Romex FancyCache Review – SSD Performance At 13GB/s and 765,000 IOPS In 60 Seconds Flat!


Now that the upper performance limit has been reached, perhaps a more real-world relevant test would show what this level of performance can achieve.  A PCMark Vantage run should do nicely.

The HDD component of Vantage puts in one of the highest scores around at 622865  points.  Without caching, the distressed Mushkin Chronos Deluxe could only manage a score of 52877. Each scoring component is moving at least 1700MB/s.  These are fairy tale numbers, and just the beginning of what could be possible with such an arrangement on a faster system.  The midrange components used here are fast, but still outclassed by much faster systems.  Intel’s X79 motherboard and Sandy Bridge – E processors could very well crush these mainstream components.


All things considered, the IO performance and huge throughput that RAM provides when acting as cache for a SSD is a wonder to behold.  Further, getting more performance when overclocked is just more icing on a hellaciously-fast cake.  Low and high QD randoms, sequential, access times, and IOPS are all marvelous when operating out of the cache.

But there are some caveats.  First, since RAM is volatile, the cache is not persistent.  When the system turns off, the cache disappears.  More importantly, the risk of data loss is omnipresent — if FC is caching writes to RAM, any fault with the RAM or sudden system crash could result in lost or corrupted data.  Using ECC RAM where possible could help in this respect, but it is seldom seen in most enthusiast systems.

The upside is obvious, though.  In specific applications, the performance increase could pay serious dividends.  If an application is heavily dependent on storage performance, RAM caching can yield incredible performance increases over SSDs alone, much less traditional spinning media.  FancyCache is easy to set up, and much more convenient that traditional RAM drives.

Caveats aside, the worst case scenario when using FC with a SSD is that performance will only be as fast as the SSD.  In the best case scenario, almost all accesses occur directly from the RAM, and performance skyrockets.  It’s either fast or faster.

Romex Software is currently beta testing FancyCache.  As such, it’s not yet for sale but freely available to test.  There are many facets and possibilities with FC, and the surface has only yet been scratched.

Just for kicks, one more test:


One last run of 4K random reads at a QD of 4 yielded the highest numbers yet — 765137.3 IOPS. Not bad for a few minutes of tweaking and overclocking. Still, the little-2500K-that-could was maxed out, and it’s bigger brothers could probably reach nearly one-million IOPS. The 2500K is a fine piece of kit, but in these situations a bigger hammer is called for.

One day, all storage media could be this fast. It sure won’t be anytime soon, but perhaps in the future, as new technologies supplant the NAND flash used in today’s SSDs. For today, if you can utilize the raw, unadulterated speed, there are brilliant performance increases to be had.  But for most uses, the bottleneck just gets shifted to other components, and such speed increases could be hard to realize.

It is also perilous to make blanket statements concerning RAM caching.  In a worse case scenario, it may not help much.  But under certain conditions, there is no substitute.  It is similar to a RAM drive in some respects, but FancyCache exhibits a level of flexibility unmatched by RAM drives.

FancyCache, freshly off a new beta revision, is certainly worth a try. Romex is aiming for an eventual commercial release, though some bugs are still being ironed out. With the potential pitfalls in mind, the potential benefits are certainly worthy of investigation. It’s most definitely a topic that requires more insight, but it is unquestionably interesting.

User Rating: Be the first one !


  1. blank

    This article talks about making a hybrid drive as an option. FancyCache requires an “L1” (Memory) amount minimum of 128 MB, there are no “L2” only caches (which would be an SSD). Also, the SSD cache is not persistent either – there is no mechanism in place to recover the data from a power failure or blue screen. Hopefully this hybrid drive option will be added later.

    Additionally, there are not only data loss issues but data corruption issues when using block based lazy writes. FancyCache’s main competition has had many issues of drives slowly becoming more and more corrupt over time. FancyCache calls out specific scenarios when you should use their product – in general if the windows read caching solution is insufficient for your program. The write caching doesn’t even come into play – only if you have self error checking programs and data (or are dealing with a scenario where data corruption only adds a bit of static) is it an acceptable risk.

  2. blank

    Please compare with other similar solutions eg. SuperCache by Superspeed.

  3. blank

    Wow, just wow! I have a 3930K that can hit 5.1Ghz, though it takes 1.5v so I typically keep my baby around 4.6Ghz (1.38v, under water btw), and is using 16GB of DDR3-2133 9-11-10-28 (G.Skill Ripjaws Z) running at 2400 10-11-11-30, with primary storage being a Samsung 830 256GB SSD….

    I am going to try this out this week, and see if I can’t break 1mil IOPS in both read and write! 50% more cores, 3x as many threads, AND a higher clock speed as well as 2x as many memory channels, faster memory…. I am pretty pumped to see what this can do!

  4. blank

    Shouldn’t the operating system do this automatically? I know Linux does, and I’d say Windows should have the same feature.

    • blank

      The main difference (besides the difference in the cache algorithm) is the ability to delay writes to the hard disk for seconds or even longer. I set my write back to 2 minutes on the drive I use to develop software (since all code gets checked into svn there is limited danger). The default cache in windows does not allow you to do that because it is potentially dangerous.

Leave a Reply

Your email address will not be published. Required fields are marked *