Does Kingston's reasoning behind my slow write speed performance make sense?

[Berryracer]

I bought a Kingston HyperX 3K SSD 480 GB

I feel my OS is snappy just as it should on an SSD but in my Benchmarks on CrystalDisk Mark and AS SSD Benchmark, I got 500 MB/S read which is good and 170 MB/S write only

Now the advertised speed of the SSD on the box says 540 MB/S Read and 450 MB/S write

So as we can see here, my write speeds are in no way near what is advertised.

I then upgraded the firmware from the original firmware 501 to 503 and that made my write speeds double from 170 to 316 MB/S which is better, but still no where close to the advertised speed.

Upon contacting them, they advised me to run the ATTO Disk Benchmark and then I got 530 MB/S read and 450 MB/S write as advertised; reason being, is that the ATTO Disk Benchmark they say uses compression in the benchmarks which is where this drive excels and the other benchmarks such as CrystalDisk Mark and AS SSD Benchmark don't use compression.

That is all well and good but my question is, in my real world usage, does Windows or any other program for that matter use compression? I mean, how does their 450 MB/S write speed in compression mode work in real life? or is it just a benchmark gimmick because when copying files from the SSD onto itself, example:

I try to copy a 16 GB movie from D: (which the 2nd partition in this SSD) onto another location in D:

The write speed I see in Windows goes from 420 MB/S at first, drops to 370 MB/S within 2 seconds, then keeps dropping gradually all the way until it reaches a stable 135 MB/S

This sound very crappy to me and I think that compression in ATTO Disk Benchmarks is just a gimmick because who uses compression??

Hope someone can shed some light on the above

Thanks

 

[Hellhammer]

It depends on the data you're dealing with. Some data formats are more compressible than others and what SandForce does is to take advantage of the compressibility of data to improve performance (basically they are taking e.g. 100MB of data but compressing it to 80MB before writing to NAND, which means they need to write less than other SSDs).

Usually data is not fully compressible (like ATTO uses for testing) but not totally incompressible (like AS-SSD and Crystal use) either. That means your performance falls in-between the numbers you've seen in benchmark apps, although this is scenario dependent. For example movies are often already in a compressed format (lossless files take soooo much space, that's why), which means SandForce cannot compress the data anymore and that's the reason for poor performance.

 

[Berryracer]

It depends on the data you're dealing with. Some data formats are more compressible than others and what SandForce does is to take advantage of the compressibility of data to improve performance (basically they are taking e.g. 100MB of data but compressing it to 80MB before writing to NAND, which means they need to write less than other SSDs).

Usually data is not fully compressible (like ATTO uses for testing) but not totally incompressible (like AS-SSD and Crystal use) either. That means your performance falls in-between the numbers you've seen in benchmark apps, although this is scenario dependent. For example movies are often already in a compressed format (lossless files take soooo much space, that's why), which means SandForce cannot compress the data anymore and that's the reason for poor performance.

Do you think if I right click on my C: and D: partitions (which are the SSD Partitions) and compress the drives, that would yield better performance for me in real life or is that a worse idea?

 

[Hellhammer]

Do you think if I right click on my C: and D: partitions (which are the SSD Partitions) and compress the drives, that would yield better performance for me in real life or is that a worse idea?

The would decrease performance because then Windows would do the compression and the drive would only receive incompressible data.

 

[Berryracer]

Just uninstalled this crappy SSD and taking it back for a refund tomorrow

I put my old LAT-256 back in for now

The only other option they have in stock (crap like this is limited in Dubai) is a Transcend 512GB 2.5" SATA III SSD720 Solid State Internal Drive

but I cannot find any reviews on it so I will take my laptop and clone my disk to it there and test it over there!

sigh....

 

[dave_the_nerd]

The Transcend uses the same Sandforce 2281 controller as the Kingston, and would give you the same performance penalty using incompressible data.

If write performance with incompressible data is the most important thing to you, the LAT-256 is your best pick of the three, although the Sandforce drives will outrun it in pretty much any other benchmark.

If I were you, I'd drop that Kingston back in.

 

[Hellhammer]

Just uninstalled this crappy SSD and taking it back for a refund tomorrow

What made it crappy? Figures that manufacturers promise are always best-case numbers, it's nothing new that the write performance varies depending on the data type when it comes to SandForce based SSDs. The Transcend you linked is based on the same controller so it suffers from the exactly same "issue". Like you said, your OS and everything feels as snappy as they should, so I don't see your reasoning for returning the SSD.

 

[Berryracer]

I see, thanks a lot for the advice guys you are a true help and thanks for saving me the time and hassle of getting the Transcend only to get the same chipset.

I guess I'll drop it back in, it was snappy at everything it's just the benchmarks that were killing me I will stop benchmarking

 

[Cerb]

That is all well and good but my question is, in my real world usage, does Windows or any other program for that matter use compression? I mean, how does their 450 MB/S write speed in compression mode work in real life?

It's much more useful for reducing WA than it is for improving desktop performance. Let's say you've got an erase block of 512KB (I haven't seen them specified for most recent MLC NAND, so maybe it's 1MB, or maybe 256KB, instead), filled with mostly text or fixed-size numerical data, and need to move it, to write new stuff there. So you need to write like 300K somewhere else. But, it's compressible, so you might get away with writing 100K somewhere else to free up that erase block (a whole block must be erased at a time, but it can then be written to at 4K/8K granularity). Controllers without compression must copy over the in-use blocks at their original size, thus making them unable to reach or get below 1 WA, where Sandforce controllers can.

 

[Berryracer]

It's much more useful for reducing WA than it is for improving desktop performance. Let's say you've got an erase block of 512KB (I haven't seen them specified for most recent MLC NAND, so maybe it's 1MB, or maybe 256KB, instead), filled with mostly text or fixed-size numerical data, and need to move it, to write new stuff there. So you need to write like 300K somewhere else. But, it's compressible, so you might get away with writing 100K somewhere else to free up that erase block (a whole block must be erased at a time, but it can then be written to at 4K/8K granularity). Controllers without compression must copy over the in-use blocks at their original size, thus making them unable to reach or get below 1 WA, where Sandforce controllers can.

Can you simplify this in layman's terms please?

I thought the whole issue with my drive is it being sporting the SandForce controller which is a major failure?

 

[dave_the_nerd]

Can you simplify this in layman's terms please?

Compressing data before writing means less data is written. This means: 1) Faster. 2) Less writing.

Less writing means the SSD lives longer.

I thought the whole issue with my drive is it being sporting the SandForce controller which is a major failure?

The first couple versions (1000 series) were, and early firmware releases sucked, and OCZ rushed them to market before the fixes were out, and had crap support. Sandforce and OCZ cut prices to compensate.

If you have a modern (SATA-3, SF-2281 or higher) SandForce drive from a reputable manufacturer (like Kingston) and the latest firmware revision available, you'll have a perfectly adequate and reliable (and fast) drive for $10 less than the competition.

I won't try to convince you that OCZ drives are totally better now and not sucky, because that's not at issue.

 

[Berryracer]

I see, thanks for the explanation

well I gave the SSD back to the store and waiting for them to liaise with Kingston directly, let's see how it goes, after what you've said, in both cases if they accept to refund it or not it's a win win situation

they tried convincing me to get another SSD from the store rather than refund it all they had were sucky SSDs based on the sand force chipset like Crucial M4, or Corsaid, or Transcend

 

[zokudu]

The crucial m4 doesn't use the sandforce controller.

 

[Berryracer]

The crucial m4 doesn't use the sandforce controller.

oh, but when I googled it this is the first thing that came up so that's why I assumed it was the crapp SandForce chipset as well:

Crucial M4 (512GB) review: A fast-reading, slow-writing, high-capacity SSD

It's read speed is around 416 MB/S while the crappy Kingston HyperX 3K SSD's read speed is around 500 MB..hmm...


What's your write speeds according to Crystal Disk Mark and AS SSD Benchmark


would appreciate a few screen shots and please tell me if you are using the IRST dirver and which version or just the chipset drivers?

 

[spandexninja]

I then upgraded the firmware from the original firmware 501 to 503

Didn't they take down the firmware update?
http://www.kingston.com/us/support/technical/downloads?product=sh103s3&filename=SH103_503fw_win

Did you update to 503ABBF0?

 

[Berryracer]

Didn't they take down the firmware update?
http://www.kingston.com/us/support/technical/downloads?product=sh103s3&filename=SH103_503fw_win

Did you update to 503ABBF0?

yeah they sent me the username / pass for

ftp://ftp.kingston.com

its there

 

[zokudu]

oh, but when I googled it this is the first thing that came up so that's why I assumed it was the crapp SandForce chipset as well:

Crucial M4 (512GB) review: A fast-reading, slow-writing, high-capacity SSD

It's read speed is around 416 MB/S while the crappy Kingston HyperX 3K SSD's read speed is around 500 MB..hmm...


What's your write speeds according to Crystal Disk Mark and AS SSD Benchmark


would appreciate a few screen shots and please tell me if you are using the IRST dirver and which version or just the chipset drivers?

I'm on an amd platform I can't run Irst drivers and I don't benchmark my ssd. I just don't care enough.

 

[Berryracer]

I'm on an amd platform I can't run Irst drivers and I don't benchmark my ssd. I just don't care enough.

cmon bro can u please run them once just for me to decide if it's worth while getting it?

 

[zokudu]

You know you're never going to be happy so why even bother. Done on my laptop, which is Sata2.

 

[Cerb]

Can you simplify this in layman's terms please?

http://en.wikipedia.org/wiki/Wear_leveling
http://en.wikipedia.org/wiki/Write_amplification

When you write new data (or send TRIM after deletion), data that was written is now re-usable, but the whole block needs to be erased, even though not all of the data may be unused. What blocks are going to be made dead data is effectively random (not exactly, but it's not highly predictable, by any stretch).

So you might have a set of pages with mixed live (L) and dead (D) data:

LLLDLLDDLDDDLDDD

To write new data, if there aren't any spare erase blocks, all those "L" blocks will need to be remapped and moved to another place, so the new block might look like so:

LLLLLLL---------

So, that's around 1.5 WA (1x from the initial writing of the whole block, .5 from moving the in-use data to another one). Let's say those blocks are all pretty compressible (C), and can be reduced to 3/7ths:

CCC--------------

Well, now, that's more like 1.2 WA.
Now, let's say you're saving uncompressed new (N) text data for an mbox inbox, and it's 6 pages worth:

LLLLLLLNNNNNN---

Now, if that were compressed (c), it might be more like:

CCCcc------------

So that's 7 pages of old host writes, taking 10 total writes so far, and 6 new pages of host writes taking 2, offering around 0.9 WA.

 

[BrightCandle]

Here is an M4 512GB drive and its performance numbers on X79:



Its decidedly a last generation drive now and when it came out it was a middle road SSD, its performance on writes has always been low. But then very few people use their SSD because it has high sustained write speed, its the sustained read and 4k reads that really matter to the feeling of performance on a modern PC as you do those most often. Crucial do have an update out with the M500 but it doesn't really perform much better, its bigger and cheaper for similar levels of performance.

A few drives do handle highly compressed files at high speeds for sustained transfers, you need to obviously avoid the Sandforce drives but there are other chipsets out there that perform well and deliver good sustained writes and often don't loose out too much in other areas. Here are 3 drives that often do well in sustained writes.

Samsung 840 Pro

Samsung 830

OCZ Vector

I think based on what you want from your drive (copying large compressed files) you may find the OCZ Vector is the cheapest of the reasonable options. However I will add I don't really trust OCZ, they have a bad history stretching back a decade of dishonest speccing of their products. So while no big scandal has broken for the Vector it might one day be proved to be fatally flawed in some way. Its been out long enough now that I suspect its not a complete lemon or highly unreliable but just be aware of their history and the companies financial difficulties.

 

[Berryracer]

I just ordered an ASUS RT-AC66u wireless router and a Samsung Pro 512 GB SSD! At last I found an online retailer that has these in stock in Dubai!

 

[johny12]

It depends on the data you're dealing with. Some data formats are more compressible than others and what SandForce does is to take advantage of the compressibility of data to improve performance (basically they are taking e.g. 100MB of data but compressing it to 80MB before writing to NAND, which means they need to write less than other SSDs).

Usually data is not fully compressible (like ATTO uses for testing) but not totally incompressible (like AS-SSD and Crystal use) either. That means your performance falls in-between the numbers you've seen in benchmark apps, although this is scenario dependent. For example movies are often already in a compressed format (lossless files take soooo much space, that's why), which means SandForce cannot compress the data anymore and that's the reason for poor performance.

Hellhammer, it is just a perception that incompressible data will cause slow performance. Have a look @ this link which has techical data sheet info on Intel 520 SSD explaining how intelligently Intel uses SandForce controllers to optimize data compression.

 

[Hellhammer]

Hellhammer, it is just a perception that incompressible data will cause slow performance. Have a look @ this link which has techical data sheet info on Intel 520 SSD explaining how intelligently Intel uses SandForce controllers to optimize data compression.

Link?