3GBPS vs 6GBPS

A

Aumakua

Limp Gawd
Joined
Dec 14, 2009
Messages
333
okay so if I understand correctly the max read speed of 3Gbps on SATAII is about 250 mb/s, and the new 6Gbps will put that over 500 bm/s

but most drives max read/write are still around 200-250 mb/s so why is everyone so excited about new 6Gbps drives? also factor in that most laptops still only have the SATAII controllers and USB 2.0.

Will the new SATAIII drives be compatible with old SATAII? I would guess they are, but also seems like a waste to put one of those into your SATAII notebook if you won't see the full benefits you are paying for.
 
3Gbps = 300MB/s
6Gbps = 600MB/s

SATA uses 8/10-bit encoding.

HDDs are still limited to SATA/150 speeds as disks typically are limited under 150MB/s.

SATA 6Gbps is useful for SSDs arriving at Christmas this year; featuring newer 6Gbps SATA->NAND controllers. It's just marketing for anything else like regular HDDs. Though the cached speeds will be higher, this is typically not relevant to real performance but makes a good marketing tool.
 
I should add: there is a minor performance increase when using SATA 6Gbps on devices that can only do 150MB/s, as it takes less time to get the data through the cable the latency will also be slightly lower. This latency is called propagation delay, and becomes an issue in SSDs but not HDDs really due to them having so high random access latency themselves; and latency is virtually unimportant for sequential transfers.
 
sub.mesa said:
3Gbps = 300MB/s
6Gbps = 600MB/s

SATA uses 8/10-bit encoding.

HDDs are still limited to SATA/150 speeds as disks typically are limited under 150MB/s.

SATA 6Gbps is useful for SSDs arriving at Christmas this year; featuring newer 6Gbps SATA->NAND controllers. It's just marketing for anything else like regular HDDs. Though the cached speeds will be higher, this is typically not relevant to real performance but makes a good marketing tool.
Click to expand...

thanks for the info, I did read somewhere that you don't get the full 300MB/s for some reason, most people get anywhere from 250-285 even with drives capable of over 300...
 
300MB/s is the bandwidth, but most people do not understand the difference between bandwidth and throughput.

Bandwidth is the raw datarate which an interface is capable of.
Throughput is what the interface can transfer, excluding any command overhead.

I'll give you an example: when you download a webpage, you connect to the webserver via HTTP protocol. You issue several HTTP-specific commands and finally get the page you requested for. Now assume the page is 10KB large and it took 2 seconds since the first byte sent and the page requested fully returned.

In this example, the throughput is 5KB/s since the actual data transmitted was 10KB. But the bandwidth used was higher, since it also had to transfer data related to HTTP which is not part of the actual page that you requested. If this equals 2KB, then the total data transferred was 10+2=12KB. So 12KB / 2 = 6KB/s would be the used bandwidth, while 10KB / 2 seconds = 5KB/s was the throughput.

So to transfer 200MB of data, you actually need more than 200MB going through the write; the command overhead. Usually this is small, though. But it does mean you can never get the full 300MB/s of throughput on a SATA/300 interface; for that you might need SATA/320 or something. ;-)

This also applies to network (gigabit) bandwidth. 1 gigabit = 125MB/s but you can never get 125MB/s throughput, as TCP/IP adds command overhead and 125MB/s is the raw datarate at which zeroes and ones go through the wire; and some of those will be command overhead not actual data.

So talk about bandwidth when you determine how fast a cable/interface can transfer data, but talk about throughput when you talk about actual transferred "end-user" speeds you get; for example when benchmarking your harddrive.

Throughput depends on many factors (drive, interface, OS) while bandwidth is constant for a given interface. Bandwidth could be considered the speed limit of a highway (the cable), while the actual speed of your cargo truck driving there is called the throughput.
 
Aumakua said:
thanks for the info, I did read somewhere that you don't get the full 300MB/s for some reason, most people get anywhere from 250-285 even with drives capable of over 300...
Click to expand...

sub.mesa gave a good explanaiton of how overhead works. Also, I think (correct me if I'm wrong) but the maximum throughput on a SATA II line is something like 286MB/s to 293MB/s. Benchmarks typically are not in bits for hard drives.
 
I thought 280MB/s was about what the Crucial C300 RealSSD did when used with SATA/300; so 20MB/s would be lost due to command overhead/inefficiency. Which is still a good result i think. Better than USB 2.0 with 480MBps bandwidth but only ~28MB/s throughput when transferring data to an external drive.
 
The greatest immediate advantage for the SATA-III speeds will occur when writing short-burts or reading data already stored in the drive cache. Neither of these activities are limited by the actual read/write speed of the media - they can run at the full rate of the cache, which should have no problem sustaining 550-600MB/s.

This advantage applies to both SSD and spinny-media drives. Even though the spinning drive may be limited to 100-120MB/s actual read/writes, as long as each write event does not fill the entire cache then it will be done (much) faster. In general, this is the same argument used to support 300MB/s SATA-II speeds. Until we had SSDs, there were few, if any, spinning disk systems that actually needed >150MB/s SATA-I. The gain was in cached reads/writes only. Which is why after SATA-II was introduced you started to see drives manufactured with larger and larger cache.

Most of the 'typical' [H] subscribers will never see the benefit. It would have limited value in any office/desktop, gaming or even media storage applications. The biggest application benefit would be for transaction database servers. IOP-based benchmarks would show the difference.

Of course it will also be highly beneficial for the next gen SSDs which can do >300MB/s sustained read and (maybe) write.
 
Well my point against a huge HDD cache is that its always less than the RAM; and RAM serves as filecache also. So if a sector that previously has been read will be read again, it will be read from RAM instead not from the disks' cache.

What the DRAM chip is useful for is write buffering; and i believe that's where your database workloads benefit; due to the larger DRAM size, this chip can also be used for write buffering (which is entirely different from caching). It can collect writes and write them in bursts/chunks. This is especially beneficial for multi-stream read/write. If you will be seeking after every sector the speeds will be lower than 0.1MB/s; but when performing burst/transaction-like buffering (i.e. 1 second reading; 1 second writing) it can achieve very high speeds. However, reading one second would already be more than a 64MB DRAM chip can store.

So i think the cache aspect is useless, but buffering can yield significant benefits. Still not very useful to most desktop users, and since a single DRAM chip costs nothing it makes of course a nice marketing tool.
 
Another thing to consider is that right now the 6g SATA controllers found on motherboards are using junk 3rd party chipsets for the most part.

Even if I had a 6G sata drive, on current hardware I'd rather plug that into the reliable ICH10R than some pos afterthought of a 3rd party controller that they glued on.

Hopefully we'll get some real 6G Sata motherboards with Sandy Bridge and the ICH11.
 
I thought, perhaps incorrectly, that write cache to DRAM could be lost in a power outage situation leading to potential data loss/ corruption hence why drive cache is usually for reads. That is why you want a BBU on a raid controller if you enable write caching and why NAND + a cap is better than the old BBWC setups.

I totally agree that 64MB or .5s of burst is not much... unless you have the need to only read the same small file(s) over and over not making changes.
 
I totally agree that 64MB or .5s of burst is not much...
Click to expand...

It's more like 0.2s for SATA2 and .1s for SATA1

Edit: I think you mean .5 seconds if it went to the media and I am quoting burst rates..
 
Last edited:
GotNoRice said:
Another thing to consider is that right now the 6g SATA controllers found on motherboards are using junk 3rd party chipsets for the most part.
Click to expand...

Are you serious? There is no other SATA6 chipset maker other than Marvell, which use in ARECA, HPT, DELL, HP... and many other...

I wonder do you know junk 3rd party chip set? Just curious!
 
Yes, SB850 has 6 full SATA/600 ports; unlike new intel controllers (Sandy Bridge?) which only have two.

And LSI SAS2008 also has 8 full SATA/600 ports on 4GB/s full duplex PCIe x8 Gen2.
 
You must log in or register to reply here.
Back
Top