SSD question: Crucial c300 64gb or samsung 470 64gb

[xman01]

thinking about going ssd route, which one would u choose

[josh1980]

I have a C300. It works well. However, I am going to be upgrading to Intel very soon. My machines that have Intel SSDs just feel more responsive.

[xman01]

even though the read/write speed is lower?

[Johnny87au]

c300, better 4k speeds which is used by 90% most of the time..

[zalbard]

Yeah, C300 has much better performance for daily use.

[xman01]

good to know
thanks

[Johnny87au]

Yeah, C300 has much better performance for daily use.

zalbard u post whore slow down lol, but +1

[mbreslin]

My machines that have Intel SSDs just feel more responsive.

Aside from copying I don't notice a difference between 1 c300 and 8, and 1 intel drive. If it weren't for synthetic benchmarks and big file copy/unzip I would have no idea how many or which controller of ssd was in my system.

Noticeable difference between ssd and mechanical drives: Huge.
Noticeable difference between ssd controllers: None.

(opinion obviously)

[Anvil]

For most tasks there is virtually no difference between the high-end SSDs.

There are of course differences in the long run and so the best SSD for you depends on both usage and personall skills. ( maintainance)

[josh1980]

Noticeable difference between ssd and mechanical drives: Huge.
Noticeable difference between ssd controllers: None.

I agree with that statement. My basis for considering switching from my c300 to Intel X25M is that I have X25M in all of the machines in my house except my SR-2. I went with C300 for the SATA 6Gb/sec feature that isn't available by Intel. For whatever reason the system seems to have bouts of what I can only describe as "waiting". For instance, I click Start and once in a while it takes a few seconds(yes, seconds) for the Start menu to appear. The computer doesn't appear to do anything except sit there. Only an occasional blink of the hard drive light occurs and the CPU utilization is less than 10%. This might happen when I try to copy files, open Office, just about anything I do can stop and the system and pause the whole system for a few seconds. During these times the mouse pointer is an hourglass.

As an experiment I ghosted my C300 to an X25M that was about to go into a new machine. After 3 days with the X25M I never experienced the hesitation once. I then did a few tests with the C300, checked firmware versions, checked BIOS settings, etc and everything seems in order. I then ghosted the X25M back to the C300 and within 5 minutes of booting up my machine I saw it start hesitating again. While I will freely admit that the C300 might not be the cause but merely a symptom, the Intel drive seems to "fix" whatever issue I am having.

[Jansa]

More than agree with previously said about X25M. Beside being "old design" it is much better option for OS drive than C300. Just from my own experience.

[josh1980]

Glad I'm not the only one that feels this way.

Benchmarks only mean so much. Benchmarks are only good at proving that one device/computer is better than the other at a given set of data. Outside of the given data set, there is no guarantee that the device will perform as expected.

I had 4 ANS-9010s on a 128G RAID0 and they didn't "feel" any faster than my Gen 2 X25-M despite having more then 4 times the transfer rates. As far as I am concerned, nobody is going to convince me that SATA 6Gb/sec is anything more than a gimmick. It might perform faster in the benchmarks, but there's no way I'll ever buy a 6Gb/sec drive over a 3Gb/sec drive again just because it can exceed the 3Gb/sec speeds in benchmarks. To be honest, I'm not sure if the "gimmick" nature of SATA 6Gb/sec will be anything more than a gimmick for the next 2 or 3 years. It just doesn't add value to the desktop experience.

[flamenko]

Many good points but in my decision I would rely on the small 4ks here first and foremost although the typical user will not see a difference in any of these drives. I might also consider the Crucial because of its aggressice GC.

[Ao1]

I agree with that statement. My basis for considering switching from my c300 to Intel X25M is that I have X25M in all of the machines in my house except my SR-2. I went with C300 for the SATA 6Gb/sec feature that isn't available by Intel. For whatever reason the system seems to have bouts of what I can only describe as "waiting". For instance, I click Start and once in a while it takes a few seconds(yes, seconds) for the Start menu to appear. The computer doesn't appear to do anything except sit there. Only an occasional blink of the hard drive light occurs and the CPU utilization is less than 10%. This might happen when I try to copy files, open Office, just about anything I do can stop and the system and pause the whole system for a few seconds. During these times the mouse pointer is an hourglass.

As an experiment I ghosted my C300 to an X25M that was about to go into a new machine. After 3 days with the X25M I never experienced the hesitation once. I then did a few tests with the C300, checked firmware versions, checked BIOS settings, etc and everything seems in order. I then ghosted the X25M back to the C300 and within 5 minutes of booting up my machine I saw it start hesitating again. While I will freely admit that the C300 might not be the cause but merely a symptom, the Intel drive seems to "fix" whatever issue I am having.

I suspect that TRIM/ GC implementation may be behind what you can observe.

Could I pursued you to run hIOmon to see if it can highlight what the issue is? A summary using the presentation client is quite easy to do and it should highlight the difference you can see between the C300 & X25-M if storage is the issue.

[Jansa]

There are as well more components to consider than just SSD drive. Have noticed up to 40% difference between same drive performance with different MB-s. Actually el cheapo P35 being better than P45 EXTREME board with X25M
And beside 4K MB/s there is access time. HDTune Random Access 4K is good indicator what to expect from OS drive. Believe me or not, by these numbers my 2 old Mtron 7535 drives in R0 beat X25M and this beats C300. And it really feels that way as well, despite all of the rest fancy numbers benchmarks show.
Can be just anomalies with specific boards I have but more and more I start to believe just 4K random access IO numbers. Correct me if I am wrong

[Computurd]

Bandwidth=sheer output of the device.
Latency=the speed at which you receive that bandwidth.


one thing that must be considered here is the difference in latency of course. here is a scenario:

DRIVE A: does 4k random @ qd 4 at 10,000 iops with a latency of .25
DRIVE B: does 4k random @ qd 4 of 7000 iops with a latency of .125

so in desktop usage, which is faster? the drive that does 7000 iops! you have twice the speed of the transfer. the latency is that much better. so in the amount of time drive A kicks out 10,000 IOPS (.25), drive B kicks out 14,000 (.125 x 2 =.25).

that is where the base 4k comes into play more than anything. Latency seems to scale very close to linear. SO, your base 4k latency @ QD1 should be indicative of overall latency of the device.

so a device that is 'faster' at the higher QD, but has higher latency, is not really faster in my estimation. that is why i have really begun to focus in on latency measurements when i do my testing across the whole spectrum of a drive and/or array.

this really goes back to Gullars plotting what he referred to as QOS (Quality of Service) charts where the bandwidth was taken into consideration with the latency, and a 'score' given for a calculation of the two. an excellent brilliant idea that was and is, just not sure the formula he used was perfect, but the concept is stellar.

that is a problem with some of the drives out there that are 'newer and faster'. they have higher latency it seems overall. is this what intel is addressing that is taking so damn long? lol!

but where is the tipping point? at what point does the bandwidth increases begin to outweigh the latency, or begin to make latency less relevant? that i do not know of course, but big sequential speed can still equal big performance in some sceanrios. now whether those scenarios are indicative of a normal users usage pattern is another thing entirely. if copying large files all the time, which desktop users dont do a whole lot of, then you never really are going after that sustained transfer.

s. Actually el cheapo P35 being better than P45 EXTREME board with X25M

i couldnt agree more. there is a difference with motherboards. i began to blame the NF200 for a lot of problems that i had with a certain mobo (e759), and when i switched to the e760 with no NF200 things were great
however, i have seen other boards where the NF200 didnt seem to make much of a difference, if any. BUT i didnt see latency plots like we were speaking of. a test such as this would really tell the tale: compare the latency with the different boards!

[johnw]

Your definition of latency appears to be non-standard. Most people I have dealt with consider latency to be the wait time between issuing a command and the time that the command begins to generate results.

Perhaps an analogy would help. Consider the stacks desk at a University library. There are a number, call it N, librarians behind the desk, and a number, QD, students coming to the desk to request books (assume the desk is wide enough that all students can stand side-by-side). Each student can only request one book at a time, and each librarian can only retrieve one book at a time. After each student requests a book, a librarian goes down into the stacks, finds the book, brings it back to the desk, and gives it to the student.

If each student has a stopwatch and times how long it takes between issuing their book request and receiving their book, then you can query each student and get a list (with QD entries) of service times. If you average all of those service times, you will have the average latency of book retrieval.

Alternatively, you could measure the total number of books retrieved per minute by all the librarians together, divide that by QD to get the average rate per student, and invert that number to get an estimate of the average book retrieval latency.

Note that if the number of students, QD, is less than or equal to the number of librarians, N, then the average latency will be about the same as for a single student and single librarian. Maybe it will be a little higher than for QD=1, since the stairways in the stacks are narrow, and the librarians may occasionally have to wait while another librarian climbs the stairs.

But if QD > N, then you are going to see the average latency start to rise quickly as QD increases. You can see this in your data, where it looks like N is between 64 and 128 for the 9260, but between 8 and 16 for the 1880. But with the 1880, each "librarian" is much quicker than each "librarian" for the 9260. There are just fewer librarians with the 1880, even though each librarian is quicker.

[Mungri]

I've been wondering just how exactly can windows get any faster than it is on my C300 :p

I have a C300. It works well. However, I am going to be upgrading to Intel very soon. My machines that have Intel SSDs just feel more responsive.

How much difference do you really think you will see over a C300? I really cant see it getting any more responsive with anything else.

[Anvil]

DRIVE A: does 4k random @ qd 4 at 10,000 iops with a latency of .25
DRIVE B: does 4k random @ qd 4 of 7000 iops with a latency of .125

which is faster? the drive that does 7000 iops! you have twice the speed of the transfer. the latency is that much better. so in the amount of time drive A kicks out 10,000 IOPS (.25), drive B kicks out 14,000 (.125 x 2 =.25).

Hi CT,

I'm equally intrigued by this but there is a flaw in this scenario.

If drive B delivers 7000 iops it delivers 7000 i/o per second, not 14.000

To make 14.000 iops happen you would have to add another worker, which is what johnw is explaining with the library analogy.

I haven't spent much time trying to find a good explanation to this since I and GullLars produced those datasets, I'll check if he is available for comments on this.

[Ao1]

Bandwidth=sheer output of the device.
Latency=the speed at which you receive that bandwidth.

Queue depth*(1/latency in ms) = IOPs
IOPS * Transfer Size In Bytes = BytesPerSec

Drive A:
4k random read @ QD 4 = 1/0.00025 x 4 = 16,000 IOPs
16,000 IOPs x 4,096 = 65,536,000 = 62.5MB/s

Drive B:
4k random @ qd 4 with latency at 0.125 = 1/0.000125 x 4 = 32,000 IOPs
32,000 IOPs x 4,096 = 131,072,000 = 125MB/s

As a cross check for the first entry for the 9260-8i
1/0.0001329 = 7,524 IOPs
7,524 IOPs x 4,096 = 30,818,304 = 29.39MB/s

Edit

As a cross check for the QD 32 entry for the 9260-8i
1/0.0004045 = 2472 x 32 = 79,110
79,110 IOPs x 4096 = 324034610 = 309.02MB/s

[Computurd]

well the comparison wasnt meant to be between the raid cards actually. i am just using that as a highlight of the methods that i am using to compare things. that is a good analogy john.

so wipe that data out, lol. i know which one is faster



but the drive scenario is what i am more interested in. i am very interested in you guys thoughts on that theory.

if the drive has faster latency at a given QD than another drive, shouldnt the amount of iops generated be more in total in regards to typical desktop usage.?


because, using johns analogy, if the book is already fetched, doesnt that mean that the librarian can then serve the next request faster, than say a slower librarian??

see IOPS is input output per second. but if that input/output reaches you faster, wouldnt that equate to a higher performance? definitely in the 'snappiness' aspect, but i mean overall?

in a desktop scenario, considering that according to A01 testing that he is done (kudos), that the average file transaction is finished so quickly, its not like you are having long sustained bursts of files/information.

its just blip blip blip. and the faster the blip, the faster your on to the next blip!
the IOPS dont take a whole second. not even close. so the iops served faster should be on to the next iop.

am i missing something here?

[mbreslin]

Glad I'm not the only one that feels this way. ... To be honest, I'm not sure if the "gimmick" nature of SATA 6Gb/sec will be anything more than a gimmick for the next 2 or 3 years. It just doesn't add value to the desktop experience.

it is much better option for OS drive than C300.

Why am I the only one that will address these ridiculous comments?

The drives that exceed sata-ii speeds do so via SEQUENTIAL TRANSFERS, the benefit is when dealing with LARGE FILES, that is the benefit to sata6gb/s. That this isn't useful in your usage scenarios doesn't make the interface a gimmick. Plenty of people transfer large files and find the speed improvement while doing so to be welcomed.

And as to the "snappyness feeling" I maintain it's similar across most all modern drives, and IF there is some difference it is SLIGHT at most. Your experience of clicking the start button and taking SECONDS for the menu to appear is not remotely relevant as a quality indicator in general as it suggests either 1) you have one horribly bad egg of a drive, or 2) your system was horribly configured.

Even slight stuttering hasn't been experienced since the the very first iteration of ssd controllers.

On Moddern SSDs if you're getting stuttering or freezing, you're doing it wrong.

[Ao1]

well the comparison wasnt meant to be between the raid cards actually. i am just using that as a highlight of the methods that i am using to compare things. that is a good analogy john.

so wipe that data out, lol. i know which one is faster



but the drive scenario is what i am more interested in. i am very interested in you guys thoughts on that theory.

if the drive has faster latency at a given QD than another drive, shouldnt the amount of iops generated be more in total?


because, using johns analogy, if the book is already fetched, doesnt that mean that the librarian can then serve the next request faster, than say a slower librarian??

see IOPS is input output per second. but if that input/output reaches you faster, wouldnt that equate to a higher performance? definitely in the 'snappiness' aspect, but i mean overall?

considering that according to A01 testing that he is done (kudos) that the average file transaction is finished so quickly, its not like you are having long sustained bursts of files/information.

its just blip blip blip. and the faster the blip, the faster your on to the next blip!
the IOPS dont take a whole second. not even close. so the iops served faster should be on to the next iop.

am i missing something here?

Lower latency = faster IOPs

Double the QD with the same latency and you double the IOPs

Bytes per second = transfer size x IOPs

At QD 256 the latency has gone up significantly = less IOPs however this is offset by the multiplication factor of 256.

At QD 128 you have ~ half the latency, but half the QD multiplication factor so MB/s is around the same.

Increasing the file transfer size will reduce the IOPs.

[Anvil]

but the drive scenario is what i am more interested in. i am very interested in you guys thoughts on that theory.

if the drive has faster latency at a given QD than another drive, shouldnt the amount of iops generated be more in total?

To make your scenario work we could turn it around and say that there is a finite number of books to handle e.g 7000.

DRIVE A is capable of handling 7000 books per second
DRIVE B is capable of handling 14000 books per second

As there were only 7000 books to deliver drive B would be finished in .5 seconds.
So yes, the fastest drive is the drive that can handle most iops. (or books per second )

Why am I the only one that will address these ridiculous comments?

As long as the industry is selling MB/s someone has to comment on these questions, ever so often
If only there was some other measure.

[Computurd]

basically what i am trying to say is that it all boils down to latency. the lower the BASE latency (4kqd1) the faster the device is period. especially in a desktop scenario.

Why am I the only one that will address these ridiculous comments?

sometimes even 'ol comp gets tired of arguing LOL
of course sata 6gb/s is better! dude, they arent taking into consideration the other things that sata 6gb/s brings to the table. like better latency, asynchronous I/O, and inhanced NCQ which is crucial to SSD's.
there are always those that think that a new protocol is all about the sequential bandwidth. and it never is the only parameter that is effected. of course.
i agree with you totally mr. Mbreslin. good to see ya hangin round btw