Never mind the money, feel the width
Some time yet when SSD capacity sales overtakes HDD capacity. I'm guessing around 3 years.
Gartner gnomes think revenues from enterprise sales of SSDs will exceed those from HDDs in 2017. Well, let's look into that claim. We all know SSDs are taking over from fast disk drives as the main storage medium for primary data, and now Gartner’s esteemed number-crunchers have estimated when the two revenue lines will cross …
The SSD market is, despite the best efforts of drive makers, NOT slavishly following the "enterprise" classification beloved of HDD makers. Enterprises are moving to appliance level redundancy and that implies cheaper SSDs are good enough.The figures from Gartner don't include this large SSD segment and so understate the impact of SSD on "enterprise" HDD sales by a considerable factor.
What is surprising is that there are still good sales levels for those very expensive "enterprise" drives. Prices are 2 to 3x those of comparably sized mid-range SSD, which makes any rationale for staying with these HDD unfathomable!
"SSD MLC, SLC 2D or 3D is a waste of money and unreliable"
Nonsense. If you have a problem with latency and I/O performance now, you can't just wait until something else comes along as the "perfect" solution. If your company's systems have hit the throughput buffers, if you are wasting money on expensive call centre staff hampered by slow responses or driven away customers due to dreadful web application performance, you need it fixed now.
As for SSD being unreliably, they are the fundamentally much better than HDDs. Of course both HDDs and SSDs need some form of redundancy to achieve the sorts of availability that major enterprises require, but there are well proven ways of achieving it. As for write endurance issues, with proper controllers (and the right type of SSDs for the right workload) this really isn't an issue and normal maintenance contracts will deal with replacements of failures.
That's even before you look at the reduced power. In short, for the right job, SSD is an absolute bargain whatever the future might bring.
'Given the gap in per GB cost between hard drives and SSDs, and the fact that the first to die will be 10K/15K SAS drives in favor of bulk capacity SATA drives, >90% of the total bytes sold will still be hard drives when the revenue crossover occurs.'
This. ^^
SSDs will become the norm for the very high performance apps. Until SSD is cheaper than NL-SAS 7.2k spindles, the rest will migrate to hybrid where you can accelerate NL-SAS to near SSD performance at a fraction of the cost and a much reduced footprint. I see a swift demise of 10/15k drives as their performance advantages are now a thing of the past.
Not really, every process shrink adds 2.25x more capacity per wafer. They don't need to build fabs to make more flash.
The memory business has always been boom and bust. A few lean years cause a few players to go out of business and older fabs to not get upgraded. Then prices stay flat for a few years (i.e. you make 2.25x more revenue per wafer) and the remaining players add new capacity to capitalize, a glut results, and the cycle begins anew.
We're currently still kinda sorta in a boom cycle, but as the industry has consolidated so much the booms and the busts are reduced in magnitude. Maybe they are getting smarter, but more likely they are secretly colluding. Which is fine, as with OPEC there will be enough cheating to keep the market well supplied and avoiding a bust will allow a longer if more moderate boom to continue longer.
A process shrink hasn't given >2X bits/wafer in a number of years. Nearer +50% now, although you could argue with the first couple of generation of Samsung 3D, there's been a drop in bits/wafer.
I believe overall NAND bits produced per year had increased buy 30-40% per year, primarily from
bits/wafer rather than wafer throughput increase. 3D should eventually enable a continuation of this bit growth (for a few years at least) through bits/wafer increase but the $/wafer will likely increase (due to equipment capex, additional fab space & process steps needed for 3D) so the overall price trajectories will flatten. (Not good for Hynix/Micron who have already been losing money on NAND for years).
There needs to be some consolidation & NAND price increases to make it profitable as the current situation is not sustainable. This should push out the SSD-HDD crossover.
It's interesting to see volumes, not revenue.
As Doug S also mentions - the more expensive 10K/15K RPM SAS drives are already biting the dust. Then at current $/GB prices enterprise SSDs are 10x more expensive than high-performance SATA HDDs. We already see even in hybrid SSD systems cost (i.e. vendor revenue) of SSDs is 3x > of cost HDD, let alone all-flash.