5

I watched a few videos trying to answer this.

This video (youtube.com/watch?v=eULFf6F5Ri8) shows a bunch of guys stacking 24 SSD's reaching a peak of around 2GBps r/w. That's under the limit of the worst DDR3 in this list (memorybenchmark.net/write_ddr3_amd.html) - that shows DDR3 memory performance varying from 2.78 to 6.55 Gb per second, but that video is over 3 years old.

This video (youtube.com/watch?v=27GmBzQWwP0) shows a more optimistic situation, but for PCI-E SSD drives:
5 drives peaking at around 4Gb.

And this other video shows that stacking up more than 3 SSD's doesn't realistically offer a substantial added performance.

This and the fact that in all benchmarks the drives act quite poorly when dealing with small files (5k file read/write averaging from 10MB to around 30-40MBps) as opposed to how native memory handles such files, seems to indicate a definite NO to this question.

Also, the write life cycle is indeed limited and the drives might wear out quickly, as kindly pointed out by paddy.

However, I wanted to get more opinions on this.

Would it be possible to at least obtain current memory performance with SSD's in RAID 0? And if so, in what circumstances?

I am assuming using this configuration with a Windows OS that has a memory pagefile resident to that stack of SSD's, thus making it very fast to work with.

8
  • This chart places DDR3 memory performance between 2.78 and 6.55 Mb per second. link May 30, 2014 at 9:04
  • Dave, it is already in the question, but I had to place it as text because I was limited to 2 links only. But true, I might have chosen to make that link more visible instead of the others. :) May 30, 2014 at 9:07
  • 2
    Given enough added hardware and enough patience you could substitute punched paper tape for RAM. May 30, 2014 at 17:34
  • 1
    The operative word for RAM is random. If programs always asked for data and instructions in sequential order it would make a lot of engineers much happier (but also put a bunch out of work). May 31, 2014 at 0:51
  • 1
    In fact, all you really need to do is to reproduce the design of the IBM 650. Just be sure to also create a new version of the SOAP assembler. May 31, 2014 at 0:54

2 Answers 2

3

24 SSD's reaching a peak of around 2GBps r/w.

"Peak" is the operative word here.

DDR3 memory performance varying from 2.78 to 6.55 Gb per secon

Whatever speed DDR3 is running at, it will run at that speed all the time on every access. You get the peak rate from an SSD only under an ideal condition (data is in cache, data is aligned properly, all flash lines not in use, etc.)

If you had a drive that was essentially battery backed RAM such as this you could achieve something near to this performance possibly (though the interface will be a limiting factor and never as fast as direct-to-CPU connected RAM), though it would be costly, power hungry, probably less reliable, and likely not fit in your PC if you want more than 32GB or so.

5

I'm sure given enough SSDs you could match the throughput, but for overall system performance, any sized RAID 0 array of SSDs cannot replace RAM. A quick look at Wikipedia shows that SSDs have seek latencies around 0.08-0.16ms, but SDRAM has latencies around 10ns which is 10000 times faster. Since typical memory operations don't involve copying huge gigabytes of RAM from one place to another, bandwidth will not be the determining factor here, but latency is critical.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.