cda
07-28-2008, 07:04 PM
Hi there,
we have tested some of the OCZ Core 128 MLC SSDs for sustained write and have had very disappointing results.
I guess the real question I have is whether MLC SSDs *should* be able to support a sustained minimum write speed of at least 30mb/sec for very long writes, and whether you all have tested any that would meet our requirements. We may have to opt for the higher priced but seemingly reliable SLC drives.
thanks.
Charles
below for context, my nth email to OCZ tech support, which has been dodging the question of real expected sustained transfer rates on the Core series drives:
"Primarily we used a utility from Aja, our video capture card's manufacturer, to observe the write speed over time...it simply measures the rate that data is being transferred, so it's objective in that sense. We also used HD2 with identical results.
The behavior is the same regardless of the system specs, controllers, disk mode, AHCI etc. As noted elsewhere, XP install stability is best with ACHI and cacheing off, as would be expected.
The write speed starts off at the full write speed expected for the disk mode and/or raid setup, then begins to fall at around 1GB and reaches a fairly constant 2-5mb/sec by 4GB or so into the write. There are also regular periodic CPU spikes to 100% from the "normal" of 60% during a capture with a core2 duo or 20% during a capture wth a Quad.
The same thing happens copying a large file, so it's unrelated to video capture per se.
The way we noticed the CPU spikes and slowing write speed was that the capture would freeze for increasing periods after the filesize reached 1-2 GB, and the time between freezes grew shorter over time.
If it were a small variation on the spec'd "best case" write speed, I wouldn't be concerned since as with all things "results may vary", and such variation could possibly be accounted for in one's choice of method of measurement, but the fall in this case is way beyond any expected real-world performance.
We'd just like to know whether these drives are *supposed* to perform at some reasonable fraction, say 80-90% of their advertised rate. If we can't get at least a constant 30mb/sec over extended writes, we can't use them."
we have tested some of the OCZ Core 128 MLC SSDs for sustained write and have had very disappointing results.
I guess the real question I have is whether MLC SSDs *should* be able to support a sustained minimum write speed of at least 30mb/sec for very long writes, and whether you all have tested any that would meet our requirements. We may have to opt for the higher priced but seemingly reliable SLC drives.
thanks.
Charles
below for context, my nth email to OCZ tech support, which has been dodging the question of real expected sustained transfer rates on the Core series drives:
"Primarily we used a utility from Aja, our video capture card's manufacturer, to observe the write speed over time...it simply measures the rate that data is being transferred, so it's objective in that sense. We also used HD2 with identical results.
The behavior is the same regardless of the system specs, controllers, disk mode, AHCI etc. As noted elsewhere, XP install stability is best with ACHI and cacheing off, as would be expected.
The write speed starts off at the full write speed expected for the disk mode and/or raid setup, then begins to fall at around 1GB and reaches a fairly constant 2-5mb/sec by 4GB or so into the write. There are also regular periodic CPU spikes to 100% from the "normal" of 60% during a capture with a core2 duo or 20% during a capture wth a Quad.
The same thing happens copying a large file, so it's unrelated to video capture per se.
The way we noticed the CPU spikes and slowing write speed was that the capture would freeze for increasing periods after the filesize reached 1-2 GB, and the time between freezes grew shorter over time.
If it were a small variation on the spec'd "best case" write speed, I wouldn't be concerned since as with all things "results may vary", and such variation could possibly be accounted for in one's choice of method of measurement, but the fall in this case is way beyond any expected real-world performance.
We'd just like to know whether these drives are *supposed* to perform at some reasonable fraction, say 80-90% of their advertised rate. If we can't get at least a constant 30mb/sec over extended writes, we can't use them."