Target PC :: Quantum Atlas V

Bonnie Benchmarks

100mb test file			500mb test file

Writes:	Per Char	7184	Writes:	Per Char	7026
	Block	44236		Block	34156
	Re-Write	29075		Re-Write	12645
Reads:			Reads:
	Per Char	7487		Per Char	7290
	Block	178471		Block	28377
Random Seeks:		18032	Random Seeks:		255

To test the raw throughput of the drive, I downloaded and newely released RedHat Linux 6.2 ISO which weighs in at a hefty 650mb. Performing partition to partition moving of the image took an average of 53 seconds. Do that math and you see that the throughput was only in the 12.5mb/sec range, not very impressive when looking at the synthetic benchmarks. But remember that in actuality the drive was both reading and writing at 12.5mb/sec, which fits very closly with the other benchmarks.

After I finished benchmarking the drive under Linux, I again tried to get NT to format the drive with no luck. A quick check on Quantums web site netted me the the telephone support phone number, I was quite happy to see that there was no little note "a $20 service charge is applied to each case". After wandering through several automated voice menu systems, I found myself on hold for only about 2 min before I was speaking with a support technician. I described my system setup and the problems I was having, we went through several routine checks just to make sure I hadn't gotten up to early that morning and forgot to plug the darn thing in. The drive checked out fine, and it was concluded that there was something screwey with the partition table that nether a low-level format nor fdisk could fix. The solution was a 60 second session in the old DOS utility Debug, which we used to systematically erased every single byte on the drive. A quick fdisk once this was finished and I was into benchmarking the drive. Both NT, Win98, and DOS were now able to format the drive at will, and my Windows benchmarking commensed.

First up under NT was SiSoft Sandra 2000 Pro. Sandra interfaces with the drive via the file system (which is why we see some large differences in the Win98 and NT benchmarks) and streams data from system RAM to ensure that all bottlenecks encountered are those of the drive and the are not higher up in the I/O subsystem.

NTFS
Buffered Read	395MB/sec
Sequential Read	27MB/sec
Random Read	10MB/sec
Buffered Write	249MB/sec
Sequential Write	15MB/sec
Random Write	33MB/sec
Average Access Time	4ms

What can I say, this drive is smokin fast. It should be noted that during my benchmarking, everyonce in a while, I got extremely high benchmarks that I'm very suspect of. Notice the Random Write benchmark of 85mb/sec, this would seem to be a hugely exagurated figure, as correct number is more around 45mb/sec, although I could not pin down the source of the wild numbers I suspect that Sandra uses a benchmark file which is around 250MB, and sometimes NT was able to free enough RAM to cache the file, although that's just speculation. Under NTFS the drive shows it's muscle in the sequential read and write tests, both times right up with the Quantums specs.

FAT16
Buffered Read	366MB/sec
Sequential Read	27MB/sec
Random Read	10MB/sec
Buffered Write	250MB/sec
Sequential Write	13MB/sec
Random Write	19MB/sec
Average Access Time	4ms

Here we see that the drive also screams under a FAT file system. Interestingly enough, the drive did not exhibit the same wild benchmarking behavior as was seen under NTFS. It is quite evident how much of a performance difference the large cache makes when you look at the buffered read and write numbers, those just arn't comparable to drives with 2MB and less of cache.

Running HD Tach under Windows 98, the drive shows some performance on the outer tracks of the drive, with sequential reads peaking around 30MB/sec. Near the very inner tracks of the drive, performance takes a signifant hit with reads peaking at only around 15MB/sec. This is not really unexpected though, all drives have a similar read performance pattern simply because of the slower platter to head speed ratio near the inner tracks of the drive compared the outer tracks. The seek time's that HD Tach reported seem quite high, but this may be a factor of the FAT file system. The burst speeds reported are also substantially lower than reported in the other benchmarks, it seems almost as if it's reporting sequential read preformance, rest assured I did run the benchmark quite a few times, and the numbers were quite consistant. The CPU utalization is nice and low where we would expect it to be when using a SCSI based drive. Even during my streaming tests the CPU usage was quite low, hovering around the ~6% mark when under a full load.

FAT32
Buffered Read	107MB/sec
Sequential Read	28MB/sec
Random Read	9MB/sec
Buffered Write	94MB/sec
Sequential Write	15MB/sec
Random Write	13MB/sec
Average Access Time	4.3ms

One thing that you should be very aware of when running SCSI drives is heat. SCSI drives historically have run quite hot, and I expected that would probally need to use a drive cooler for benchmarking. However when I inquired, Quantum assured me that I wouldn't need any sort of cooling on the Atlas, that it is easily able to dissapate the heat it produces through the metal housing. I was interested to see just how accurate that statement was, as you can easily cook your supper on top of an older, uncooled Seagate Cheetah drive. During normal system use, Word Processing, MP3 listning, Net surfing, and The GIMP open in the background, the drive was barely warm to the touch, only 5C above case temperature (22C). When under a heavy load of benchmarking with Bonnie serving up MP3's at the same time it's a different story. The drive was much warmer during this marathon, hovering around the 34C mark (94F) which is definitly warm in my books considering my IDE drives run in the 25C range. I was surprised at how large of a difference there was in temperature when idle and under load, but even 35C doesn't hold a match to the nuclear rod temperatures of the Cheetah drives. So if your only going to be running a single Atlas V in a system and have decent system cooling, then a cooler is definitly not a requirement as long as you take into account the heat produced by the drive. However if your going to be running a dual or quad setup in a server, make absoutly sure you have an adequate cooling system as the heat produced by 4 drives under a constant load is nothing to scoff at.

When 7200RPM SCSI drives first appeared, and later IDE variants, they generally ran very loud, and many users found them very annoying. The Atlas V incorporates their Quantum's new "Quiet Drive" technology. When I first booted up the test-bed with the Atlas installed and the side off, I had to double check that indeed the drive was running. Boot-up is the loudest that the drive ever is, during it's 3 second spin-up and self POST, but once passed this stage the drive is amazingly quiet. The most noticable thing is that it doesn't "grind" during heavy use as IDE drives commonly do, the drive sounds almost as if it's muffled. If you sit fairly close to your system when your using it, you'll definitly notice that the Atlas runs nice and quiet.

One problem that may arise when the Atlas V is used in multiple drive configurations is power draw. The drive is rated at a peak power-up current draw of 2.2A on the +12 line. Alone, this is not a large draw, but when used in multiple disk configurations, the power requirements get fairly hefty and a 300watt or higher power supply should definitly be a requirement. Considering that your average 300 watt power supply kicks out 15Amps on the +12v line, even when running a dual disk setup, during power-up the drives will be sucking about 37% of the total available power, and power-up is when most devices draw the most power, so if your running a smaller power supply power draw should be a concern.

Despite the installation woe's I experienced, I believe this to be an isolated case, and not refelective of all drives. The specs the drive sports are very impressive, as equally are the benchmarks. The 4mb buffer and 7200 RPM spindle speed definitly shine through in the sequential read tests, the buffered reads are nothing short of spectatcular and lend the drive to being well suited for server situations. The Atlas V's performance with smaller sized files (under available system RAM size) where the 4mb of onboard cache and the system file caching are able to work in unison is simply astounding, and lends itself especially well toward server situations. Even when the file size was cranked up the benchmarks were still fairly good across the board, with numbers being pretty consistant. For an average user, the Atlas V's performance would very rarely be realized, and the investment for the drive and a SCSI controller would not make it very attractive. For power users and gamers who crave speed, the Atlas V certinally helps with map load times and should be considered, although the investment may deter a lot of users who do not already have a SCSI based system. But servers are the Atlas's stong point. Where most large SCSI hard drives of this size are in relativly the same price range, and performance is the backbone of a network is where the Atlas V shines. Power draw and heat production should be kept in mindwhen purchasing the Atlas V, although I would expect Power users and servers to already have adequate power and cooling systems, and these would not really be of concern.

Overall the Atlas V gets a 9½ for performance, bottom line is it's fast, very fast. I deducted ½ for combined power concerns and heat production, and another ½ for the installation problems I had. Leaving the Atlas V with a solid 8½ rating and skirting dangersouly close to receiving the coveted Editors Choice award, but the installation problems were the deciding factor. Although Tech Support was very good, I can't imagine a user new to SCSI having to first trouble shoot a problem like this, and then end up mucking around in MS Debug. © 2000

Andrew Oliver
00/04/29