Gigabyte i-RAM

Benchmarks
 
Naturally, you’re here for the juice.

Our testbench machine was an Athlon 64 3500+ with 1GB of DDR on a Gigabyte GAK8NXP- SLI nForce 4 SLI-based motherboard.

Considering at Computex the de-facto demonstration of the i-RAM’s capabilities was its use as a Windows boot drive, we started our tests by installing a slipstreamed Windows XP and SP2 onto one of the i-RAMs.

The motherboard’s BIOS had no trouble recognising the i-RAM as a SATA drive (showing up as ‘Gigabyte i-RAM v0.96’), and so did the Windows installer, happily treating it as just another hard drive.

And the results? The install itself was marginally faster, but the real mind-fsck experience was seeing it boot up for the first time... in just four seconds, from post BIOS initilisation to desktop. And that’s just one i-RAM, we hadn’t RAIDed the pair yet. Various hoots of awe and loudly phrased self-deprecating expletives could be heard escaping the confines of Atomic Labs at this result. We had just witnessed, after all, the fastest Windows boot in Windowsdom.

Ironically, and as is the case with Windows, after installing approrpriate drivers for the system (NVIDIA Forceware set, graphics drivers etc) the bootup time extended to eight seconds as Windows got cosy with its hardware.

Next up we blanked the drive, booted off the hard disk, and setup one i-RAM as a data drive. For measuring throughput we used HDTach and SiSoft Sandra.

Sandra registered a maximum throughput speed of 136MB/s and a sequential read speed of 137MB/s. The best part – the random read speed also maintained a speed of 137MB/s. To understand the importance of this score we’ll take a look at the HDTach.

HDTach also showed a speedy 126MB/s for average read speed, but the two most important results to notice a seek time of 0ms (while obviously not zero, it’s so low it’s off the scale for HDTach) and the sequential read speed graph – it’s compeletely flat. A hard drive, by comparison, shows a consistently declining read speed as the heads move to the inner platters where the effective rotation is slower. With the i-RAM, of course, the whole data space is perfect. This means regardless of wether the data to be read is sequential or randomly scattered, the i-RAM delivers a consist throughput of 126MB/s.

Then, natrually, we setup RAID.

The i-RAMs were configured as dynamic disks and an 8GB RAID-0 software RAID array was made. Testing again with SiSoft Sandra and ATTO to focus on throughput, we witnessed the insanely fast result of some 270MB/s in Sandra, reflect in ATTO with a maximum throughput of 267MB/s. ASCII is too feeble a communication medium to express the type of smiley we’d like to print.

However all this being as it may it’s important to note that the i-RAM is severly bottlenecked by the SATA interface. Protocol overheads aside, the RAM is capable of many gigs a second transfer rates but the SATA interface can only push 150Mb/s at best. Even when the next generation i-RAM comes and supports SATA II at 300Mb/s, it’s barely touching what the RAM drive is capable of pushing.

It’s also worth noting that the i-RAM is still in development, and these preview units may not resemble the final product. For a start, these units are fussy over the compatible memory modules it can use, and it’s also rather fussy over the chipsets it likes to run on – Intel based boards are favoured at the moment (you need to be running at least the nForce 4 for AMD 6.67 drivers for the i-RAM to even show up on AMD boards) and it seems to prefer operating on SATA controllers in regular mode over RAID mode.

According to Gigabyte, all of these issues are being looked at and will be rectified as the product develops.