IDF SPRING 2007
While the SATA 3.0 update is expected to be set in cement in about two weeks (in case you missed it), Intel reminded us that the Serial ATA 2.6 update was published last month. SATA 2.6's white paper has an outline for a whole new slimline connector for 1.8” hard drives.
This is designed to reduce the space used in a notebook or UMPC as it allows the drives to mount lower power and fewer pinned SATA plugs on the drives' shorter edge. Currently the SATA connector is a bit too big to fit on 1.8” disk drive and when the original specification was put through, they didn’t expect SATA to be adopted into these components.
NCQ, as we previously discussed in the introduction to hard drive technology article, is a questionably useful inclusion into consumer hard drive technology. Since a consumer level hard drive never fills the request queue the performance is marred by the large NCQ overhead.
In SATA 2.6, NCQ has an update to include the NCQ Unload
feature. The NCQ priority settings have been updated for improved responsiveness and a reduced overhead to increase performance.
A quick question to Knut Grimsrud, Director of Storage Architecture, SATA-IO chairman, ONFI Chairman and Intel fellow (his business card is twice the size of mine) yielded some more detailed information. The additional priority settings allows selecting commands that are not necessarily on the drive in the correct order, but what is actually needed
first, even if it sits in the furthest track away. However, he did say this was specific to when the queue reaches a certain depth, meaning commands issued first could be then serviced far
later on, which is still not a consumer level use.
Despite suggesting that selective NCQ would possibly increase performance better for consumer machines, whereby only when the queue reaches a certain depth it enables NCQ, he dismissed the idea. This man really
likes his NCQ, and considering his position he must know what he’s talking about.
Also, there has been a recent kerfuffle about the use of both hybrid hard drives and Intel Turbo Memory technology with Windows Vista, citing the use of ReadyBoost being incapable of use with two separate NAND storage options. Mr. Grimsrud was quick to point out that Intel Turbo Memory can be used as both ReadyBoost and ReadyDrive, whereas hard drives are ReadyDrive only, so both can
be used together after all.
Previously there was no need to have a SATA certified program as there was only a few manufacturers to start with, and they all worked closely with the SATA-IO group. With the proliferation of new products and companies entering the market to supply SATA orientated goods, there was a need for a certification program to allow the end user to know that what they're buying will actually work. This is no doubt a good idea, and should hopefully stop some of the branding issues of old.
ONFI: Open NAND Flash Interface
Flash is the only high commodity memory that doesn’t have a unified standard. With everyone’s flash devices slightly different, making a single controller that supports all of them was incredibly difficult. The ONFI created the standard 1.0 interface which was finished and published a mere eight months after conception. It creates a uniform interface definition, timing, electrical properties and pin-out to talk to flash.
It has made it possible for the devise to tell the host what kind of Flash it uses, and the features it contains. By the time the Intel Turbo Memory (Robson) platform ships and becomes established with Santa Rosa laptops and future PC motherboards, new flash modules hit the market at such a high turnover rate that the interface might not be able to read them before it’s updated.
ONFI has protected itself against this by only keeping the interface and device-flash function constant, but allowing the flash to manage itself with the information it holds, like ECC for example. Version 2.0 is already in the works and plans to quadruple the interface speed, which should see massive improvements to I/O performance over a standard hard drive.
We were treated to a real life benchmark of a Google Earth simulation & Adobe Photoshop Elements on two identical Santa Rosa platforms (2.2GHz Core 2 Duo and 1GB of memory): one with Intel Turbo Memory and one without. The one with Intel Turbo Memory was nearly twice as fast as the one without, clocking in at 68 seconds to a non-Boosted 120 seconds. Whilst Google Earth didn’t differentiate them much, Adobe Photoshop Elements image manipulation used the high I/O the cache offers far more providing the large gap in the result.
It obviously depends how much you hammer the hard drive for random writes, so this makes it excellent for a page file, however no indication was given as to whether the benchmark was run before, therefore caching the images into the flash memory beforehand. However, remember this is merely a demonstration, not a final definitive benchmark. You’ll have to wait until we get a hold of some Turbo Memory kit in the bit-tech
labs for that!
Are you excited about future NAND or SATA developments? Or do you think they should have done more? Let us know in the forums