Accelerate Your Mac! - News and Reviews

Performance reports on Adaptec Ultra160
64-bit PCI SCSI cards

Published: 3/16/2000

Updated 3/12/2001 for G4/500 MP Triple X15 Cheetah RAID
Updated 9/16/2000 for
G4/MP RAID report
Updated 7/12/2000 for more 33/66Mhz PCI slot tests
Updated 8/20/2001 for ATTO Ultra3 vs Adaptec 29160 comments



1.02 Firmware Update: [11/10/2000] Although it's not listed at Adaptec's 39160 product page as of 11/10/2000, their FTP site has a 1.02 updater for the 39160. You should use the PowerDomain 6.1.1 control panel with this update (also at their FTP site). The readme notes this fixes issues with multiple 39160s in B&W G3 and later Macs, but warns that installing more than one 39160 in a Beige G3 will cause the system not to boot. I retested my Series 1 (slower model) Atlas 10K and the differences in v1.02 vs. 1.02 firmware performance was nil, but that drive is not a great performer compared to the current series (or other brands of 10K drives) and clearly a limiting factor in performance.

Direct download links:

I didn't see any performance benefit with my single Atlas 10K series 1 drive but reader Tim Seufert reported that in a G4 Gigabit ethernet system, he saw significant improvements in write speeds with his RAID array. (Lower than expected write speeds was a common complaint here previously, noted on the 39160 performance page).

" Mike,
Just wanted to drop you a note that with the new firmware version, I'm now getting full write performance on my 39160. With a 3-drive RAID 0 volume (Seagate Cheetah 18XL disks, all on one channel of the 39160) Atto's utility reports ~110 MB/s peak and ~90 MB/s sustained write rates.

It also seems to have fixed a Linux problem I was having where the generic driver for all Adaptec cards (aic7xxx) would lock up during boot if there were no drives on one of the 39160's channels. Must have been some kind of bad hardware initialization.

There is one other variable: the 39160 is now in a Gigabit Ethernet Sawtooth, not a B&W G3 like before. I can't test whether the firmware update would improve performance in a B&W. Prior to the update, I was still getting limited write rates in the Sawtooth. The limit was higher than the B&W, about 60 MB/s instead of 40, but definitely there.
Tim "

On 11/16/2000, an Adaptec Test Engineer replied to the above with comments:

" Regarding the posting [above]:

First let me say that no changes made in the 1.0.2 release specifically targeting any speed related issues with the 39160 card.

Regarding the user who saw a speed increase from what he saw in his B&W G3, he would see the speed drop back to his previous results if he were able to go back an retest in a B&W G3. We have seen that the B&W G3 limits the write speed of the 39160. We've also seen particular drives and/or particular drivers which could artificially limit the speeds achieved with the 39160 even outside of a B&W G3 environment.
--
John Tessier
Test Lab Engineer"



The page contains comments from readers and news pages relating to experiences with Ultra160 Adaptec SCSI Cards (39160 dual channel and 29160 single channel). Many owners originally noted (as I did) lower than expected write performance. The original post from the 3/16/2000 news page appears first, then reader replies on their experiences follow.

Ultra160 SCSI Adaptec 39160/Quantum Atlas 10K Performance: I recently acquired an Adaptec 64-Bit dual-channel Ultra 160 SCSI card (39160) and a Quantum Atlas 10K (10,000RPM/U160 SCSI) 18.2GB drive. The performance in ATTO's benchmark was impressive on reads (over 110MB/sec peak rates) but there appears to be literally no write caching (little difference in peak/sustained writes) on the drive. Testing with Apple's OS 9 drivers and Softraid made no improvement in write performance (Softraid did have very slightly higher rates). To ensure that my G4/450 (2.6 stepping) wasn't affecting things I also tested the card/drive in a rev 2 B&W G3 with similar results.

Ultra160 Performance

I'm running the latest 1.01 firmware for the 39160 card and PowerDomain 5.1 control panel shows the interface is set for 80MHz.

A past reader review of a U2 SCSI/Atlas 10K also showed much higher peak read than write performance, so perhaps this is just the nature of the Quantum drive.


More on 39160 Write Performance: Rod Paine of Astec (a Mac VAR) replied to my earlier post that he's also seen lower write performance with the 39160 than a 2940U2W:

"I have seen the same thing with the v1.0 39160 and SoftRAID 2.2.1 with Seagate Cheetah ST39103LW's and also with the 29160 and a 3-drive SoftRAID 2.2.1, which I am currently running in my B&W G3 450. (see attached)

Ultra160 3 drive RAID results

Oddly enough, any one of the Seagate's used by itself with a 2940U2W retains decent Write data rates in a Beige G3. (see attached)

2940U2W single drive results

I've played with Page Mode settings on the Cheetah's, but can't improve the Write data rate to any extent. I also tried Remus 1.4 Raid software and it exhibited the same slow Write data rates.

It's an interesting problem and I've seen it with IBM drives too, so it isn't limited to Quantum or Seagate and may well be a G3/G4 or Adaptec issue. Since our clients are so pleased with the Read data rate (file opening time) I haven't done any thing further about this Write data rate quirk. But you're correct, it's there!
-Rod"

I suspected this was a Quantum drive issue, but based on his comments that's apparently not the case. Note that what I commented on was lower peak writes (as if no write caching was enabled), since the sustained rates were near what Quantum lists as the max for the drive. On Adaptec's suggestion, I tried disabling Softraid's read/write acceleration which didn't help. As noted earlier using the Apple OS 9 drive also didn't help.

I've not yet tried testing with other than the default 'immediate' PCI Bus settings (PowerDomain v5.1 has a slider to adjust how the card shares the bus, which was called Video Tuner in older versions).


More on Low SCSI Write Speeds: Although many report the issue may be related to the 29160/39160 cards (as noted in yesterday's news), one common thread seems to be B&W G3 or G4 systems. Here's the latest reader comments on the issue:

" I too have the problem of slow write speeds on a rev 1 B&W G3. I have 2 9GB IBM 7200 RPM drives stripped using softraid 2.2.1. A jackhammer card (using 1.4.5 control panel) is handling the connection. I used to have this setup connected to a power computing power centre 132 and on this I get equal write and read speeds of about 15MB/s. Move the card over to the G3 and read ups to 17.5 MB/s but the write speed seems to top out at 7MB/s - quite literally a flat line in ATTO's benchmarking tests.

From what I can see this is related to the PCI bus in the B&W G3s. [ Note: I saw the same thing in a Rev 2 B&W G3 and a G4/AGP (Sawtooth) system as noted in the news yesterday-Mike] I have since put the card back into the power computing machine and the read and write speeds return to normal (i.e. the same speeds for read and write).

I have tried using OS8.6 and 9.0, various drivers even using a single drive instead of 2 stripped and I always get this 7MB/s write ceiling on the B&W G3.

It sounds very similar to the report concerning the slow write speeds you posted on the 16th of March.

I wonder if the problem is not that the SCSI cards in question have a conflict with the new PCI bus of the new G3s - hence why a beige G3 does not exhibit the problem.

Hope you get to the bottom of this as I'd love to solve my slow write problem too!
Davide "

[Note: A B&W G3 owner with another brand of SCSI card replied to this post saying he's seen good write performance. This may yet be some issue with the 29160/39160 cards.]

Another reader reports he returned a 39160 due to low write speeds:

"hi
we recently returned a 39160 because the write rates were much slower than with even a single channel atto lvd card. We had two 9 gig IBM ultrastars striped as a raid array. We were capturing uncompressed video through the ICE sdi option and the 39160 was dropping too many frames - however the exact same drives connected to the atto card didn't drop any frames.

i'm planning on examining atto's dual channel ultra160 card soon to see what sort of results we get with it.
regards
steve sloan
dB Film Company"

I've asked Steve what Mac model he's using.


Beige G3/333 Results with Quantum Atlas V: (Added 5/3/2000)

" Here is my test of a PD 39160 Dual Channel card in a Beige G3MT/AV 333 Mhz system. I tested it with the Quantum Atlas V hard drive that has 4 Meg of Cache and spins at 7200 RPM. Mounted Internally.

I have to also note that the Memory Read and write (performance) When up from 64 Meg/sec to 68 Meg/sec. Over all system feel is much faster than the ATTO PCI SCSI Wide card that came Stock with this Mac. With the Atlas V drive connected to the ATTO PCI card it Maxed out at 38 MB/s on the reads and writes. The PD 39160 card to me Makes the system so much faster, (at least the feel of it).
Sincerely,
David Brown"

Beige G3 results w/Atlas V

Note: His comment on 80MHz bus speed means the Adaptec SCSI Card clock speed (80MHz = Ultra160 SCSI). The standard bus speed of the Beige G3 is 66MHz (although it can be set to 83MHz).

What's interesting is that his sustained rates are higher than the same 39160 (64-Bit PCI) card in my G4/450 with 64-Bit PCI Slots and a faster CPU and a Atlas 10K 10,000 RPM drive (series 1). I can only deduce that the series 1 Atlas 10K's are not as fast as the lower priced Atlas V drive. (The Atlas V has a 4MB cache David says, my Atlas 10K has a 2MB cache.) Insiders tell me the series 2 Atlas 10K's are much better performers than the series 1. Series 2 drives should be available soon.

Summary:

Considering the large sum of money the Adaptec 39160 Ultra160 SCSI Card costs ($400+ retail) as well as the price for the drives, I don't consider the value to be appealing for most users unless you need a large RAID Array. For instance the IBM 75GXP 45GB drive delivers about 35MB/sec with a PowerMac G4 onboard ATA/66 controller and the under Maxtor DiamondMax Plus 40GB drive delivered nearly 30MB/sec sustained read and write rates with the onboard ATA/33 controller in the B&W G3.

Although SCSI is far more flexible and supports more drives, if you have an modern IDE controller you may want to give one of the large, fast, affordable IDE drive a try before you drop a lot of money on high end SCSI hardware. If you have a ton of money (say $5000-$6000) - see the report at the bottom of the page for a 8 drive X15 Cheetah RAID array's performance that is stunning.


More 66MHz PCI Slot Performance Tests: [added 7/12/2000]
The 2nd reader to note much better performance (esp. writes) when using a fast SCSI card in the 66MHz PCI slot. Not all SCSI cards are 66mhz PCI slot compatible, but the Adaptec U160 and ATTO Ultra3 cards are. (Below this post is a May 2000 table of test results comparing 33mhz vs 66mhz PCI slot performance.)

" Hey Mike,
Some cool SCSI info for your readers. I've been reading a ton of bad things on the internet about the Adaptec Ultra/160 cards. Everyone's complaint seems to be about the poor sustained write performance. I just wanted you to know that the EXACT same problem exists with the new ATTO Ultra/160 card (at least in my Yikes G4). The only way I've managed to get around this strange performance problem is when I put the ATTO card in the 66mhz PCI slot where the video card normally resides. I personally felt a bit mislead as MacGurus made it clear that this problem only existed with the Adaptec cards. I just thought some readers should be aware of this before they went out and bought the VERY expensive ATTO card as it is not exempt from this strange performance glitch.

In the 64 bit/33mhz slot I was flat-lining at 50 MB/sec for sustained writes. After putting the card into the 32 bit/66mhz slot check out my ATTO benchmark results. TRULY AMAZING!!!!
andre

Yikes G4 @ 350mhz
1 Gig of ram
ATTO Ultra/160 Dual Channel SCSI card
Six 18 Gig Seagate Cheetah's (15,000 RPM X15 models) "

66MHz PCL Slot RAID Array performance

I asked if he could re-run the test with a sample size of 2 (vs 1). He later said the results with sample size of 2 was practially identical to the above graph.

Adaptec 39160 Ultra160 SCSI Low Write Performance Discovery: [added 5/13/2000]
Tim Seufert wrote with some very interesting test results regarding the low write performance of the Adaptec 39160 64Bit/Ultra 160 SCSI card noted here previously:

" Mike,
I've just made an interesting discovery regarding the slow write rates with the 39160. I'm using it in a B&W G3 with some 9GB Cheetah 18XLs (4th-gen Cheetah) which I just bought. They're about 34-35 MB/s read and write, and they're very quiet for 10K RPM drives -- somehow they damped out the high frequencies so you mostly just hear a low frequency hum, and the seek noise is barely louder than idle. However I'm going to have to get one of them replaced because it has 8000+ bad sectors. (!)

Like everybody else, I noticed that the Atto benchmark didn't show high peak rates for writes. (These Cheetahs do ship with write caching enabled BTW.) Peak read rates from the Cheetah onboard cache are very fast, 140 MB/s or so.

When I configured two of the drives as a RAID, I found that write speed seems to just hit a wall 39 MB/s. The write speed graph looks totally flat except for the very beginning.

But here's the really interesting thing: I tried moving my video card from the 66MHz 32-bit slot to one of the 33MHz 64-bit slots, and then put the 39160 to the 66MHz slot. (The 39160 is 66 MHz capable as well as being 64-bit.) The results are interesting to say the least:

Atto results for 2-drive RAID on 1 channel of 39160 (RAID volume 70% full):

PCI Slot
Used
33MHz/64-Bit PCI Slot
66MHz/32-Bit PCI Slot
Peak Read
140.05 MB/sec
97.13 MB/sec
Sustained Read
50.12 MB/sec
47.83 MB/sec
Peak Write
39.00 MB/sec
62.92 MB/sec
Sustained Write
38.23 MB/sec
56.44 MB/sec

Odd that reads decrease a bit, but the increase in write performance is substantial. Also, the Atto write graph for the 32-bit/66MHz slot doesn't flatline at a barrier value like it does for 64-bit/33MHz.

The two conclusions I draw from this are:

1. There's some kind of interaction limiting PCI write performance when going across the PCI-PCI bridge which connects the 33MHz 64bit bus to the 66MHz 32bit bus. Could be a driver problem, could be a hardware problem, could be that Apple isn't setting up the PCI-PCI bridge right.

2. Adaptec's driver isn't optimized well for the case of the card sitting in the 66 MHz slot, but despite that write rates improve simply because they're no longer bottlenecked by whatever the problem is when you go through the PCI-PCI bridge chip.

Tim Seufert "

One note: G4/AGP owners don't have the luxury of moving the card to a 66MHz/32Bit PCI slot, since the AGP slot replaced that one for video card use.


8 Drive RAID Array of X15 (15,000 RPM) Cheetah Drives: This report is from a 7/12/2000 entry in the Drive Compatibility database from Macgurus:

"base machine: G4 Sawtooth AGP 512MB w ATTO UL3D Dual-Channel Ultra3 SCSI host. PCI bus was also populated w Igniter w Component, and an ATI Nexus video card for a second display. MacOS 9.04 w G4 Update 2.4.

RAID 0, eight ST318451LW 18GB 15,000 rpm "X15" Cheetahs, bus duplexed with four drives per channel. Running SoftRAID 2.2.2b11 drivers.

MacBench 5 scores:

Disk: 4438
Pub Disk: 3956

Sust Reads 1024k: 184488.92 (184MB/sec)
Sust Writes 1024k: 124052.60 (124MB/sec)

Intech QuickBench Scores:

Xfer Size : 1 MByte
Seq Read : 168.098 (168MB/sec)
Seq Write : 130.774 (131MB/sec)
Random Read : 123.611 (124MB/sec)
Random Write : 120.135 (120MB/sec).

Raw benchmarks, including single drive, pairs, quads, etc., are posted in the benchmarks archive at ftp.macgurus.com. Discussion and narrative on the Upgrades Forums at www.macgurus.com.

yeee-haw!"


G4/500 500 Dual Processor w/3 15,000 RPM 18GB Cheetah Drives Report: (from the 3/12/2001 news page)

" Mike,
I thought I would drop a line and update you on my array experience. The problem turned out to be the APS drives had been built with SE [single ended-Mike] cables. The 39160 wouldn't operate faster than 20mhz. Apparently I alerted them to the fact that many of the drives had been put together with the wrong cables. To their credit, they took them back.

Today I received a 3 drive array from Megahaus (I contacted Macgurus but they were swamped and took almost two weeks to return inquiries) of 18 gig Cheetah 15K. Tomorrow I receive a 4 drive array of 9 gig Cheetah 10K. Megahaus is letting me test them and then return the slower one to exchange for another of whichever is faster. I have included a screen capture of my atto benchmark test for the 3 - 15k drives. I get a sustained Read of almost 100 mbs and a sustained Write of about 74 mbs with Softraid 2.2.2. I will test the 4 drive array tomorrow to see if it fairs better. ... I thought about experimenting with the Atto U3LD but read one of your readers' notes that it doesn't always perform any better than the 39160.
Thanks for all the info,
Stan

The Cheetah 18GB 15K RPM drives have a 4MB cache (per spec), but you don't see a big cache spike in this graph (only a small read spike). Interesting to see how the graph of reads shows progressively higher rates as the I/O size increases. I wish ATTO's benchmark had larger than 8MB I/O size options.


G4 500 Dual Processor/Cheetah 39160 RAID Report: (From the 9/15/2000 news page)

Although past results seemed to indicate the Adaptec 39160 Ultra160 SCSI card had less than expected write performance (noted above), a reader with a G4/500 MP system and 39160 card sent some impressive results with a 3 drive 10,000 RPM Cheetah RAID array - over 100MB/sec sustained writes. Perhaps the new G4 motherboards have better PCI bus performance than previous Sawtooths or the dual processors are a factor (I assumed OS X would be needed for full MP support at the file IO level, but maybe not).

He's also going to be testing some 18GB 15,000 RPM Cheetahs in the same system soon and promised to send the results. When my G4/500 MP system arrives (due around 9/20/2000 or so) I'll try some tests with my 39160, but I don't have drives as fast as the 10K Cheetah to use.

Note: A reader replied to the above results saying his 72GB Cheetah RAID/39160 setup in a G4/450 MP did not deliver near the rates shown above. I've asked him to send a graph of his results for posting here.

Adaptec 29160 vs ATTO Ultra3 SCSI Card Comments: [added 8/20/2001]

" Hi Mike,
You may remember I purchased an Adaptec 29160 Ultra160 card for my G4 733 for use with two Ultra160 (Fujitsu) drives, and had some problems. Using other brand hard drives, pre-release drivers from Intech, and support from Adaptec, didn't fix problems with OS X, sleep, option bootup and SCSI mode corruption.

I'm happy to note that buy.com refunded my 29160, and I purchased an ATTO UL3S.

Unfortunately since the ATTO UL3S doesn't include a LVD cable like the 29160 kit, the ATTO was over $100 more in the end. However I'm very happy with it so far:

  1. The ATTO scores 41mb/sec on 10mb+ files, versus 38mb/sec on the Adaptec.
  2. The ATTO does not corrupt my drives to SCSI 1. All of my drives have remained Ultra160 even in OS X.
  3. OS X support out of the box! No beta drivers, no firmware update. OS X installed and runs perfectly.
  4. Holding the option-key at bootup lets me choose my startup drive without crashing to OpenFirmware!
  5. It works in both PC and Mac! Making the card easy to move/resell. The 29160 is made per platform.
  6. Unlike the 29160, I can attach slower such as a CDROM, and the Ultra160 bus doesn't slow down.

Sleep seems to be disabled at the moment, in OS X and OS 9. I believe PCI sleep with some cards doesn't work in the G4 period, but that's okay.

... I'm sure their [Adaptec] Ultra160 Mac cards will mature eventually. However, if you could put a brief comment regarding the benefits of going with ATTO versus Adaptec for Ultra160, on your site, it might help to save other users my headaches.
Thanks!
Rich "

For more info on the ATTO Ultra3 SCSI cards, visit www.attotech.com.




Related Links (from 2000/2001):


Back to XLR8YOURMAC.COM

Copyright © 2000, all rights reserved.
All brand or product names mentioned here are properties of their respective companies.

Disclaimer: Users must read and are bound by the Site Terms & Conditions of Use.