Yesterday Western Digital and SanDisk announced their collaboration for hybrid hard drives (or SSHDs as they are now called). The idea behind the move is that SanDisk will supply Western Digital with iSSDs (I’ll explain what these are in a bit), which WD will then integrate with their hard drives to act as a read/write cache. The new 2.5" WD Black that was first showcased at IDF last year will be the first fruit of the collaboration, and it will also be WD’s first SSHD.

WD’s approach with SSHDs is slightly different from Seagate’s. Seagate simply put a NAND package on the PCB and used their own controller manage it. SanDisk’s iSSD, on the other hand, is a standalone SSD with a SATA 6Gbps interface in BGA-156 form factor (sometimes called µSSD). Both implementations obviously have their pros and cons: Seagate has total control over the NAND (garbage collection, etc.) but it also means the NAND performance is up to Seagate’s engineers, whereas WD can rely on SanDisk’s expertise on the NAND frontier and concentrate on caching and hard drive technologies.

Now, before we get too excited, putting an iSSD inside a hard drive won’t magically solve the biggest problems that SSHDs have. The first generation WD Black will only have 8GB-32GB of NAND, which is better (or the same) than Seagate’s SSHD (or Momentus XT as it was called previously) but it’s still not enough to provide performance that is even close to the SSD-only experience. SanDisk promises pretty impressive numbers for the iSSD (450/350MBps sequential read/write, 9K/1K IOPS random read/write), but those are for the largest 128GB SKU. Remember that one of the key elements of SSD performance is high parallelism: The more NAND you have, the better the performance is because you can access multiple dies simultaneously. As the Black SSHD will only have up to 32GB of NAND, it won’t be able to achieve a similar level of performance as the 4x larger 128GB model.

With increasing NAND densities the performance difference between capacities has become an even bigger issue because performance decreases with every process node, yet capacity per die goes up. This is a double-whammy that results in lower parallelism (and hence performance) at the smallest capacities. With SLC NAND you could somewhat dodge the performance issue since SLC NAND is faster to begin with, but unfortunately the iSSD is MLC based (which shouldn’t surprise anyone given the pricing of SLC NAND).

When you combine the very limited amount of NAND with a low-power integrated controller, it’s simply impossible to get performance that’s anywhere close to a decent standalone SSD. Of course there’s the caching side too because only a small portion of your data can be stored in the NAND, so in most cases you will still be limited by the spinning platters. With 32GB it should finally be possible to cache Windows in full, although the hurdle of hardware-level caching is that you have no say in what goes where.

What’s really special about this announcement is the timing as Seagate introduced their first consumer SSDs only a day earlier. I have a feeling that WD and SanDisk had not planned to go public with their partnership yet but Seagate’s announcement changed their plans. With WD’s biggest rival entering the SSD market, it’s clear that shareholders want to know WD’s strategy in order to maintain credit on the company.

Aside from keeping investors happy, there are also concrete reasons for the partnership. By far the most important one is the fact that SanDisk is a fab owner (a joint-operation with Toshiba where SanDisk gets 49% of the NAND output). Nowadays if you want to do something that requires NAND, there’s no other way to guarantee a steady NAND supply than to partner up with a NAND fabricator. There have already been several NAND shortages in the market (and it's only going to get tougher this year) and the brutal fact is that the ones without a fab or partnership are the last ones in the supply chain.

I’ve already heard from several fab-less SSD OEMs that they have not been able to keep up with demand because there’s not enough NAND in the market. For someone like WD a steady NAND supply is even more important because at least in the beginning the WD Black SSHD is aimed towards OEMs (there's a custom connector so it doesn’t work in regular systems without an adapter). If your production is dependent on the fluctuations of the NAND market, OEMs will likely not choose your product because they don’t want to take the risk of halting their own production due to the lack of drives. It’s not a coincidence that for example Apple sources their SSDs from Samsung, Toshiba, and SanDisk, which are all NAND fabricators.

If WD ever decides to re-enter the SSD market, the partnership will obviously be even more important. I wouldn’t be surprised if WD and SanDisk were actually working on an SSD together. SanDisk has controller IP thanks to the acquisition of Pliant in 2011, but that arm of SanDisk has mainly been focused on the enterprise segment. The consumer market has much lower profits and it’s usually not profitable to design a consumer-grade controller on your own, but with a partner like WD it can turn out to be a good investment as R&D costs can be shared and WD has an enormous distribution channel for providing the product to the market.

All in all, I have a feeling that the real fruits of this partnership won’t be seen today or tomorrow. The WD Black SSHD is definitely an interesting product and we will try to get one in for reviewing as soon as possible, but it's likely that you will still be better off with a small-ish SSD accompanied by a hard drive for storage. My gut is saying that this is more of a transitory product as WD gets ready to re-enter the SSD market. That doesn’t mean it’s the end of story for SSHDs, but this announcement should have happened two, preferably three, years ago.

Source: SanDisk Press Release

Comments Locked

30 Comments

View All Comments

  • Kristian Vättö - Friday, May 10, 2013 - link

    The difference is, SanDisk ReadyCache is a standard 2.5" SSD and not a BGA package. The iSSD has much stricter space and power limitations, which also have a significant impact on performance (9K/1K random read/write is not exactly impressive and the 32GB SKU will be even worse).

    Maybe I'm just a pessimist. I was very excited when the first MomentusXT came but I ended up being quite disappointed, which might have lowered my expectations when it comes to hybrid drives. The iSSD is something that could be hyped as "the first real SSD+HD combo!!!" but when you look at the facts, it's still not going to be a real SSD (even if all your data was cached). I hope I'll be positively surprised, though.
  • name99 - Friday, May 10, 2013 - link

    There were multiple problems with MomentusXT.
    One was the cache size was just too small. A second was that writes weren't cached.
    32GB is a lot closer to a reasonably useful size. IF that is coupled with write caching, we may have something that works well. The write-caching is very important, in my experience with home-constructed hybrid disks. It's not that you want writes to be fast, it's that writes force the head to bounce around between the data you are reading and flushing out various bits of random stuff.
    If you can cache the bulk of metadata in the SSD for reading, and likewise can have the myriad misc small writes happen to the SSD, then your head basically sits in one place tracking the "main" file that you care about.

    Personally I wouldn't worry much about the performance details of the SSD --- this is a device that's supposed to act like a very fast HD, not like a mid-range SSD.

    What I would worry about iso
    (a) the quality of the caching algo
  • name99 - Friday, May 10, 2013 - link

    (oops) hit the wrong key.

    What I would worry about is
    (a) the quality of the caching algorithms (most obviously whether write caching occurs, but also how block usage is tracked, how blocks are replaced, how much free space is kept available for new writes, that sort of thing(
    (b) the peak power draw when lots of back-to-back writes occur, especially for what should be bus-powered USB devices. (Will they only work on USB3 and intermittently hang on USB with its lower power budget?)
  • Kristian Vättö - Friday, May 10, 2013 - link

    Write caching is supported, though like you said, it also brings a bunch of other issues (which is one of the reasons why it was better to go with SanDisk iSSD than standalone NAND).
  • Wall Street - Friday, May 10, 2013 - link

    If it is anything like the ReadyCache which uses a Diskkeeper product for the Windows software, then the write cache will be in RAM which somewhat alleviates the performance issue of few NAND channels (writes suffer more from this than reads) and endurance.
  • stickmansam - Thursday, May 9, 2013 - link

    32GB will make it very interesting if they get it into the standard 7-9 inch form factor with the regular connector. Combine that with 7.2k speeds (which Seagate dropped) and it'll be a very interesting solution for laptops with only 1 bay. Running a 64GB SSD to cache (SRT) my 2TB 7,2k on my desktop speeds up my OS + few fav games + chrome quite a bit. 32gb should do great for say a 500gb-1tb HDD. I echo the sentiment that WD is unlikely to enter the SSD market while Sandisk is doing fine by themselves.
  • Kristian Vättö - Friday, May 10, 2013 - link

    From the technical angle, 32GB will definitely make it to the 7mm form factor at least. It's 0.2mm thicker than 8GB and 16GB packages so if 8GB can be fit into 5mm, there is no reason why 32GB can be fit into 7mm. The real question is, will we see 32GB in 5mm FF? That's crucial for Ultrabooks as Intel requires at least 20GB of NAND (though Intel may also require that the SSD is managed by Smart Response Technology, which isn't possible with an SSHD).
  • Romberry - Friday, May 10, 2013 - link

    I don't really think that the observations concerning that fact that integrating these kinds of SSD caches into mechanical hard drives won't make these hybrid drives as fast on an absolute level as pure SSD's is particularly relevant. Hybrid SSD/mechanical drives are about achieving massive storage at reasonable cost and bringing in a really significant performance improvement/advantage at an affordable price point. You can graft these SSD caches and controllers onto any size hard drive and not drive the price through the roof. Imagine terabyte+ drives with 32 or 64 gigs of SSD cache. Now imagine the potential improvement in write performance, and to a lesser degree (depending on file type, file size and prediction algorithms) read performance (up to the limits of the cache.) This is a middle ground strategy designed to give massive storage at a price we mere mortals can afford to pay, not a strategy to make mechanical drives into SSD's. I really disagree with the author's conclusions. This sort of strategy is exactly the answer to the current gap between high cost per gigabyte SSDs and lower performing by orders of magnitude but dirt cheap per gig purely mechanical hard drives.
  • risa2000 - Friday, May 10, 2013 - link

    Your observation is sound and someone who runs two 2TB WD Green drives (solely as media drives) I can concur that I do not care much about performance for the these two drives. So technically any speed-up might be welcome?
    Not really. Do I take hybrid drive with likely higher power consumption and one additional piece of puzzle in processing chain which may fail? What would have to be price incentive, or speed-up incentive (about which I do not care much) to justify that? I do not know and I am not sure.

    But if WD simly sees this as en evolution of standard HD = no drives without iSSD will be produced anymore, then there is no dispute, I will buy this, or maybe some other brand which will provide classic technology.
  • Kristian Vättö - Friday, May 10, 2013 - link

    I edited the last paragraph a bit to clarify what I meant (you are right that this is exactly what will tighten the gap between SSHDs and SSDs).

Log in

Don't have an account? Sign up now