G.Skill Titan 256GB SSD

[imported_TimBob]

Originally posted by: aka1nas
It applies to all SSDs. The performance does not continue to degrade after successive writes, it's just that every write to a block after the first time has to also do that cache+erase operation. I think that some manufacturers may also do a full write cycle prior to shipping the drive, so write performance in those drives would be static.

Why aren't the blocks just cleared when data is deleted? Why does it happen at write time?

Edit: Is that what diskeeper's hyperfast is attempting to do behind the scenes? Found this quote regarding hyperfast.
"Diskeeper and Apacer developed the Hyperfast Flash technology for the new SSD+ Optimizer product. This combines to offer a SSD that intelligently eliminates free space fragmentation, and the performance degradation that comes with it. In addition, Hyperfast Flash technology also reduces the erase-write cycles, which in effect, will enhance reliability and extend SSD life spans. "

 

[coolVariable]

Also sems like not everything is well in g.skill-ville:

http://ht4u.net/reviews/2009/g...n_256gb_ssd/index7.php

The last three benches are writing access times.

Avg. write access time is 100x SLOWER than the intel
Min. write access time is 25x times slower than the intel
Max. write access time is 1.5x times slower than the intel (and 20x times that of a Western Digital HDD)


And these ATTO tests also do not bode well:
http://ht4u.net/reviews/2009/g..._256gb_ssd/index14.php

 

[Denithor]

That's why I'm waiting to see how AT likes the drive.

 

[Viper GTS]

Originally posted by: TimBob

Originally posted by: aka1nas
It applies to all SSDs. The performance does not continue to degrade after successive writes, it's just that every write to a block after the first time has to also do that cache+erase operation. I think that some manufacturers may also do a full write cycle prior to shipping the drive, so write performance in those drives would be static.

Why aren't the blocks just cleared when data is deleted? Why does it happen at write time?

Edit: Is that what diskeeper's hyperfast is attempting to do behind the scenes? Found this quote regarding hyperfast.
"Diskeeper and Apacer developed the Hyperfast Flash technology for the new SSD+ Optimizer product. This combines to offer a SSD that intelligently eliminates free space fragmentation, and the performance degradation that comes with it. In addition, Hyperfast Flash technology also reduces the erase-write cycles, which in effect, will enhance reliability and extend SSD life spans. "

Erasing a block when files are deleted would burn a write cycle, effectively doubling the write amplification of the drive.

It's just an inherent limitation of flash, there is no way around it. You can reset a drive to its new state by doing a secure erase, but as soon as everything is written again you're right back where you started.

Viper GTS

 

[aka1nas]

Originally posted by: TimBob

Originally posted by: aka1nas
It applies to all SSDs. The performance does not continue to degrade after successive writes, it's just that every write to a block after the first time has to also do that cache+erase operation. I think that some manufacturers may also do a full write cycle prior to shipping the drive, so write performance in those drives would be static.

Why aren't the blocks just cleared when data is deleted? Why does it happen at write time?

Edit: Is that what diskeeper's hyperfast is attempting to do behind the scenes? Found this quote regarding hyperfast.
"Diskeeper and Apacer developed the Hyperfast Flash technology for the new SSD+ Optimizer product. This combines to offer a SSD that intelligently eliminates free space fragmentation, and the performance degradation that comes with it. In addition, Hyperfast Flash technology also reduces the erase-write cycles, which in effect, will enhance reliability and extend SSD life spans. "

You can't just append data to a block (IIRC, current SSDs have 8MB blocks) with an SSD, you have to first erase the block(and of course you have to hold anything that was already in that block somewhere) and then write the entire thing out again. So, if you are trying to write data anywhere in a block that already has something in it, you have to go through this cache+erase cycle first. Intel calls this effect Write Amplification, and is something that their controller is supposed to significantly minimize.

I'm not sure how completely empty blocks behave, I would hope that the controller can be smart about it and not have to do an erase cycle then.

 

[imported_TimBob]

Thanks for the explanations.

 

[imported_TimBob]

Finally arrived. Vista 64 sp1 installed in 20 minutes. Unfortunately, I probably wont be able to play with it too much more until the kids/wife go to bed tonight.

 

[taltamir]

Originally posted by: Viper GTS

Originally posted by: TimBob

Originally posted by: aka1nas
It applies to all SSDs. The performance does not continue to degrade after successive writes, it's just that every write to a block after the first time has to also do that cache+erase operation. I think that some manufacturers may also do a full write cycle prior to shipping the drive, so write performance in those drives would be static.

Why aren't the blocks just cleared when data is deleted? Why does it happen at write time?

Edit: Is that what diskeeper's hyperfast is attempting to do behind the scenes? Found this quote regarding hyperfast.
"Diskeeper and Apacer developed the Hyperfast Flash technology for the new SSD+ Optimizer product. This combines to offer a SSD that intelligently eliminates free space fragmentation, and the performance degradation that comes with it. In addition, Hyperfast Flash technology also reduces the erase-write cycles, which in effect, will enhance reliability and extend SSD life spans. "

Erasing a block when files are deleted would burn a write cycle, effectively doubling the write amplification of the drive.

It's just an inherent limitation of flash, there is no way around it. You can reset a drive to its new state by doing a secure erase, but as soon as everything is written again you're right back where you started.

Viper GTS

Don't forget it also means erasing takes as long as writing the file originally. So a file that took a minute to write will take a minute to erase (instead of a fraction of a second).

even if you has an SLC, the block is 8MB, to prevent huge wastes of space, each block can contain more than one file... So a block containing 10 500kb files still has 3MB of free space, to write another 500kb file to it you need to read the current 5MB, erase the entire block, and then write back the original 5mb + the 500kb of new files.

There are a few other issues with "immidiately erasing"

 

[taltamir]

Originally posted by: aka1nas

Originally posted by: TimBob

Originally posted by: aka1nas
It applies to all SSDs. The performance does not continue to degrade after successive writes, it's just that every write to a block after the first time has to also do that cache+erase operation. I think that some manufacturers may also do a full write cycle prior to shipping the drive, so write performance in those drives would be static.

Why aren't the blocks just cleared when data is deleted? Why does it happen at write time?

Edit: Is that what diskeeper's hyperfast is attempting to do behind the scenes? Found this quote regarding hyperfast.
"Diskeeper and Apacer developed the Hyperfast Flash technology for the new SSD+ Optimizer product. This combines to offer a SSD that intelligently eliminates free space fragmentation, and the performance degradation that comes with it. In addition, Hyperfast Flash technology also reduces the erase-write cycles, which in effect, will enhance reliability and extend SSD life spans. "

You can't just append data to a block (IIRC, current SSDs have 8MB blocks) with an SSD, you have to first erase the block(and of course you have to hold anything that was already in that block somewhere) and then write the entire thing out again. So, if you are trying to write data anywhere in a block that already has something in it, you have to go through this cache+erase cycle first. Intel calls this effect Write Amplification, and is something that their controller is supposed to significantly minimize.

I'm not sure how completely empty blocks behave, I would hope that the controller can be smart about it and not have to do an erase cycle then.

The anandtech article said that intel found a way to append data to blocks. Or at least claims to have done so.

 

[Old Hippie]

Looking back, the G. Skill makes Titan a very good impression. In our benchmark parcours can Solid State Drive with consistently high performance results convincing and makes Intel's X25-M, which is still a measure of all SSDs applies may, life difficult.

From the initial disadvantage of a MLC flash memory is from these results no longer be seen. On the contrary, the reading will coincide with the dual-controller concept again even better than the 64-GB SSD from. When writing reaches G. Skill large files even when the performance of a traditional hard drive and comes in FileCopy copy test even consider a WD Raptors RAID-0 approached. Only small block sizes - which are particularly interesting to write temporary files on your operating system around - she continues to lose.

Whether you are on the increase in energy demand should make sure, probably every user must decide for themselves. Compared to magnetic storage, he is still much lower - as compared to SSDs smaller but significant in the amount jumped. The increased technical complexity for the rest seem to care. Energy efficiency hears in this sense at least something different.

Unfortunately we are missing at this time, both information about the targeted price as well as on the availability in trade. These data are obviously after we, once we have definitive evidence from G. Skill available. Until that can be taken from the G. Skill Titan 256 GB SSD at least once a convincing performance to demonstrate results.

I think that's good. :laugh:

It does look like the page file needs disabled.

Newegg has raised the price by 10.00.

 

[aka1nas]

It was 279.99 when I ordered mine.

 

[Old Hippie]

Originally posted by: aka1nas
It was 279.99 when I ordered mine.

Well, horse poop.

I paid 300.00.

 

[JMapleton]

$500 for the a top of the lind dds is quite reasonable. It wasn't only a few years ago 500GB platter drivers were near that price.

 

[aka1nas]

I imaged the Vista install from my raptor and spent the last few hours tweaking. I had considerable stuttering initially, but after applying most of the non-extreme tweaks(i.e. no RAMdisks or Steadystate) from the OCZ guide performance is much better.

Firefox seems to be one of the biggest culprits for stuttering, as I think it is aggressively writing profile and history data to the disk by default. Installing more than one game at a time from Steam to the SSD is also causing a little stuttering.

I may try a clean install over the weekend just to make sure that imaging the OS over with Acronis didn't screw up partition settings.

 

[SteveGo]

Sounds like not much has changed. Time to wait and see if the OCZ Vertex might be the answer.

 

[aka1nas]

I moved my Firefox profile off of the SSD, and there is virtually no stutter now during normal use, even when installing multiple apps at once. I still have SuperFetch and Prefetch enabled on Vista, and I am thinking that a lot of the stuttering occurs when SuperFetch is still populating the cache. Once the machine has been up for a while, performance smooths out significantly.

Performance numbers from Crystalmark 2009:

[ HDD ] 24380
Read : 209.26 MB/s ( 6092)
Write : 145.15 MB/s ( 4903)
RandomRead512K : 187.92 MB/s ( 5758)
RandomWrite512K : 68.23 MB/s ( 2729)
RandomRead 64K : 96.54 MB/s ( 3861)
RandomWrite 64K : 25.94 MB/s ( 1037)

When I was still getting stuttering, Seq Writes were only benching at around 50-60MB/s on Crystal and most other benchmarks.

 

[coolVariable]

Sorry. I don't see any sense spending $300 for something that stutters. At $200 I might get weak about the Titan but at this price ...
I agree with SteveGo. I will pass and wait ... I am very doubtful about the Vertex (sound like Vaporware ... already over a 1 month late coming out, ugh). Maybe Samsung can wow us.

 

[taltamir]

Firefox seems to be one of the biggest culprits for stuttering, as I think it is aggressively writing profile and history data to the disk by default

Its the firefox crash protection... If firefox crashes / the computer loses power. You can restore the session, it will reopen all the tabs (even of pages that are not available), it will even keep a full "back" and "forward" list for each individual tab. This is a great feature when you need it.

 

[noname98967]

Why aren't reviewers testing for something as simple as this? Is it that hard to open Firefox and browse the web for a few minutes. Who cares if an SSD loads Crysis in two seconds if it can't even run a web browser without stuttering.

 

[taltamir]

thats the difference between a BAD review site and a good one . Anandtech has been good enough to point out this failure of SSDs. beware where you shop for advice.

 

[BlueAcolyte]

Originally posted by: aka1nas
I moved my Firefox profile off of the SSD, and there is virtually no stutter now during normal use, even when installing multiple apps at once. I still have SuperFetch and Prefetch enabled on Vista, and I am thinking that a lot of the stuttering occurs when SuperFetch is still populating the cache. Once the machine has been up for a while, performance smooths out significantly.

Performance numbers from Crystalmark 2009:

[ HDD ] 24380
Read : 209.26 MB/s ( 6092)
Write : 145.15 MB/s ( 4903)
RandomRead512K : 187.92 MB/s ( 5758)
RandomWrite512K : 68.23 MB/s ( 2729)
RandomRead 64K : 96.54 MB/s ( 3861)
RandomWrite 64K : 25.94 MB/s ( 1037)

When I was still getting stuttering, Seq Writes were only benching at around 50-60MB/s on Crystal and most other benchmarks.

Hmm, are you sure you want Superfetch/Prefetch still running? I thought that was one of the optimization tweaks (from the OCZ website) to disable them.

 

[taltamir]

Originally posted by: aka1nas
I moved my Firefox profile off of the SSD, and there is virtually no stutter now during normal use, even when installing multiple apps at once. I still have SuperFetch and Prefetch enabled on Vista, and I am thinking that a lot of the stuttering occurs when SuperFetch is still populating the cache. Once the machine has been up for a while, performance smooths out significantly.

Performance numbers from Crystalmark 2009:

[ HDD ] 24380
Read : 209.26 MB/s ( 6092)
Write : 145.15 MB/s ( 4903)
RandomRead512K : 187.92 MB/s ( 5758)
RandomWrite512K : 68.23 MB/s ( 2729)
RandomRead 64K : 96.54 MB/s ( 3861)
RandomWrite 64K : 25.94 MB/s ( 1037)

When I was still getting stuttering, Seq Writes were only benching at around 50-60MB/s on Crystal and most other benchmarks.

what ssd are you using?

 

[Old Hippie]

I had considerable stuttering initially

Just curious, how does the stuttering manifest itself?

Is the screen flickering or does everything get jerky?

Mine will be here on Monday but I'll be putting a fresh install of XP Pro on it.

 

[imported_TimBob]

Well, here's my results so far.
Just ran crystaldiskmark 2.2 w/500mb selected.

seq 229.4, 131.3.
Random 512k 185.9, 46.37
Random 4k 20.42, 1.566

But, my results seem all over the place. Often I'm only 80ish in sequential writes (maybe something running in the background that I'm not aware of). I've turned off defrag, indexing, pagefile and pre and supperfetch. Installed the x48 chipset drivers and Intel Matrix Storage Manager.

Unfortunately, I'm also experiencing some stuttering at times. Hippie, when the stuttering happens the system basically freezes for a couple seconds and is completely unresponsive. You just have to wait it out.

Other than the occasional stuttering which can be pretty annoying, I'm pretty happy with it. Everything loads so quickly.

Does it matter whether you set the controller in your BIOS to Raid, AHCI or IDE? I first installed with Raid since that's what I had before with my 2 74gb Raptors. I tried reinstalling with AHCI and didn't notice a difference.

 

[taltamir]

Can you please fill up the drive and then test it?