PDA

View Full Version : Intel Application Accelerator???


BBA
09-02-2002, 12:13 PM
I have come to the conclusion the Intel Application Accelerator is a little screwy.

I just put in a WD 80M drive as my system drive ( removed the SCSI 10K stuff ). It's HD Tach performance is as good as the 10K SCSI drive I had, except the disk access times are a little higher.

I was checking Sandra for file system performance...and the 80G drive was showing about 25M/sec, and Sandra said disk caching was not enabled. Of course, it looked like it was enabled to me from device manager...so I decided the Intel drivers may do it.

I loaded the latest version of Intel App Accellerator, V2.2, and re-ran Sandra. The performance went down to 16M/sec with 10-14% processor useage as reported by HDTach, and HDTach throughput did not change. Something screwey was going on...so I ran the Intel App Accel aplication, and configured the drives for max performance. I had to reboot.

Sandra now showed 17M/sec, 10-14% CPU still. Not much difference. I uninstalled the Intel App Accelerator, rebooted and rechecked sandra, still shows only 17M/sec, but CPU use is now down to 4-5% where it should be for ATA drives.

Of course, the whole time HDTach showed the perf starting at about 48MB/sec, then tapering to about 28MB/sec, which I think is normal. The only thing is HDTach does not measure file system performance...only hardware level disk throughput. So I know my actual file system perf is screwey now ( as evidenced by Sandra )

How do I get fs performance back or is Sandra just really that screwed up?

NDC
09-02-2002, 12:37 PM
Well, I don't know the rest of you who have used Sandra for HDD benchmarks, but for me, it has been the most inaccurate benchmark util for me. Sure Sandra does seem to have consistency on HDD benchmarks, but not very accurate when comparing two different drives side by side. Sandra would report that my IDE 7200rpm drive is faster than my SCSI 10,000rpm drive which blew me off my seat! That's when I broke out the HDTach utlity and did a benchmark and saw more accurate readings for burst speed as well as seek times. I would go with the HDTach readings vs. Sandra readings anyday...