iRAM Update

First, the real news, which is no news. We’ve haven’t heard anything from Gigabyte yet on getting an iRAM to test.

Second, someone recent brought to my attention a review of the iRAM that was done before Computex.

You likely didn’t hear about it because it was in Chinese, but Joe posted a link to it the other day, and Stephen Lin was nice enough to send me a translation of that article, so I could see what else they had to say about it.

There’s two particular statistics of significance in the article. The first measures boot time from first sight of the Windows logo to the Windows musical intro (not too sure why they measured it that way, but we’ll take anything we can get:

Seagate 7200rpm 8Mb cache hard drive: 13.17 seconds
iRAM: 6.42 seconds

Presuming the rest of the boot process shows equivalent time savings, that would mean that you’ll boot with an iRAM in about half the time you’ll boot today. That’s pretty good.

On the other hand, if you thought an iRAM would let you do an entire boot in two seconds or less, that’s pretty bad.

The problem isn’t with the iRAM, but your expectations from RAM drives, any RAM drives, either now or for the foreseeable future.

The truth is, it simply takes a (relatively) long time to boot, for a CPU to do all the work an OS wants it to do. The time it takes for an iRAM to actually transfer the amount of data a CPU needs to open up Windows XP is only about a second. Nonetheless, it takes much longer than that for the computer to do the work necessary to get everything operational.

I think we’ll find the same will be true in loading game levels. If it takes thirty seconds for you to load a level now, an iRAM will probably get that down to around fifteen seconds. It won’t get it down to one or two.

Nor is this a matter of the SATA interface being a bottleneck. If the computer can’t do all the work necessary to process information coming in at 130Mb a second in real time, it’s not going to do a lot better when it comes in faster. The CPU and the OS become the bottlenecks.

Can something be done with these bottlenecks? Sure, eventually, but that will be done to some degree by the CPU/mobo people, to probably more degree by the OS people.

The other area of note is in data transfer. On average, the iRAM is about three times as fast as a fairly mundane 7200rpm drive, maybe it would be a little less than 2 1/2 times as fast as a Raptor.

This is a very big jump up in performance, but miraculously instantaneous it’s not. Again, keep in mind that if you want your CPU to do something to the data flow, it’s going to take time for it to do its thing.

This is an instance where the SATA interface is more of a bottleneck, limiting throughput to somewhat less than the 150MB/s SATA limit, but again, once you have to do something with the data, that becomes the bottleneck. A much smaller bottleneck, but still a bottleneck.

If The Wrong People Get Their Hands On This . . .

It seems to me the last few years, expectations have grown, while results have shrunk. People are expecting more radical improvements than before, and if they don’t get it from something, they tend to trash it.

This is my big fear about iRAM. When the comprehensive reviews come out, it will be found that it will be very good at cutting down the amount of time it takes to do certain things, but on the whole, it won’t greatly increase performance because, like memory, increased speeds in one area has relatively little overall impact on total performance compared to, say, ramping up CPU speed.

I’m afraid a lot of people have impossible expectations for this device, and when the iRAM numbers from the reviews don’t meet it, they’ll just conclude, “This sucks.”

Even worse will be those people with impossible expectations who spend a lot of money better spent elsewhere on the thing, then get disappointed and start really screaming, “This sucks!”

Let me put it this way. If you have a two-three year old system (or one that could have been bought back then) without a very specific (most likely work-related) reason for wanting this, you have no business buying this.

My own experience with RAMDrives tells me that the benefits in real-time use are mostly going to be a split-second here, a split-second there. If you don’t think that’s worth paying for the card to get that for a few applications (if you’re recycling a little RAM) or even for quite a few apps and Windows (if you buy the card and a lot of RAM), you’re probably right. If cost is at all a serious factor in your buying decisions, and you have to prioritize, this is a pretty low priority.

But that doesn’t mean it sucks; it just means you have bigger fish to fry, and it’s not ready for you yet, which is not surprising given that having these for regular desktop operations is in its infancy.

The iRAM is one of the first steps beginning the move of solid-state drives (in whatever form) towards the mainstream. It lets the pioneers in; it doesn’t make this mainstream.

I still want one (actually, I want two, fully loaded with 4Gb each), but I know I’m gilding the lily. This is my version of wanting two 7800GTXs in SLI; it might be good for me, but it’s bad general advice.

Ed