Conclusions
With so many benchmark results to consider, it's no wonder I'm having a hard time picking a real winner here. None of the cards we tested is good at everything. They all have their Achilles' heels, and no card really rises above and dominates throughout. Let's run through them all and see what we've got.
Overall, Adaptec's 2400A turns in probably the worst performance of any of the cards in this comparison. Sure, it has its bright spots, but it's got real problems with RAID 1, especially in response times. The 2400A's rebuild times are atrocious, and its disk access times and file read performances aren't that hot, either. The 2400A's one saving grace, performance-wise, is its incredibly low CPU utilization across all of our IOMeter tests.
Other than a generally subpar performance, the 2400A has a lot to offer. Its array management software is the slickest of the lot, you can upgrade the cache to 128MB, and the card itself even comes bundled with a 32MB DIMM right in the box. All those extra features are nice, but I don't think they're enough to get by the card's poor performance, especially when you consider its $323 street price.
At the other end of the price spectrum we have HighPoint's RocketRAID 133, whose performance was really a pleasant surprise for me. I expected the diminutive two-channel card to be buried by its more expensive four-channel competition, but that wasn't the case. In only a few tests did the RocketRAID's two IDE channels become a serious limitation, and then only in four-drive RAID 0 and 0+1 configurations. Otherwise, the RocketRAID 133's performance was quite impressive, especially considering it costs $100 less than the next-cheapest card.
Because the RocketRAID 133 doesn't support RAID 5, and because its two-channel design hinders four-drive performance in server-class benchmarks like IOMeter, the RocketRAID 133 is more appropriate for an enthusiast's desktop than a high-volume server or even a workstation. As happy as I'd be running the RocketRAID 133 in my main rig, the one year warranty bothers me.
3ware's Escalade 7500 is an intersting card to talk about because its results are all over the place. In particular, the card excels at disk read tests, but fares quite poorly when it comes to disk writes. In the end, the two extremes can average out to yield great overall performance, but it all really depends on whether the tests are more biased towards reading from the disk, or writing to it.
The Escalade 7500's IOMeter performance bodes well for the card's suitability for demanding server and workstation environments, and its low rebuild times and RAID 10 support make it especially attractive for mission-critical servers. Still, it would really be nice to have an expandable cache and some error code LEDs for troubleshooting and array monitoring. 3ware does offer a set of open-source Linux drivers that are apparently pretty decent. The card's $252 street price seems pretty reasonable, too.
Finally, we come to Promise's FastTrak SX4000, the only card we tested that's not ready to roll right out of the box. But that's all right, since even with the cost of a 128MB PC133 ECC DIMM included, the FastTrak's price is still the second lowest of the lot. When you tally everything up, the FastTrak SX4000 is probably in a dead heat with 3ware's Escalade 7500 for the overall performance crown. But the FastTrak SX4000 has an edge because its performance is far more consistent from test to test.
The Promise card's Achilles' heel is its CPU utilization in IOMeter's server and workstation simulations, which is the highest of any card we tested, regardless of array type. The FastTrak is an affordable solution, and its performance is generally quite good, but its CPU utilization might be enough to sink it among high-end workstation and server users, as may its shorter, two-year warranty.
Now comes the hard part: picking one to throw my weight behind. Honestly, I can't pick a winner. Each card is good for specific applications, so here are my picks:
127 comments — Last by Anonymous at 2:13 AM on 04/10/03
| Intel's 520 Series solid-state driveA new muse for Intel's 25-nm NAND | 81 | |
| Samsung's 830 Series solid-state driveThe dark horse rides again | 61 | |
| SSD performance scaling across the spectrumSize matters, but how much? | 88 | |
| OCZ's Octane 512GB solid-state driveIndilinx returns with Everest | 41 | |
| Seagate's Momentus XT 750GB hybrid hard driveAdaptive memory enters its second generation | 66 | |
| A quick look at OCZ's 2.15 SSD firmwareBSODs begone? | 69 | |
| SSD scaling outside the sweet spotA look at performance scaling from 120-320GB | 44 | |
| SSDs in the sweet spot: Modern contenders at 120-128GBSurveying the meat of the market | 171 |
| Friday night topic: The trouble with Best Buy | 131 |