Long term performance of SSDs. Analysis of Intel SSDs.

[fuzzybabybunny]

Just read this recent article:

Long-term performance analysis of Intel Mainstream SSDs
http://www.pcper.com/article.php?aid=669

It shows very significant performance decreases as SSDs are used for longer periods of time.

I've currently got a 64GB Runcore 1.8" ZIF PATA MLC (80/45 MB/s read/write) on order to replace my slow internal 1.8" rotary drive (25MB/s read/write). Considering the performance hit that SSDs take over time, should I cancel my order and wait for SSD tech to get better? I don't want to pay $200 for a smaller capacity drive only to have it get significantly slower over time.

 

[Idontcare]

Wow. That is eye-opening to the extreme.

PC Perspective just "Anandteched" the Intel SSD's.

Welcome to OCZ's world Intel. You've been served. :laugh:

 

[bradley]

This information was previously documented in a few places.

http://www.xtremesystems.org/f...howthread.php?t=213226

Perhaps recent pricedrops will further expose limitations of Intel's SSD to a wider audience -- while overall limitations of SSD tech itself is better known. This further demonstrates the mistaken importance placed on theoretical performance (and synthetic benches) vs. everyday real world usage. Tony (OCZ) likely also deserves a mea culpa, unless posing competition against huge multinational corporations is a crime.... which seemingly is the overall impression reinforced on this forum, and others like it.

 

[Idontcare]

Originally posted by: bradley
This information was previously documented in a few places.

http://www.xtremesystems.org/f...howthread.php?t=213226

Yeah IntelUser2000 did a pretty decent job documenting it here on these forums as well.

Just as stuttering was "outed" to varying degrees in forums across the web prior to Anand's article, the PC Per article on the subject is largely value-add for providing closure to existing Intel SSD owners who feel their systems weren't performing to expectation. Now they know their expectations were set to high by Intel marketing and early reviews (as were OCZ customers with their SSD's).

Originally posted by: bradley
Tony (OCZ) likely also deserves a mea culpa, unless posing competition against huge multinational corporations is a crime.... which seemingly is the overall impression reinforced on this forum, and others like it.

Two wrongs don't make a right.

OCZ long held onto the view that the customers were inept and not properly using/characterizing their SSD's when the initial reports of stuttering came to the surface. Just because Intel's drives suffer a performance affliction over time doesn't mean OCZ's are suddenly going to stop stuttering now.

All I see here is we now have yet another SSD SKU to add to the pile of "if you buy this drive, be aware of these unadvertised performance caveats...".

I'm not really following your last statement there. Got any examples?

 

[coolVariable]

I would be curious whether this also applies to the Samsung drives (e.g. the Corsair S128).

And I am sorry, bradly, but the OCZ drives to date remain crap!

 

[BTRY B 529th FA BN]

This makes my recently purchased vRaptor more appealing and "the right choice" x

 

[frostedflakes]

Pretty odd. Sounds like this isn't an issue with SSDs in general, but something to do with the Intel controller.

FWIW, I've been running XP on a Samsung MLC drive for a few months and performance doesn't seem to have changed much. Still about 100/80 for large read/write.

http://xs136.xs.to/xs136/09076/atto_samsung491.png

 

[Zap]

Originally posted by: BTRY B 529th FA BN
This makes my recently purchased vRaptor more appealing and "the right choice" x

+1

I've gotten three VR 300GB and now I regret it even less. I'll let other people be guinea pigs for SSDs.

 

[Tiamat]

Originally posted by: Zap

Originally posted by: BTRY B 529th FA BN
This makes my recently purchased vRaptor more appealing and "the right choice" x

+1

I've gotten three VR 300GB and now I regret it even less. I'll let other people be guinea pigs for SSDs.

Looks like I will be amending my build to include the VR 300. I'll just have to remove it from its 3.5" adapter and keep it in AAM so that it doesn't exceed my noise level requirements (21dB)

Edit, nevermind. Seems to be high amount of sample variation in the VR300. Guess I'll just have to hold out for now.

 

[IntelUser2000]

I'm pretty sure you guys are about now bashing about the performance, but the reliability claims by Intel looks very correct.

Think about this. There is two things to mitigate the 10K write cycle limitation:

-Write Levelling
-Lower Write Amplilfication

Write levelling means that every part of the drive is made to be written evenly. After writing 100 times in average to the entire drive, cell #1053 should be on 105 writes, cell #2838 should be on 102 writes on an ideal drive. A bad write levelling drive would mean one of the cell would be on the 300th write while the average is only 100, deviating from the average greatly.

Of course. But what if its handling it in blocks anyway?? It means sure every block is being written evenly. But blocks are divided into much smaller portions called pages.

Write Amplification is basically the ratio of actual amount written to the SSD and the amount asked to be written by the controller.

If 32KB is written but the block size is 512KB and the SSD really writes 512KB the write amplification is 512/32 or factor of 16.

That's where the write combining algorithm of the Intel drive comes into play.

Downside is there of course but I think significant amount of the users wouldn't experience the worst-case performance scenario.

Most people probably don't need the extensive reliability features and live with cheaper SSDs. The basis of the reliability over absolute performance mentality comes because the Enterprise-oriented X25-E uses the exact same controller as the consumer-oriented X25-M. The enterprise users would probably see their money's worth in the reliability features of the Intel drives.

EDIT: The X25-M is also the lowest power SSD out there. Some things will have to be sacrificed to lower power.

 

[BTRY B 529th FA BN]

The technology just needs to improve for me to go head first into a high $$$ per GB ratio drive

 

[Idontcare]

Originally posted by: IntelUser2000
If 32KB is written but the block size is 512KB and the SSD really writes 512KB the write amplification is 512/32 or factor of 16.

http://www.anandtech.com/cpuch...howdoc.aspx?i=3403&p=3

Just posting a link which supports exactly what you said in case would-be lurkers want to read up on write amplification in more depth.

 

[n7]

Mine's gone to crap.
I get 30 MB/s write for everything now...including just copying files onto the drive.

It's pretty frustrating to spend $550 & then realize the performance advertised only really occurs when you use it as an empty benchmark drive...

Here's my numbers today.
http://ense7en.com/pics/Intel%...iskMark%2002.15.09.JPG

Yay for 30 MB/s max write.

The drive still feels snappy & amazingly fast loading stuff, but i see even my read numbers have gotten worse.

 

[Zap]

Originally posted by: Tiamat
Looks like I will be amending my build to include the VR 300. I'll just have to remove it from its 3.5" adapter and keep it in AAM so that it doesn't exceed my noise level requirements (21dB)

Edit, nevermind. Seems to be high amount of sample variation in the VR300. Guess I'll just have to hold out for now.

What's this sample variation that you speak of?

If you really don't want the IcePAK then you can buy a version of the VelociRaptor without it. Tough to find, but it is available.

WD3000BLFS

This ultra fast 2.5-inch hard drive is engineered to fit into blade servers and other enterprise storage configurations.

Cheapest I've seen it for is $280, but it might be worth the $50 difference to retain the 5 year warranty.

I've got three VelociRaptor 300GB. Two are the original style and the thirs is the "backplane-ready" which has a little PCB to relocate the SATA ports. They are all very quiet to my recovering-quiet-freak ears.

 

[frostedflakes]

Originally posted by: n7
Mine's gone to crap.
I get 30 MB/s write for everything now...including just copying files onto the drive.

It's pretty frustrating to spend $550 & then realize the performance advertised only really occurs when you use it as an empty benchmark drive...

Here's my numbers today.
http://ense7en.com/pics/Intel%...iskMark%2002.15.09.JPG

Yay for 30 MB/s max write.

The drive still feels snappy & amazingly fast loading stuff, but i see even my read numbers have gotten worse.

People are too obsessed with benchmarks IMO. If the drive feels just as fast as it always has, I wouldn't worry too much about it.

 

[Tiamat]

Originally posted by: Zap

Originally posted by: Tiamat
Looks like I will be amending my build to include the VR 300. I'll just have to remove it from its 3.5" adapter and keep it in AAM so that it doesn't exceed my noise level requirements (21dB)

Edit, nevermind. Seems to be high amount of sample variation in the VR300. Guess I'll just have to hold out for now.

What's this sample variation that you speak of?

If you really don't want the IcePAK then you can buy a version of the VelociRaptor without it. Tough to find, but it is available.

WD3000BLFS

This ultra fast 2.5-inch hard drive is engineered to fit into blade servers and other enterprise storage configurations.

Cheapest I've seen it for is $280, but it might be worth the $50 difference to retain the 5 year warranty.

I've got three VelociRaptor 300GB. Two are the original style and the thirs is the "backplane-ready" which has a little PCB to relocate the SATA ports. They are all very quiet to my recovering-quiet-freak ears.

Sample variation in terms of noise from SPCR's review sample. This was covered in the thread over there. I will probably get the 150GB version and buy a Novibes for 2.5" drives.

 

[coolVariable]

Originally posted by: frostedflakes

Originally posted by: n7
Mine's gone to crap.
I get 30 MB/s write for everything now...including just copying files onto the drive.

It's pretty frustrating to spend $550 & then realize the performance advertised only really occurs when you use it as an empty benchmark drive...

Here's my numbers today.
http://ense7en.com/pics/Intel%...iskMark%2002.15.09.JPG

Yay for 30 MB/s max write.

The drive still feels snappy & amazingly fast loading stuff, but i see even my read numbers have gotten worse.

People are too obsessed with benchmarks IMO. If the drive feels just as fast as it always has, I wouldn't worry too much about it.

Agree.
So does it effect normal use or is it only something that shows up in benches?

And I still would like to see a review whether this applies to the Samsung SSDs (e.g. Corsair).

 

[n7]

Originally posted by: n7
I get 30 MB/s write for everything now...including just copying files onto the drive

So yes, i notice it for real world stuff now too.

Accessing stuff on the drive is still great, though the fact those numbers have dropped isn't good either.

But writing to has gotten worse than half of what it was at the very beginning.

 

[Red Squirrel]

This is why I'm not even considering SSDs for production use yet. It's a great concept for expensive disposable storage, but with pollution and crap, the last thing we need is another disposable item.

 

[EarthwormJim]

Originally posted by: RedSquirrel
This is why I'm not even considering SSDs for production use yet. It's a great concept for expensive disposable storage, but with pollution and crap, the last thing we need is another disposable item.

I don't see how it is more disposable than a Hard Disk. They do fail too, and have wear and tear associated with usage.

 

[FireChicken]

Originally posted by: Zap

Originally posted by: Tiamat
Looks like I will be amending my build to include the VR 300. I'll just have to remove it from its 3.5" adapter and keep it in AAM so that it doesn't exceed my noise level requirements (21dB)

Edit, nevermind. Seems to be high amount of sample variation in the VR300. Guess I'll just have to hold out for now.

What's this sample variation that you speak of?

If you really don't want the IcePAK then you can buy a version of the VelociRaptor without it. Tough to find, but it is available.

WD3000BLFS

This ultra fast 2.5-inch hard drive is engineered to fit into blade servers and other enterprise storage configurations.

Cheapest I've seen it for is $280, but it might be worth the $50 difference to retain the 5 year warranty.

I've got three VelociRaptor 300GB. Two are the original style and the thirs is the "backplane-ready" which has a little PCB to relocate the SATA ports. They are all very quiet to my recovering-quiet-freak ears.

CDW has them but they are pricey

 

[FireChicken]

So does Intels warranty apply to these drives drop in performance.

"...it will be free from defects in material and workmanship, and will substantially conform to Intel?s publicly available specifications for a period of three (3) years beginning on the date the Product was purchased in its original sealed packaging in the case of an Original Purchaser,...."

Intels warranty

Not sure what intels version of "substantial" is...

 

[Cookie Monster]

Guess us "peasant" folks aren't missing out on anything yet with the SSDs.

This sure does make the vRaptors more attractive than it use to.

 

[IntelUser2000]

Originally posted by: RedSquirrel
This is why I'm not even considering SSDs for production use yet. It's a great concept for expensive disposable storage, but with pollution and crap, the last thing we need is another disposable item.

See comments like this corrupts the thread. People who think like this probably don't need anything faster than regular HDD in the first place.

Everyone thinks the Vertex or the next upcoming SSD will completely solve everything PLUS the benefits of the SSD technology. Oh how hurt they will be after they find out technology advances aren't big as they thought to be.

There is no free lunch.

See the conclusion is the Intel X25-M is still an excellent SSD drive. It won't freeze like the cheap SSD drives just by doing simple everyday duties like instant messaging. Slowdown is there because the controller ensures that every place is written evenly so maybe in few years people won't come back and say I can't use the drive because its dead. It still offers all the benefits of an SSD which the main thing is the read access times are so low and your computer is responsive.

Dead drives offer 0MB/s.

 

[Tiamat]

Originally posted by: FireChicken

Originally posted by: Zap

Originally posted by: Tiamat
Looks like I will be amending my build to include the VR 300. I'll just have to remove it from its 3.5" adapter and keep it in AAM so that it doesn't exceed my noise level requirements (21dB)

Edit, nevermind. Seems to be high amount of sample variation in the VR300. Guess I'll just have to hold out for now.

What's this sample variation that you speak of?

If you really don't want the IcePAK then you can buy a version of the VelociRaptor without it. Tough to find, but it is available.

WD3000BLFS

This ultra fast 2.5-inch hard drive is engineered to fit into blade servers and other enterprise storage configurations.

Cheapest I've seen it for is $280, but it might be worth the $50 difference to retain the 5 year warranty.

I've got three VelociRaptor 300GB. Two are the original style and the thirs is the "backplane-ready" which has a little PCB to relocate the SATA ports. They are all very quiet to my recovering-quiet-freak ears.

CDW has them but they are pricey

Thanks for the link. The 150BLFS seems to be for 210$. While the price is somewhat steep, at this point, I don't really care all that much. Otherwise, Id have to buy a torq No. 8 I think, and risk killing the warranty.

Just checked on pricewatch, there are a couple other providers like ewiz that have it for $183 shipped