I want to see results for the weakest spot of SSD technology, that is latency of small random writes (example - http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3403&p=8 ). Without it any SSD test is quite meaningless.
Did you do any specific tests with regard to small random writes? The first couple of OCZ drives with the JMicron controller, the Core and Core V2 series, had absolutely horrible stuttering/hanging problems with small writes, leading to (in some cases) a maximum write latency of almost a second, and regular hanging of the OS.
I don't want to stick my neck out because I agree that the Apex drive seems a bit erratic in its performance but this part of the article is just plain wrong.
"Our initial impressions of the Apex were mixed. The figures from HD Tach 3 show the average write speed - 80MB/s - is barely faster than a decent hard drive and lags far behind the Intel X25-M, which is 70 per cent faster, at 136MB/s."
You seem to have reversed the scores here - the Apex drive trounced the Intel one on average write speed. It is the Intel drive that is "barely faster than a decent hard drive". This is clear in both the HD Tach screenshots and the Average Read and Write Speeds graph on the following page.
I just wanted to put that right because I really appreciate what OCZ are doing for the SSD market - creating more affordable drives with awesome performance. As stated in the article, the huge problems people have with these drives are caused by Windows not being up to scratch. I'm sick of hearing people saying Intel are the only performance option in SSDs. Keep it up OCZ!
When it comes to burst speed, the Apex shoots to the front of the pack by a margin of ten per cent. This is such an improvement over the average read and write speeds that we have to see it as evidence of the RAID feature in action.
And yet The chart shows burst speeds of X25=260 and Apex=228....
Apologies, all. The graphs present the correct HD Tach numbers - and now so does the review text. It doesn't radically alter the outcome of the evaluation, and the Verdict and Rating stand.
Apologies for this folks - the figures on the graphs are correct however I managed to trip over the reams test results and make some silly mistakes in the copy which Tony has sorted out.
ref Bronek's comments: Tony and self talked about hard drive testing a while back and came to the conclusion that benchmarks are all well and good but file transfer tests are more real world.
Transferring files and timing the results should reflect any funny business such as excessive latency. If this is not the case I'd be happy to hear the flaw along with any suggestions about how we can realistically measure it.
I think here you will find just the right tool (admittedly, benchmarking is not its purpose) : http://managedflash.com/downloads/index.htm . To see what I mean, download the documentation http://downloads.managedflash.com/documentation/090107_windowsinstall.pdf and jump to bottom part of page 22 (section 2.8).
Alternatively, just adopt any SQL IO performance test - they all measure random writes performance pretty well. BTW, I hope this explains why random writes latency is important.
WD My Passport Essential 500GB Portable Hard Drive - Black (USB 2.0)
WD Caviar Black 1TB Hard Drive (Serial ATA-300, 7,200 RPM, 32MB)
WD My Book Essential Edition External 1TB Hard Drive (USB 2.0, 16MB)
Intel X25-M Gen2 160GB 2.5" Solid State Drive (SATA)
WD My Passport Essential Portable 320GB Hard Drive - Black (USB 2.0)
OCZ Apex 120GB SSD
Bronek Kozicki
sigh, the most meaningful figure is missing #
Posted Wednesday 25th February 2009 10:18 GMT
I want to see results for the weakest spot of SSD technology, that is latency of small random writes (example - http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3403&p=8 ). Without it any SSD test is quite meaningless.
MarkP
Odd behaviour? #
Posted Wednesday 25th February 2009 10:18 GMT
Could the odd behaviour of the OCZ be down to the internal RAID?
Simon Ball
Random writes... #
Posted Wednesday 25th February 2009 10:47 GMT
Did you do any specific tests with regard to small random writes? The first couple of OCZ drives with the JMicron controller, the Core and Core V2 series, had absolutely horrible stuttering/hanging problems with small writes, leading to (in some cases) a maximum write latency of almost a second, and regular hanging of the OS.
tadju
A performance reversal. #
Posted Wednesday 25th February 2009 10:47 GMT
I don't want to stick my neck out because I agree that the Apex drive seems a bit erratic in its performance but this part of the article is just plain wrong.
"Our initial impressions of the Apex were mixed. The figures from HD Tach 3 show the average write speed - 80MB/s - is barely faster than a decent hard drive and lags far behind the Intel X25-M, which is 70 per cent faster, at 136MB/s."
You seem to have reversed the scores here - the Apex drive trounced the Intel one on average write speed. It is the Intel drive that is "barely faster than a decent hard drive". This is clear in both the HD Tach screenshots and the Average Read and Write Speeds graph on the following page.
I just wanted to put that right because I really appreciate what OCZ are doing for the SSD market - creating more affordable drives with awesome performance. As stated in the article, the huge problems people have with these drives are caused by Windows not being up to scratch. I'm sick of hearing people saying Intel are the only performance option in SSDs. Keep it up OCZ!
Josh
SSD friendly OS #
Posted Wednesday 25th February 2009 11:52 GMT
Is Mac OS X 'SSD friendly', as per the configs you can buy from Apple with SSDs?
Would be interesting to see results here.
Anonymous Coward
Geeze... #
Posted Wednesday 25th February 2009 11:52 GMT
When it comes to burst speed, the Apex shoots to the front of the pack by a margin of ten per cent. This is such an improvement over the average read and write speeds that we have to see it as evidence of the RAID feature in action.
And yet The chart shows burst speeds of X25=260 and Apex=228....
Please clarify!
Tony Smith, Editor, Reg Hardware
@All #
Posted Wednesday 25th February 2009 12:20 GMT
Apologies, all. The graphs present the correct HD Tach numbers - and now so does the review text. It doesn't radically alter the outcome of the evaluation, and the Verdict and Rating stand.
Leo Waldock
Figures, figures #
Posted Wednesday 25th February 2009 12:57 GMT
Apologies for this folks - the figures on the graphs are correct however I managed to trip over the reams test results and make some silly mistakes in the copy which Tony has sorted out.
ref Bronek's comments: Tony and self talked about hard drive testing a while back and came to the conclusion that benchmarks are all well and good but file transfer tests are more real world.
Transferring files and timing the results should reflect any funny business such as excessive latency. If this is not the case I'd be happy to hear the flaw along with any suggestions about how we can realistically measure it.
Bronek Kozicki
testing random writes on SSD? #
Posted Monday 2nd March 2009 10:05 GMT
I think here you will find just the right tool (admittedly, benchmarking is not its purpose) : http://managedflash.com/downloads/index.htm . To see what I mean, download the documentation http://downloads.managedflash.com/documentation/090107_windowsinstall.pdf and jump to bottom part of page 22 (section 2.8).
Alternatively, just adopt any SQL IO performance test - they all measure random writes performance pretty well. BTW, I hope this explains why random writes latency is important.
Bronek Kozicki
... or just ... #
Posted Monday 2nd March 2009 13:17 GMT
use www.iometer.org with 4KB random writes.