Now we come to RAID 5. nVidia added RAID 5 to the nForce4 for Intel chipset, an improvement over the nForce4 for Athlon 64, which doesnâ€™t have this feature. The reason it was able to do this is because as Intel chips donâ€™t have an integrated memory controller nVidia needed to add one to its chipset. To do this it reverted to a traditional two-chip piece of hardware so it used an â€˜off-the-shelfâ€™ Southbridge, which happens to have RAID 5. The key phrase is that the chipset â€˜supports RAID 5â€™ i.e. itâ€™s done in software but if it could deliver the goods effectively free of charge this would be big news.
nForce4 includes a Windows based utility called Media Shield which you can use to control your RAID array without fiddling around in the BIOS. We deleted the mirrored RAID array and plugged in a third Seagate drive, so we had three individual drives showing in Media Shield. Then we selected the type of RAID that we wanted. The wizard is intelligent and only shows the options that are supported so you wonâ€™t see RAID 5 if you only have two drives installed.
If youâ€™re migrating from a single drive which contains data you can mark the drive at this stage and the data will be preserved while the array is created. We used this process and it worked perfectly but if youâ€™re paranoid enough to use RAID 5 then we feel sure that you wouldnâ€™t consider this step without making a full back-up to DVD.
Some hours later we had a healthy RAID 5 array and could get on with our testing.
Well, what can we say? The test results were an eye-opener as the read speeds looked good. However, the write speed in HD Tach was absolutely appalling and crawled along at 4.6MB/second which was ten per cent of the speed of a single drive. In our 4GB transfer test the write time was 15 minutes 34 seconds which was six times slower than we would have expected, and PCMark05 also returned a low score.
Naturally we doubted the test results but you have to remember that the HD Tach test is run with the array as an unformatted Slave while the transfer test uses it as a simple data drive. Itâ€™s not until we get to PCMark05 that we install Windows so thereâ€™s no question of a driver installation problem. For the record we used nForce4 7.13 drivers. We had been using the quick format option during Windows installation so we re-ran the tests using a full format which took five and a half hours and the results were exactly the same. On this evidence we strongly suggest that you forget about using the integrated RAID 5 feature in nForce4 for Intel, but we did note one interesting point which is that CPU usage was very low. In the past we have seen software RAID 5 with a CPU overhead of 40 per cent so we have to consider the possibility that nForce 4 was designed to give your CPU an easy life with RAID 5. Whether this is the case or not, performance is so low as to make the integrated RAID 5 unattractive but nonetheless we checked out its capabilities by unplugging one of the three Seagate drives and Media Shield correctly alerted us that the array was degraded.
When we reconnected the drive the software reported that all three drives were present and correct but split into a degraded array and a healthy drive. It took many hours for the array to rebuild itself but this is quite usual.