE: Needed? by quiksilvr, 256 days ago

Why is there a Megabit and a Megabyte? Cant we just choose byte and be done with it?

Reply

RE: Needed? by sidefx1979, 256 days ago

The MB vs MiB issue has nothing to do with bits vs bytes; It refers to the difference between mega and mebi (i.e. decimal vs binary) prefixes for bytes and the confusion that arises when they're not used consistently... mostly because the binary prefixes weren't defined until relatively recently and so the decimal versions were used to mean 2 different things and still are in many places - such as hard drive specs.

Reply

RE: Needed? by ameatypie, 255 days ago

See http://compnetworking.about.com/cs/basicnetworking/f/bitsandbytes.htm.

That basically explains the difference between bits and bytes, and also why they are nessercery.

Reply

RE: Needed? by yyrkoon, 255 days ago

Uh . . . no . . . bandwidth, not storage.

Also, Veerappan, you do know what maximum theoretical means do you not ? 2b encoded is still information sent. But why say "8b/10b encoded", when you can say 20% communications/controller overhead.

Spivonious, because a SATA channel is not limited to a single drive. Read about SATA port multipliers sometime.

Reply

RE: Needed? by Targon, 256 days ago

As with any drive technology, there are two aspects to how well a drive can work. First, you have the physical system where the information is read off the media and into the memory on the drive, and then you have the transfer ability from the drive to the system.

Now, while current drive technologies may be limited by the controller on the hard drive itself(the electronics on the hard drive), we also need to be worried that the push to improve the hard drives will be slowed down by latency in the electronics. By pushing the transition from 3Gb/second to 6Gb/second, it encourages hard drive manufacturers to improve things on the drive side.

Another way to look at it is that if there is a bandwidth limitation in transferring the data, there is very little motivation for companies to release faster products, since there is no way to show you have a better product than your competition. Why make a drive that has the internal ability to read/write at 6Gb/second when the interface to the computer is only 3Gb/second since people will not see the benefits?

The same applies to video cards, where if you are CPU limited, there is no benefit to a faster video card. There is a balance between all the components in a computer, so the key is to reduce or eliminate potential bottlenecks.

AMD has become something of a platform company over the years. When the people saw that CPU access to memory was being limited by going through the chipset, the memory controller was moved to the CPU. Now, AMD is addressing the need for data from the hard drive, and that is what this article was all about. In time, we will probably see AMD push for a better connection between the video card, CPU, and memory as a way to make an AMD platform a better choice for some people.

Reply

RE: Needed? by finalfan, 256 days ago

See http://www.sata-io.org/members.asp

It is just a demo from AMD and Seagate. AMD is not even a board member while Intel is which is writing the spec. The article gives too much credit to both companies which are apparently not leaders of today's industry.

Reply

RE: Needed? by Targon, 256 days ago

Isn't the term leader supposed to indicate who is actually leading the way? If AMD beats Intel to releasing a SATA 6Mb/second controller, doesn't that make AMD a leader?

Reply

RE: Needed? by Sahrin, 253 days ago

Intel competes with Seagate, AMD does not. That's the explanation you're looking for.

Also, and this is just speculation on my part, given the way Intel was behaving on release of USB3.0 design spec, I'm not surprised to find AMD is being more proactive about having their own standard ready. Intel may have the majority of the chipset market, but if all the other players use a single (incompatible) spec, it will have to be respected by manufacturers.

Reply

RE: Needed? by v413, 256 days ago

Yes, it is needed. The reason why is mainly because of SSDs. And SSDs will soon easily eat up even the provided bandwidth of SATA3. Micron for example are currently developing an SSD with 1GB/s throughput (800MB/sec demonstrated) that will use PCI express.

Reply

RE: Needed? by mariush, 253 days ago

Actually, they got up to 2100+ MB/s using 10 Samsung SSD drives and 800 MB/s using 24 in a RAID (hardware controller was simply not capable of using all 24 at a time and keep the throughput)

Here's the Youtube video:

http://www.youtube.com/watch?v=96dWOEa4Djs

Reply

RE: Needed? by UNHchabo, 256 days ago

The big reason is so that we can have more headroom.

My WD 640 can't even max out my SATA 1.5Gbps link except maybe on cached workloads, but that doesn't mean I wouldn't like a 6Gbps link.

Reply

RE: Needed? by piroroadkill, 255 days ago

Faster SSDs that can easily saturate it? DRAM based SSDs? I don't see a reason why we shouldn't make it faster.

Reply

RE: Needed? by blyndy, 255 days ago

Many people are waiting for the release of power over SATA. Where is it?

Reply

RE: Needed? by 9nails, 255 days ago

Wait a second... Which hard drives are breaking 250 MB/s? You're talking transfer rates right? That seems super quick, even my favorite little VelociRaptor's don't touch 130 MB/s read transfer rate.

Reply

RE: Needed? by Sabresiberian, 234 days ago

Comments like this are always on the goofy side, in my opinion. What do we need for any improvement in the computer world? What do we even need for computers, we can do all of the things a computer does without one, anyway.

The fact is better, faster technology improves our lives in terms of what we want to get out of our time, well, alive. EVERY ADVANCE IN TECHNOLOGY GOES TO IMPROVING THE LIVES OF ALL OF US! How can you not understand this? Every advancement allows for further development in the rest of the industry, and all related industries - related meaning anyone who has use for a computer. Changes like this don't just effect load times for World of Warcraft players and benchmarkers, but allow the whole industry to move forward. Faster drives WILL be built, and not understanding the picture in at least that one, small area is paramount to not thinking at all. I find it ludicrous to think that anyone taking the time to read any article at Anandtech doesn't have the capacity to do so, yet comments like these keep getting made.

Reply

What's with that tiny steps all the time? by davepermen, 256 days ago

Why not 4x as s-ata2? why not 8x? why not arbitary?

why can't they just let both ends diskuss how fast they are and fit the fastest they can together?

now i can have my ssd up to 300MB/s, next year up to 600MB/s. what if i'd like to have 1GB/s? do i have to wait for s-ata4? and then s-ata5?

why don't they create a standard that actually scales?

Reply

RE: What's with that tiny steps all the time? by v413, 256 days ago

I guess they are approaching physical limitations for signaling. SATA3 runs with 6Gbps speed using the old SATA cables. Raising the speed will require either limiting the possible length of the SATA cables (longer cables = more noise, signal deteriorates and hence more errors) or using different cables - most probably optical cables. So 6gbps is probably the limit for the easiest and backward compatible solution for the time being.

Reply

RE: What's with that tiny steps all the time? by Starcub, 256 days ago

It's a shame they are using the same old cables, at least on the external side. Every eSATA cable I have ever used performs intermittently at the slightest bump. They really need to implement a locking connector design.

Reply

RE: What's with that tiny steps all the time? by DeesTroy, 255 days ago

None of my eSATA cables worked well if at all until I took an exacto knife to the connector and removed some excess material. Once I removed a little of the fattest part of the plug, the cables could finally be inserted fully, allowing a reliable eSATA connection. It's hard to blame the cables though as the fault could also lie with my eSATA to PCI-Express adapter and its back plate getting in the way.

Reply

RE: What's with that tiny steps all the time? by garydale, 256 days ago

The problem is, in part, that technology can only advance so fast. If the SATA interface was spec'd at 1G B/s then controllers and disk electronics would have to handle that speed to match the spec. That could add to the price, if it was even doable a few years ago.

SCSI has always been the faster technology but it isn't as widely used because of the added cost.

So long as we can maintain backward compatibility, there isn't a huge issue with updating the standards as technology progresses. As the article and several others have commented, the issue isn't being driven by the hard drives so much as by the more recent SSD technology. As SSD becomes mainstream, the need to unblock its potential creates the consumer demand for a faster interface that doesn't require jumping through hoops to make it bootable.

So even when the new SATA standard becomes popular, I'm not going to throw out my old hard drives. However, I might invest in a new drive for the system and caching when I upgrade my processor & mainboard next time.

Reply

RE: What's with that tiny steps all the time? by MadMan007, 256 days ago

Because moderate incremental increases are more profitable.

Reply

Cached read performance is mostly irrelevant by defter, 256 days ago

Let's do the math:
64MB buffer, 290MB/s transfer speed: 0.22 seconds to read the whole buffer
With 580MB/s tranfer speed (2x) you save only 0.11 seconds if you lucky enough that all necessary data is buffered.

Or when transferring 1GB of data with HDD speed of 120MB/s:
current SATA-300: 0.22 + 7.8 = about 8 seconds
new SATA-600: 0.11 + 7.8 = about 7.9 seconds, which is just 1% faster.

For most users, this is just a marketing gimmick, just like previous versions of SATA or AGP/PCI-E initially. It will take few years before it will become really useful.

Reply

RE: Cached read performance is mostly irrelevant by UNHchabo, 256 days ago

For sustained transfers, you're right -- cache speed isn't a big deal. However, for small transfers, it makes a big difference. Think about how many files on your computer are less than 64MB. Pretty much all of them, except for video files longer than 10 minutes.

Cache latency is measured in microseconds, as opposed to normal hard disk latency, which is measured in milliseconds. That means if you ask the disk for something that's not in cache, you have to wait roughly 1000 times longer in order to start receiving data at 120MB/s, instead of getting data back almost instantly at 580MB/s.

Reply

RE: Cached read performance is mostly irrelevant by JarredWalton, 256 days ago

Isn't that the whole purpose of Windows' data prefetching and caching system? SuperFetch tries to put as many commonly used files and applications into RAM as possible. What's more important: 4GB of SuperFetch cache running at a bandwidth of 10000MB/s or more, or faster hard drive caches that run at a bandwidth of 300 (or 600 with this) MB/s? Not that SuperFetch or Windows caching in general does a perfect job of caching important data, but I'd think it would be easy enough to set up a much larger, faster cache in RAM than anything we'll see on HDDs in the near future.

Reply

RE: Cached read performance is mostly irrelevant by UNHchabo, 256 days ago

The main difference would be the caching algorithms. Hard drives use a different caching algorithm than Windows, Linux, or OSX, so the data that's in the HDD cache won't be in RAM or the filesystem cache.

Then when the CPU wants data, it checks its internal cache first, then checks RAM, then checks the disk. If the disk has it in cache, then it'll be passed up. If not, then it has to fetch the data off the platters. More disk cache necessarily means fewer cache misses, which leads to greater speed. The only way that you could fully supplant disk cache with RAM is if you set up a RAM disk large enough to contain your entire file system.

Reply

RE: Cached read performance is mostly irrelevant by epobirs, 251 days ago

And where does SuperFetch obtain that data? From the drive(s), of course. So drive performance makes a difference in how fast the system can reach its best operational state.

So long as the performance isn't greater than the other components can work with, better performance is always a win. Could be a big win, could be a small win but the improvements add up. A poster above was complaining about a situation where it would only be a 1% improvement but failed to consider how those little slices of time add up across millions of operations while using a computer. If a new machine costs no more or is lower priced than the equivalent from a year earlier but has small incremental improvements throughout, the consumer is benefiting nicely. since most non-enthusiasts go a few years between new machines, the difference si a much more appreciable.

Reply

RE: Cached read performance is mostly irrelevant by defter, 256 days ago

Actually, with smaller files the difference becomes even more insignificant. For example, transferring 10MB file from disk's cache would take 0.034 seconds with current SATA-300. With SATA-600 you would only save 0.017 seconds, would you notice such difference? Probably not.

[quote]Cache latency is measured in microseconds, as opposed to normal hard disk latency, which is measured in milliseconds. That means if you ask the disk for something that's not in cache, you have to wait roughly 1000 times longer in order to start receiving data at 120MB/s, instead of getting data back almost instantly at 580MB/s.[/quote]

I agree that cache is very useful, but this story is about faster SATA standard, which increases only the speed of sustained transfer. It will not grow the cache.

Reply

Why.. NOT? by LuxZg, 256 days ago

OK, thre are people here that sound like Bill Gates with his "noone will need this much memory" speech..

There is never "too much". Actually, if you ask me - I wonder why they haven't made SATA "1" at 6 Gbps initialy, than we wouldn't have had worried about different motherboards and all that for a while longer.

So yes, I'm all for SATA3. Only complaint is - why not 60 Gbps :P

More is better, maybe a marketing gimmick today, but in future it will again be too little. So I'd rather be prepared than sorry...

Oh, should I say that I wouldn't mind at all if they had this 10y ago, than I wouldn't have to worry about old computers without USBs, withous SATA, without ethernet, without PCIe slots etc..

The more the better.. always..

Reply

RE: Why.. NOT? by epobirs, 251 days ago

For your first sentence to make sense there would have to be no other comments. Just silence. Why? Because Gates never made any such statement. That stupid myth needs to die.

Reply

drives with hypertransport ? by William Gaatjes, 256 days ago

If the bandwith increase keeps up, it might in the near future be more usefull to implement a SSD harddrive with a hypertransport channel. The harddrive would be connected directly with the cpu. Let an adapted form of the standard MMU manage the hardrive. A cpu with 2 Memory management unit's would be needed. 1 for main dram memory and the other for flash mem. When later on the fasttrack ram and the memristor memory technology come out, we will finally have real instant on computers.

Reply

RE: drives with hypertransport ? by Wrzesien, 256 days ago

It looks like this year we will see SSD RAID cards that connect through the PCIe x8 slot. Kind of reminds me of the classic 20MB "hardcard" for the ISA slot! Except this is 800MB/s of 1TB SSD RAID with a 256MB cache. Not out of this world price-wise either!
http://www.madshrimps.be/?action=getart...ber=1&artpage=3979&articID=911

Reply

RE: drives with hypertransport ? by Wrzesien, 256 days ago

My bad...speeds for this drive are officially:

• Max Read: up to 600MB/s
• Max Write: up to 500MB/s
• Sustained Write: up to 400MB/s

Assuming the Intel SSD's are faster someone could build something for 800MB/s too.

Reply

too slow by marraco, 256 days ago

Is a minor improvement over today SATA II.

It will not support the SSD disks performance.

Reply

1 Year from now. by iwodo, 256 days ago

SATA 6Gps isn't even final yet. And by the time it reaches its final spec and implemented on Chipset. ( Which is at least 1 year from now ), we would already have SSD drive that significantly saturate it.

Since USB 3, SATA 6Gps, and PCI Express 3.0 all expected to arrive at the same time. I suspect those New and top of the range SSD will live on a 1x Slot PCI Express 3.0.

Reply

Seagate should first fix theirs drives by matheusber, 256 days ago

New technology ? they should be shamed to make such a load of dying crap. If you look for this info on their site, you will find almost nothing.

they deserve garbage thrown on theirs fabs !

Reply

RE: Seagate should first fix theirs drives by blyndy, 256 days ago

Many people are waiting for the release of power over SATA. Where is it?

Reply

RE: Seagate should first fix theirs drives by blyndy, 256 days ago

Whoops, I didn't mean to make my comment as a reply matheusber.

Reply

nothing is officially in the pipeline as of now? by supremelaw, 255 days ago

http://www.storagereview.com/newest_cheetahs

Cheetah ® 15K.7 Drive Specifications

* Capacity 600, 450, 300GB
* Interface 6Gb/s SAS-2.0, 4Gb/s FC
* Spindle Speed 15,000 RPM
* Seek Time 3.4 ms
* Reliability 0.55% AFR / 1.6M hours MTBF
* Cache 16MB
* Form factor 3.5-inch

Cheetah ® NS.2 Drive Specifications

* Capacity 600, 450, 300GB
* Interface 6Gb/s SAS-2.0 (600 and 450 GB drives), 4Gb/s FC
* Spindle Speed 10,000 RPM
* Seek Time 3.8 ms
* Reliability 0.55% AFR / 1.6M hours MTBF
* Cache 16MB
* Form factor 3.5-inch

Reply

Duty now for the Future... by Marc B, 254 days ago

While I think this is a cool development, it's current usefulness isn't that restricted by the current SATA 1.5/3 technology on motherboards. The great thing about SATA is it's forward and backward compatibility. You'll be able to not only use, but also maximize the performance of your old(new tech) HDD in your new computer/mother board upgrade.

One area I see potential in the short term for SATA 6 is for SATA Raid controllers, where the latency of SATA 3 could potentially be a limiting performance factor in large RAID 0,5,6,10, etc. configurations. Now that this technology is developed, controller companies can get cracking on SATA 6 cards for PCIx4/PCIx8 slots that could be implemented and ready for market around the time HDD manufacturers are ready to release the SATA 6 drives en mass. Current motherboard technology is already available to utilize them in this capacity.

Reply

Forget SSD, my RAID saturates too! by melgross, 254 days ago

With all the talk about the mythical high speed SSD that no one here will be able to afford for several more years, there are more pressing reasons why we need higher speeds.

Only one other person mentioned port sharing. most SATA RAIDS use port sharing. This means that four 120 MB/s drives together, saturate the SATA bus.

Even two high speed drives come close, and the buffer is past it already.

for people using one drive per port, it's not an issue.

Reply

RE: Forget SSD, my RAID saturates too! by Zak, 249 days ago

In your RAID setup each HD is connected with a separate cable to a separate SATA port, isn't it? That means that each of your 4 drives has a dedicated SATA link capable of full 300MB/s or whatever. Therefore, your RAID is not saturating SATA bus. You'd need a single drive capable of >300MB/s to do that. That would be SSD.

I've read on OCZ forums, I think, that the problem is that SATA bus is capable of doing 300MB/s but the interface hardware is not designed to run at that speed all the time since the manufacturers didn't have to really worry about that until now. Now a single SSD capable of ~250 MB/s rates would make SATA interface work hard almost all the time which may lead to premature failures.

Z.

Reply

RE: Forget SSD, my RAID saturates too! by Zak, 249 days ago

Oh, I wasn't replying to last post, port multiplier is a whole different issue, I was referring to a multiport internal RAID. Port multiplier is kind of a gimmick IMHO. Besides several HDs sharing bandwidth I think the controller talks to one drive at a time which conflict with NCQ. PM is not for real RAID applications where speed matters.

Z.

Reply

RE: Forget SSD, my RAID saturates too! by sleepeeg3, 246 days ago

"Mythically" fast SSD...

SSD uses multiple channels to deliver its speeds. All they need to do to increase transfer rate is to create controllers that can supply more bandwidth.

This seems like an abnormally low standard to keep magneto-optical drives relevant.

Reply