Intel Makes Caching Easy
SSD caching, Intel-style, is constrained by a handful of requirements.
- You have to be using a Core i3, Core i5, or Core i7 processor. As far as we know, caching will only be enabled on Z68 Express, so this narrows the list even further to Sandy Bridge-based LGA 1155 CPUs.
- You need to have the 32- or 64-bit version of Windows Vista, 7, or Server 2008 installed.
- Intel’s Desktop/Workstation/Server Express Chipset SATA RAID controller hub has to be installed and enabled (the platform controller hub needs to be in RAID mode, in other words).
- The system has to be RAID-ready with the Accelerate bit enabled (presumably, Z68 will be the only platform with this switched on initially).
- You need a SATA-based SSD with at least 18.6 GB of free space.
- You need a hard drive present with no recovery volume.
Given those requirements, there are also a handful of limitations. First and most important, the maximum cache size is 64 GB. I like this ceiling. If you have an 80 GB SSD, use it as a boot drive. At 64 GB or less, your drive is a good candidate to serve as cache instead. You can also only have one accelerated disk per system, and once you have a cache device configured, you can’t add a recovery volume.
Setup is a piece of cake. The “best practice” is to install Windows and all of your apps on conventional storage. Add the SSD, which Intel’s Rapid Storage Technology 10.5 driver recognizes and displays in the Storage System View of the Status pane.
Click the Accelerate box up at the top of RST, then click the Enable Acceleration link.
The software pops open a configuration box that lets you dial in the desired settings. First, you pick the SSD to use (a particularly easy choice in systems with a single solid-state drive). Then you pick the size of the cache, between 18.6 and 64 GB. Since the entire purpose of our 40 GB X25-V is caching, we allocate the whole thing. Choose the disk to accelerate—again, an easy decision in a single-drive configuration. Finally, pick Enhanced or Maximized acceleration mode.
Enhanced is default; it employs a write-through strategy to keep data synchronized between the SSD and hard disk. If you lose your solid-state drive, no data is lost. Maximized mode gets a performance boost from write-back caching. Because the SSD writes to the disk in intervals, however, you risk data loss if the solid-state device drops offline or fails before it gets synchronized. As we’ll show in the benchmarks, there is a quantifiable benefit to using Maximized mode, but it’s not worth rolling the dice with your data. Should you wish to play with both modes, Intel does facilitate swapping back and forth dynamically.
When all is said and done, you’re presented with a confirmation page that identifies the accelerated disk, the SSD serving as cache, and the selected mode. From there, you’re free to start using your workstation. What benefits can you expect? Let’s check out the benchmarks:
interesting read
Nice. Thanks for the quick heads up, I was just starting to build my new rig but now that I read this I'm going to wait for the Z68 MOBOs
I am one of those ssd+hdd users who prefer manually managing the drives. Recently I discovered some of the older ssd + hdd cache devices(Silverstone made one). I wasn't impressed. However,I could see the Intel cache set-up as being advantageous for me. Why? I have a boot SSD, large storage HDD, and a third SSD. I could still boot from the boot drive, then use the second SSD and HDD together. I like that idea enough to wait for the Z chipset before I ditch the H chipset.
compton, problem with that SilverStone unit was that it didn't have any intelligence built-in--it was simply mapping the first sectors of the hard drive, if what I remember reading a year ago was right...
waiting for Bulldozer...
Excellent Work, Mr Angelini! Now it's my job to make sure the motherboard manufacturers follow through!
-Your Adversarial Colleague
wow ...
I am thinking what is more restricting ... a ZFS supporting OS or the stupid Intel list for SSD caching.
The only impresive part of Sandy Bridge is the single-threaded performance. Everything else is a disaster (chipsets, QuickSync restrictions, price, linux drivers and bugs etc.) or was already available in previous generations.
Waiting on AMD Bulldozer and Llano ... I just hope those 2 won't be similar disasters.
wow ... I am thinking what is more restricting ... a ZFS supporting OS or the stupid Intel list for SSD caching.The only impresive part of Sandy Bridge is the single-threaded performance. Everything else is a disaster (chipsets, QuickSync restrictions, price, linux drivers and bugs etc.) or was already available in previous generations.Waiting on AMD Bulldozer and Llano ... I just hope those 2 won't be similar disasters.
In regards to Linux, isn't that like saying "Toyotas are junk because they're always dirty"? I mean, Linux is maintained by its "owners".
So would ssd caching work on a raid 0 setup with 2 samsung spinpoint F3's or would this add an additional risky element without much performance gain?...or say a raid 1 where I have backup...would it cache both drives or 1?..Sorry new to this and also waiting for a z68.
So would ssd caching work on a raid 0 setup with 2 samsung spinpoint F3's or would this add an additional risky element without much performance gain?...or say a raid 1 where I have backup...would it cache both drives or 1?..Sorry new to this and also waiting for a z68.
Yes, so long as all members of the array are hard disks.
P67 and Z68 are both incomplete chipsets. You still have to pay extra for virtue if you don't want to swap cable. I'm actually fine with that it's just that going down this path is like selling a modular design. I'm expecting intel's P77 chipset to be missing disk controllers but motherboard manufacturers could opt for marvell, jmicron or others to supplement that shortage. Intel does make some fine processors today, but their chipsets disappoint. But the good news is, maybe one day with the help of motherboard vendors we could pair a AMD chipset (which is believe to be superior) with an Intel processor.
P67 and Z68 are both incomplete chipsets. You still have to pay extra for virtue if you don't want to swap cable. I'm actually fine with that it's just that going down this path is like selling a modular design. I'm expecting intel's P77 chipset to be missing disk controllers but motherboard manufacturers could opt for marvell, jmicron or others to supplement that shortage. Intel does make some fine processors today, but their chipsets disappoint. But the good news is, maybe one day with the help of motherboard vendors we could pair a AMD chipset (which is believe to be superior) with an Intel processor.
Chris probably won't say anything, but as a motherboard tester I've found that the Intel features that do work "right", work better. That includes drive controllers, so it really comes down to a choice of a bunch of good features or a few great ones.
I don't understand why cached SSD/HDD is so far from pure SSD. Once something is cached to the SSD shouldn't the performance be nearly identical? Seems like this type of technology needs more work.
Had Intel not imposed such limitations on Sandy Bridge, they'd not need so many motherboard chipsets for a start, plus you can only imagine what a monster it could have been to start off with.
So yeah hooray for intel, ssd caching is just taking performance down (note, this is an option aimed for power user, they are kind of guy who pay 300$ bucks and then choose the worst way to get perf...), quick sync is a unstable restricted piece of crap, transcoding media is a top priority... what about a trim support in raid a array? something that is really needed by power user and not those wanabee features
Had Intel not imposed such limitations on Sandy Bridge, they'd not need so many motherboard chipsets for a start, plus you can only imagine what a monster it could have been to start off with.
Intel, like most other companies in this business, is known for using feature limitations to push more-expensive platforms.
So yeah hooray for intel, ssd caching is just taking performance down (note, this is an option aimed for power user, they are kind of guy who pay 300$ bucks and then choose the worst way to get perf...), quick sync is a unstable restricted piece of crap, transcoding media is a top priority... what about a trim support in raid a array? something that is really needed by power user and not those wanabee features
If you have the money for a huge SSD, go for it! But don't Sandforce controllers already have their own built-in garbage collection that practically negates the need for TRIM?
You can get some benefits of SSD caching by doing different RAID setups between a SSD disk and a partition on magnetic disk.
I could see this working out very well for a cad or cam software where your pulling up the same huge files day in and day out off of a storage drive. I will look forward to this for my next workstation.
Useless. Why would you use SSD as caching? Why would you pair your Nvidia card with that useless silicon called HD3000? Looks like Intel is running out of good ideas so they throw all this useless technology, what a waste.