i-RAM as a Paging Drive

One question that we've seen a lot is whether or not the i-RAM can be used to store your pagefile. Since the i-RAM behaves just like a regular hard drive, Windows has no problem using it to store your pagefile, so the "can you" part of that question is easily answered. The real question happens to be, "should you?"

We have heard arguments on both sides of the fence; some say that Windows inefficiently handles memory and inevitably pages to disk even when you have memory to spare, while others say that you'd be stupid to put your pagefile on an i-RAM rather than just add more memory to your system. So, which is it?

Unfortunately, this is the type of thing that's difficult to benchmark, but it is the type of thing that's pretty easy to explain if you just sit down and use the product. We set up a machine, very similar to how we would a personal system, but tended to focus on memory hogs - web pages with lots of Flash, Photoshop, etc. Of course, we opened them all up at once, switched between the applications, used them independently, simultaneously, basically whatever we could do to stress the system as it normally would be stressed.

At the same time, we monitored a number of things going on - mainly the size of the pagefile, the amount of system memory used, the frequency of disk accesses, pagefile usage per process... basically everything we could get our hands on through perfmon to give us an idea if Windows was swapping to disk or not.

The end result? There was no real tangible performance difference between putting more memory in the system and using the hard disk for the pagefile or putting less memory in the system and using the i-RAM for the pagefile. Granted, if we had a way of measuring the overall performance, it would have shown that we would be much better off with more memory in the system (it runs faster, and it is accessed much quicker than off the i-RAM).

The only benefit that we found to using the i-RAM to store our pagefile was if you happened to have a couple GBs of older DDR200 memory lying around; that memory would be useless as your main system memory in a modern machine, but it'd make a lot better of a pagefile than a mechanical hard disk.

One more situation we encountered that would benefit from storing your pagefile on the i-RAM was those seemingly random times when Windows swaps to disk for no reason. But for the most part, our system was slower when we had less memory and stored the swapfile in it than when we had more memory and less swap file.

Adobe Photoshop is a slightly different creature as it keeps a scratch disk that is separate from the Windows pagefile. We tested Photoshop and used the i-RAM as our scratch disk, but in all cases it always made more sense to just throw more memory at Photoshop to improve performance where we ran out of memory. If the operations you're performing in Photoshop can fit into system memory, then you'll never touch the scratch disk.

Overall, based on our testing, the i-RAM doesn't make much sense as a paging drive unless you have the spare memory. The problem with "spare" DDR200 memory is that it is most likely in small 64MB, 128MB or maybe 256MB sizes, which doesn't buy you much space on an i-RAM drive. For most people, you're much better off just tossing more memory in your system.

i-RAM Pure I/O Performance i-RAM as a boot drive
Comments Locked

133 Comments

View All Comments

  • crazySOB297 - Monday, July 25, 2005 - link

    I'm surprised they didn't raid a few of them... I think you could get some huge performance.
  • Googer - Tuesday, July 26, 2005 - link

    quote:

    I'm surprised they didn't raid a few of them... I think you could get some huge performance.


    Not to mention it is a way to also get around the 4gb siza limitation.
  • Hacp - Monday, July 25, 2005 - link

    Dude the article said straight out that SATA150 was the only format supported. Read the entire article.
  • Guspaz - Monday, July 25, 2005 - link

    I too am dissapointed that the article lacked any mention of SATA2, which is twice as fast as SATA (300MB/s vs 150MB/s). Considering many motherboards already on the market suport SATA2, and the 300MB/s transfer rate that goes with it, it is a bit of an oversight that the articles doesn't even MENTION if the card supports SATA2 or not. Nor do they mention what they think would happen with SATA2, or if Gigabyte is likely to produce a SATA2 version. It's a weak spot in this article, I think, considering how central the bandwidth of SATA is to the performance of the i-RAM.
  • snorbert - Tuesday, July 26, 2005 - link

    quote:

    I too am dissapointed that the article lacked any mention of SATA2, which is twice as fast as SATA (300MB/s vs 150MB/s)


    33MHz PCI only gets you 133 MB/sec theoretical, and more like 110 MB/sec in the real world. The i-RAM with SATA 1 can completely saturate a PCI bus. SATA2 would cost more to implement, and give you no speed increase at all on a 33MHz bus. If you build the card for higher-end PCI specs (e.g. 66MHz, 64 bit, 66MHz/64bit, PCI-X) then you automatically exclude most PC enthusiasts (unless they like buying server boards for their game boxes).

    If they end up doing a PCI Express version, then there would be some reason to support SATA2.

    This board is not a replacement for a hard drive. It would be incredibly useful as a transaction log though. Reliable (i.e. won't get lost if the machine crashes) write-behind caching for RAID 5 drives will give you a huge boost to write speeds. And the controller cards that support battery-backed write behind caching cost a lot more money than an i-RAM.

    -Jason
  • sprockkets - Tuesday, July 26, 2005 - link

    Also to reply here

    Keep in mind that for many years the ide/sata controllers are NOT on the PCI bus of the southbridge, so PCI is not a limitation.
  • snorbert - Tuesday, July 26, 2005 - link

    Actually, scratch my comment - I had not had enough coffee when I wrote it. I forgot that the PCI connector is doing essentially squat except providing power to this device. Of course you could have a SATA2 controller on a faster bus talking to this thing. But an SATA2 version would probably cost more. (because it would need a faster FPGA, newer SATA transceivers)

    Sorry folks,
    Jason the doofus
  • Anton74 - Monday, July 25, 2005 - link

    You did miss that reference; on page 2 it says "The i-RAM currently implements the SATA150 spec, giving it a maximum transfer rate of 150MB/s".

    Given the 1.6GB/s of the RAM, it seems completely silly not to provide a 300MB/s SATA interface instead, especially considering that the whole contraption including RAM will cost as much as 2 or more decent hard drives.

    Anton
  • ryanv12 - Monday, July 25, 2005 - link

    The controller on the card is not SATA-II...it can do a max of 1.6GB/s...not exactly SATA-II speeds there...
  • Anton74 - Monday, July 25, 2005 - link

    1.6GB/s is actually more than 5 times 300MB/s, the maximum supported by SATA-II. So 300MB/s could easily be fully utilized, and I don't understand why they didn't support that.

    Anton

Log in

Don't have an account? Sign up now