It's not similar at all. OCZ's performance mode had a strict 50% threshold -- either everything was SLC or MLC. In other words, once you filled over 50%, everything went MLC and any future writes would not get any performance increase. DWA still works after the 50% mark because it will move data to the MLC portion in the background before the SLC cache gets full, whereas OCZ's implementation kept everything as SLC until it become full.
Well, i've seen lots of reports of performance mode working just fine after 50% mark. It just needs some time to move data around and then it gets its performance back. Could be wrong tho.
Obviously, if you trash it with constant writes, it will drop performance after 50% mark. But it will regain performance back after some time. Unless those bechmarks lie somehow :)
If hammered with constant writes, the drop in write performance is bigger with OCZ's performance mode because the drive is still migrating data from SLC to MLC while processing the host writes. Once the migration has been completed, performance will recover, but not to the original clean state because the whole drive is now operating in MLC mode.
DWA will show similar behavior with constant writes, but with the difference that the SLC cache is still functional after the 50% mark (but just smaller).
>. Once the migration has been completed, performance will recover, but not to the original clean state because the whole drive is now operating in MLC mode.
Yeah, but i've seen numerous reports of Vertex4 back in the day, that had write speeds near its maximum even after 50% full. So it was either an error in benchmarking or OCZs performance mode actually treats remainder of free flash still as pSLC. It just waits till 50% before it starts doing anything at all to normal MLC. Atleast thats how i see. Either way, it would be interesting to see this benchmarked again in newer drive. As i understand they still use this. So fill the drive to 75%, wait some time (30m-1h) and benchmark again. If performance mode isn't functional beyond 50%, write speeds should stay at MLC levels.
"If performance mode isn't functional beyond 50%, write speeds should stay at MLC levels."
That could be the problem of those investigations. Do we know just how fast or slow the pure MLC write speeds are on those drives? Maybe they're not as slow as people think (pretty much everyone else is using pure MLC for a reason) tey should be, and hence performance mode may be off but since the performance measured is still pretty good people may think the drive is still performance mode. The effect of performance mode should also depend on the workload / benchmark.
Yeah, MLC gets pretty fast beyond 128GB (or 256GB in case of 128Gbit dies). Still, this whole 50% theory could be easily tested on 128GB drives. Afterall, they benefit the most. So if pSLC is still somewhat functional after 50%, this should be easily seen in benchmarks. If not, such ssd would have noticebly lower write speeds at stay 75% full, regardless of idle time.
Dynamic Write Acceleration can cause performance to fall off a cliff as the drive gets filled up. Take a look at http://www.micron.com/-/media/documents/products/t... and see the graph at the bottom of page 4. Write speed falls to what looks like about 50MB/sec once the drive is about 58% full. That graph is for the 128MB model however; hopefully worst-case write speeds won't be as bad for the higher-capacity units.
Oh, remember that DWA is only used for the 128GB and 256GB 2.5" models (and 512GB for m.2 & MSATA). So the 512GB and 1TB 2.5" models wouldn't be affected.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
15 Comments
Back to Article
hojnikb - Tuesday, September 16, 2014 - link
Hmm, that DWA looks very similar to OCZs performance mode, introduced back with vertex4...Hopefully, this will carry to their budget line aswell.
Kristian Vättö - Tuesday, September 16, 2014 - link
It's not similar at all. OCZ's performance mode had a strict 50% threshold -- either everything was SLC or MLC. In other words, once you filled over 50%, everything went MLC and any future writes would not get any performance increase. DWA still works after the 50% mark because it will move data to the MLC portion in the background before the SLC cache gets full, whereas OCZ's implementation kept everything as SLC until it become full.hojnikb - Tuesday, September 16, 2014 - link
Well, i've seen lots of reports of performance mode working just fine after 50% mark. It just needs some time to move data around and then it gets its performance back.Could be wrong tho.
hojnikb - Tuesday, September 16, 2014 - link
Obviously, if you trash it with constant writes, it will drop performance after 50% mark. But it will regain performance back after some time. Unless those bechmarks lie somehow :)Kristian Vättö - Tuesday, September 16, 2014 - link
If hammered with constant writes, the drop in write performance is bigger with OCZ's performance mode because the drive is still migrating data from SLC to MLC while processing the host writes. Once the migration has been completed, performance will recover, but not to the original clean state because the whole drive is now operating in MLC mode.DWA will show similar behavior with constant writes, but with the difference that the SLC cache is still functional after the 50% mark (but just smaller).
hojnikb - Tuesday, September 16, 2014 - link
>. Once the migration has been completed, performance will recover, but not to the original clean state because the whole drive is now operating in MLC mode.Yeah, but i've seen numerous reports of Vertex4 back in the day, that had write speeds near its maximum even after 50% full. So it was either an error in benchmarking or OCZs performance mode actually treats remainder of free flash still as pSLC. It just waits till 50% before it starts doing anything at all to normal MLC. Atleast thats how i see. Either way, it would be interesting to see this benchmarked again in newer drive. As i understand they still use this.
So fill the drive to 75%, wait some time (30m-1h) and benchmark again. If performance mode isn't functional beyond 50%, write speeds should stay at MLC levels.
MrSpadge - Tuesday, September 16, 2014 - link
"If performance mode isn't functional beyond 50%, write speeds should stay at MLC levels."That could be the problem of those investigations. Do we know just how fast or slow the pure MLC write speeds are on those drives? Maybe they're not as slow as people think (pretty much everyone else is using pure MLC for a reason) tey should be, and hence performance mode may be off but since the performance measured is still pretty good people may think the drive is still performance mode. The effect of performance mode should also depend on the workload / benchmark.
hojnikb - Tuesday, September 16, 2014 - link
Yeah, MLC gets pretty fast beyond 128GB (or 256GB in case of 128Gbit dies). Still, this whole 50% theory could be easily tested on 128GB drives. Afterall, they benefit the most. So if pSLC is still somewhat functional after 50%, this should be easily seen in benchmarks. If not, such ssd would have noticebly lower write speeds at stay 75% full, regardless of idle time.hojnikb - Tuesday, September 16, 2014 - link
say*really, still no edit ?
hojnikb - Tuesday, September 16, 2014 - link
... if the drive is indeed stuck in normal mode, until you make more than 50% free space.iwod - Tuesday, September 16, 2014 - link
Damn it, wake me up when there is an SSD with Marvell 88SS1093 ( PCI-E 3.0 x 4 )microlithx - Friday, September 19, 2014 - link
Gonna be about a year on that one...jamesd1999 - Wednesday, September 17, 2014 - link
"The rating also scales linearly with capacity now"... Right...http://oi61.tinypic.com/4l2jom.jpg
http://oi58.tinypic.com/256b9zs.jpg
Well, there IS a line...
Donuts123 - Sunday, September 21, 2014 - link
Dynamic Write Acceleration can cause performance to fall off a cliff as the drive gets filled up. Take a look athttp://www.micron.com/-/media/documents/products/t...
and see the graph at the bottom of page 4. Write speed falls to what looks like about 50MB/sec once the drive is about 58% full. That graph is for the 128MB model however; hopefully worst-case write speeds won't be as bad for the higher-capacity units.
Donuts123 - Sunday, September 21, 2014 - link
Oh, remember that DWA is only used for the 128GB and 256GB 2.5" models (and 512GB for m.2 & MSATA). So the 512GB and 1TB 2.5" models wouldn't be affected.