OCZ's new Vertex 4 SSD: Faster... and slower
You win some, you lose some
OCZ has a new SSD that's both faster and slower than the previous model. That's what can happen when you swap controllers on an SSD.
A year ago the hustling and bustling SSD supplier introduced its Vertex 3 with a bought-in SandForce SF-2281 controller inside. Since then LSI has bought SandForce and OCZ has bought its own controller company: Indilinx. That company's Everest 2 controller, using two ARM chips, is what drives the new Vertex 4 SSD.
The Vertex 4 comes in 128GB, 256GB and 512GB versions. It has a five-year warranty compared to the preceding Vertex 3's three-year warranty stretch.
Both the Vertex 3 and 4 use synchronous MLC NAND chips. We can compare their I/O numbers, looking at the maximum IOPS and MB/sec rates for simplicity:
A couple of charts make the differences clearer:
Vertex 4 is faster than Vertex 3, particularly with regard to random reads.
But Vertex 4 is not faster than Vertex 3 bandwidth-wise; in fact it is quite a bit slower at sequential reading. It appears that OCZ has optimised the thing for random I/O and been unable to get sequential I/O up to Vertex 3 levels, which is a pity.
OCZ fleshes the differences out in Vertex 4's favour, saying that its latency is "as low as 0.04ms for read operations and 0.02ms for write operations, delivering an improvement of approximately 80 per cent over the Vertex 3". The SSD-maker then claims that "[i]n typical use case scenarios, the Vertex 4 outperforms the Vertex 3 by as much as 400 per cent."
All in all though, these are pretty decent I/O numbers and we might look forward to a go-faster Vertex 4 Max IOPS edition. The controller does the usual wear-levelling, garbage collection, error correction and so on, which enables OCZ to offer its five-year warranty.
OCZ does not say anything about performance consistency over time, although it does assert: "[Vertex 4] consistently accomplishes superior real world performance over the long term regardless of whether the data streams are in compressed or uncompressed formats."
The company is expanding its SSD product range energetically and introducing new products and refreshing existing ones at an incredible rate. The chances are that it will be a survivor in the almost inevitable flash drive supplier shakeout that's coming along in a few years' time.
We see two blots on its escutcheon. Firstly, there appear to be support issues with consumers saying they get faulty drives which are fixed slowly or not at all – and that refunds are tricky to get. Secondly the drives can feature fresh-out-of-the-box performance that may tail off. If OCZ can get product quality and support issues resolved, and provide consistent drive performance over the long term, then there's probably no stopping the outfit.
The Vertex 4 SSDs ship this month and are priced at $179 for the 128GB, $349 for the 256GB and $699 for the 512GB. ®
Non-zero Y-axis and 'Quite a bit slower'?
"But Vertex 4 is not faster than Vertex 3 bandwidth-wise; in fact it is quite a bit slower at sequential reading. It appears that OCZ has optimised the thing for random I/O and been unable to get sequential I/O up to Vertex 3 levels, which is a pity."
Look, maybe it's me, but when you graph stuff (looking at second graph here) not from a zero Y-axis you should say so. Especially given the previous graph is from a zero Y-axis. I expect better from The Register.
Also, from your numbers (550 Vs. 535 MB/sec), I would not say that it is 'quite a bit' slower given it is around 3%. Yes slower, but 'quite a bit'?
No, I am no OCZ fanboi, and no, I don't have an SSD (yet) - but I am looking at these products and I appreciate balanced reports.
"in fact it is quite a bit slower at sequential reading"
Seriously? Seems a massively biased review to state that and draw graphs with non zero Y axis so to a cursory glance it looks like a 25% performance drop when 15MB/sec out of 550 is 2.7%.
I know 72% faster random reads, 3% slower sequential reads, 12% faster random writes might not sound as snappy, but this is The Register not the daily mail or metro.
So stick to the facts, put in a few funny quips and jibes when it fits but don't exagerate beyond all proportions and for goodness sake never ever repeat dodgy graphs like that
Yeah, misleading graphs
But I think the fact they're just the manufacturer's specifications, rather than actual test results, is much more of an issue than the vertical scale not starting at zero. At least the values are shown on the axis.
Real hardware testing websites have covered Vertex 4 vs Vertex 3. For some workloads v4 is superior, for others it isn't.