Sign in with
Sign up | Sign in
Your question
Solved

Need a Small Fast SSD for SSD Caching

Last response: in Storage
Share
June 5, 2013 11:23:12 PM

For some reason I cannot find a small SSD (64GB or less) with high read speeds. I want to use one for a SSD cache for my spindle drive and all of the small drives seem to have slow read speeds. I am used to my Samsung 840 120GB that stays right around 500 GB/s. I purchased this recently:

http://www.newegg.com/Product/Product.aspx?Item=N82E168...

Not impressed. Sequential read speeds are just under 200 MB/s...barely faster than my spindle drive!! The write speeds happen to be SLOWER than the spindle drive! It seems SSD makers advertise speeds of "up to" some number. Samsung advertises and benchmarks for sustained speeds. I would by another Samsung 840, but I don't need a 120GB cache...just a bit excessive I think...

Anyone have any suggestions?

Best solution

a b G Storage
June 6, 2013 3:15:01 AM
Share

Most of the speed-up from an SSD comes from the random 4k read/writes, not the sequential read/writes. A top-tier SSD is only about 4x faster than a HDD at sequential read/writes, but it's typically 30-100x faster at random 4k read/writes. These small read/writes happen almost instantly on an SSD, but a HDD has to physically move the read/write head and wait for the platter to spin the data into place (seek time).

So you really should be looking more at the IOPS rating (I/O operations per second), instead of the sequential speeds. The Mushkin drive you linked is rated at 90,000 IOPS, which is damn fast for a 60 GB SSD (hard drives typically only manage a few hundred IOPS).

Real-world performance will be a combination of IOPS and sequential read/write speeds. But unless you're in the habit of copying large media files over and over, the IOPS will matter more. For a more eclectic benchmark, you may want to try something like CrystalDiskMark, which tests sequential, 512k, 4k, and 4k with queuing read/write speeds.
Related resources
June 6, 2013 6:40:57 PM

Ahh, I see. Looks like I was judging by the wrong speeds. However, I do video editing and frequently access large source files for rendering. Would sequential read speeds be more important in this case? My goal is to reduce rendering times slightly and to improve responsiveness and real time previews prior to rendering.

Solandri said:
Most of the speed-up from an SSD comes from the random 4k read/writes, not the sequential read/writes. A top-tier SSD is only about 4x faster than a HDD at sequential read/writes, but it's typically 30-100x faster at random 4k read/writes. These small read/writes happen almost instantly on an SSD, but a HDD has to physically move the read/write head and wait for the platter to spin the data into place (seek time).

So you really should be looking more at the IOPS rating (I/O operations per second), instead of the sequential speeds. The Mushkin drive you linked is rated at 90,000 IOPS, which is damn fast for a 60 GB SSD (hard drives typically only manage a few hundred IOPS).

Real-world performance will be a combination of IOPS and sequential read/write speeds. But unless you're in the habit of copying large media files over and over, the IOPS will matter more. For a more eclectic benchmark, you may want to try something like CrystalDiskMark, which tests sequential, 512k, 4k, and 4k with queuing read/write speeds.


a b G Storage
June 12, 2013 8:52:37 PM

Tien23 said:
Ahh, I see. Looks like I was judging by the wrong speeds. However, I do video editing and frequently access large source files for rendering. Would sequential read speeds be more important in this case? My goal is to reduce rendering times slightly and to improve responsiveness and real time previews prior to rendering.

Sequential read speeds would be more important in that case. But I dunno how much more. A SSD can read a video file far, far faster than a CPU rendering video could ever process the frames. Something like a simple mux (combining two files) would go faster on the faster SSD. But if you're rendering I suspect the task is CPU-bound, not drive-bound.

Since you have both the 200 MB/s Mushkin and 500 MB/s Samsung, why not put the same video files on both drives. Then try your editing with one, then the other, and see if there's a substantial difference in speed. My hunch is the render itself is CPU-bound and SSD speed won't really matter. But the faster SSD may respond more quickly while you're doing the pre-render editing.
!