Top of the Intel SSD crop from 160 GB consumer SSD to Parallel 4-disk Intel Enterprise SSD RAID

IF YOU'RE HOLDING OUT to buy a solid state drive, your wait may be over.

SSD drives have not only become substantially cheaper, but have also improved quite a bit by now. Both read and, importantly, write speeds have been upped and made more consistent, and write reliability is seemingly far better than before. Wear levelling algorithms used by sophisticated multichannel NAND controllers on the drives, aided by ever increasing local DRAM buffers for large writes now approaching 128MB, have helped tremendously to improve write speeds, despite the still present long term write cell damage problem for NAND.

Add to this the lower power consumption, lightweight noiseless operation, and no mechanical wear and tear, and it's a way to go, as long as prices improve as well.

Intel has benefitted from SSD popularity as the means to enter the storage market in a big way via a pure semiconductor based solution, something impossible during the hard disk storage days. So has Micron, Intel's partner in creating the Open NAND Flash Interface (ONFI) standard. That allows Flash memory DIMMs like those "Braidwood" ones seen on, then suddenly removed from, the P55 Lynnfield mainboards a few months ago.

Here we look at the brand new 160GB Intel X25-M SSD model based on the most recent 34nm process NAND Flash chips, running on the Asus P6T7 WS mainboard on the Core i7 975XE processor with 6GB memory. The 7mm thin 2.5 inch drive, weighing less than 80 grams, could even fit into many slim notebooks, not to mention desktops.

The performance claimed by Intel is still 250 MB/s read and 70 MB/s write, same as the old 50nm based SSDs, but the latency and write IOPS have improved by over a quarter.

I ran the disk using Sandra 2009 SP4 tests on both Windows 7 in the IDE emulation mode, without command queuing, and the same Windows 7 in native AHCI SATA mode, with NCQ enabled. The Read, Write and detailed IOPS tests across multiple sizes were run.

Why again Sandra? I do know that some of our readers persistently comment against the benchmark, however its storage benchmark portion has comprehensive read and write test coverage as you see here:

Windows 7 IDE

Read: 241MB/s

Write: 99MB/s

Detailed iops spread:

Windows 7 AHCI NCQ

Read: 255MB/s

Detailed iops spread:

As you can see, there is a benefit - but not that great - of using AHCI, for large files, but for some reason the non command queueing IDE mode was still a little faster especially for small transfers. Either way this is great performance for an MLC cell-based SSD. It showed blazing fast reads maxing out the SATA 300 connection, and very decent write speed, above Intel's own claimed performance spec.

This 34nm refresh of the Intel X25-M drive may not have the highest absolute performance among the mainstream SSDs - the Patriot Torqx we tested two months ago is faster in some benchmarks - but it's near the top in every category. And anyway, Intel does have faster SLC-based Extreme enterprise drives available as well.

Talking about SLC drives, let's move to some results with them. Since SSD drives have great read speeds and, now, improved write performance too, what happens when you put more of them in parallel in, say, RAID0 or RAID10? Well, for one, if you don't have a dedicated hardware RAID controller sitting on its own PCIe lanes, the typical South Bridge will get overwhelmed.

Now, to pack them together nicely and fit them with the minimum hassle and only half the power connectors used, I got two A-DATA XPG hot plug 2.5-inch storage enclosures, each packing two 2.5-inch drives inside a single 3.5-inch enclosure, each with its own SATA link but two sharing a power connector. The sleek looking black contraptions are well suited for desktop PC mounting, where usually the 3.5 inch to 2.5 inch adapters would be too cumbersome to install. So, all 4 drives fit nicely into two enclosures.

The quadruplet of 32GB Intel X25-E Enterprise SSD drives, each based on SLC fast cells and providing 250MB/s reads and 170MB/s writes, can run in parallel as a huge RAID0 stripe or RAID10 (mirrored stripe) array. It easily overwhelmed the X48 and X58 chipset's south bridges before.

Here, on the X58 chipset and ICH10 south bridge, what will happen? Will a 60GB RAID10 configuration of four of these X25-E drives overwhelm the 4 PCIe lanes that link the ICH10 to the North Bridge?

The RAID10 setup was tested afresh in both Read and Write tests, as well as the detailed iops:

Read: 576MB/s

Write: 294MB/s

Detailed iops:

Wow, a RAID10 mirrored stripe, where you can still have a full parallel read from all four drives, basically maxes out the DMI link from the south bridge. Reaching nearly 600 MB/s in reads is nothing to scoff at, now I got a very fast instant-boot OS disk there, even though data and pagefile - and anything else write-intensive - would go to a secondary hard disk instead. Also, you may notice that, in IOPS, the advantage over single drive is far less pronounced in small size transfer, but nearly double in larger transfers.

Even the writes reach, where only two-drive parallelism is achievable due to mirroring, was nearly 300MB/s. But anyway, with SSD, the idea is to minimise the writes by offloading the write-intensive files and structures to another drive as mentioned above. The SLC based drives like the X25-E should have a longer write life anyway. µ

Good: Intel continues with great SSD drives, both desktop and server, simply stunning performance without much hassle.

Bad: Interface and system limitations start to show, time to re-engineer the South Bridge links?

Ugly: Until phase-change memory or something else comes in instead, every single bit write you do slowly but surely destroys your NAND Flash-based SSD.

Beers: 8/10