SAS card for G5Quad

[flyproductions]

But we will see. 😉

Went fast, this time! 😃

The card arrived today and just passed a quick test. Did not boot right out of the box for me. I'll have a closer look at this. But for the rest, works fine. It really is four lane (as shown by System Info) while the Accelsior, even with a socket of same physical size, only shows up with two.

But, what's a bit disappointing, is the performance: Marginal differences (which more likely beeing caused by the different SSDs anyway) aside, it is just the exact same speed, which the G5's onboard SATA I delivers.

LSI SAS3041E-R, SAS

Powermac G5 mainboard SATA I

It even falls behind in random writes and small blocksizes. But, as mentioned, i think this is more SAMSUNG vs. SANDISK than SAS vs. SATA I. My main suspicion is, the card is just one lane per channel. This would make sense as it delivers nearly exact one lane speed, with the little Marvell SATA 3 card getting even more out of one lane.

But after all performance falls way behind the Accelsior even only having a two lane connector. With these speeds not even a RAID 0 of two disks would come close. Would be nice, if the lanes could be user-configured to the channels via jumper etc.. But for now it stays, that i have yet to see anything faster than the Accelsior S in a powermac.

 

[flyproductions]

So performance-question seems allready answerd!

I just downloaded the manual. And there, on page 17, it states the following:
"Supports SATA as defined in the Serial ATA Specification, version 1.0a."

So, when it comes to SATA, weare talking about SATA I. Means not much more than 130MB/s, even no matter what the speed of the PCIe is. Nice try! ...i wold say. 😒

 

[DearthnVader]

I always though SAS stood for SCSI Attached Storage?

Thus I thought RAID SAS controllers used high speed and pin count SCSI drives?

Wasn't there an Apple Fiber Channel RAID card for PowerMac's?

If so, what kind of speeds could that attain?

 

[Amethyst1]

I always though SAS stood for SCSI Attached Storage?

SAS means Serial Attached SCSI and is the successor to parallel SCSI, and you can conveniently hook up both SATA and SAS drives to SAS controllers.

 

[DearthnVader]

SAS means Serial Attached SCSI and is the successor to parallel SCSI, and you can conveniently hook up both SATA and SAS drives to SAS controllers.

Are SAS drives any faster than SATA drives in this configuration( with this controller )?

Anyway, there seemed to be two versions of the Apple Fiber Channel RAID card. One PCI-X and one PCI Express.

It would seem that the PCI Express card could hit 200MB/s or 400MB/s bidirectional.

While the PCI-X card could operating in 32bit PCI 66Mhz slots( among other modes ) such as the B&W and Yikes had. Giving a max theoretical speed of 266MB/s in this configuration.

Xserve RAID (SFP) - Technical Specifications - Apple Support

support.apple.com

 

[Amethyst1]

Are SAS drives any faster than SATA drives in this configuration( with this controller )?

The controller can do 3 Gbps per channel on a SAS drive.

 

[DearthnVader]

The controller can do 3 Gbps per channel on a SAS drive.

So 3 Gigabits Per Second / 8 = 375MB/s per channel?

 

[TzunamiOSX]

So 3 Gigabits Per Second / 8 = 375MB/s per channel?

Minus overhead, real is around 280 MB/s

 

[Amethyst1]

So 3 Gigabits Per Second / 8 = 375MB/s per channel?

Minus overhead, real is around 280 MB/s

SATA and SAS use 8b/10b encoding so 3 Gbps means 300 MB/s, minus overhead.

 

[DearthnVader]

SATA and SAS use 8b/10b encoding so 3 Gbps means 300 MB/s, minus overhead.

Ok, that always confuses me. SATA is 1.5Gbps ( 150MB/s ) and so on.

With Ethernet when we talk Gbps we always divide by 8 and USB and FW because Gigabits is not the same as GigaBytes.

Anyway, one PCI-E lane per channel?

Would this mean RAID 0 with 4 SAS drives could do 4x300MB/s? or still just 1x300MB/s?

 

[Amethyst1]

Would this mean RAID 0 with 4 SAS drives could do 4x300MB/s? or still just 1x300MB/s?

I'd suspect the former but keep in mind PCIe 1.0 ×4 is ≈1000 MB/s max, as the manual says (though it's inaccurate in terms of the SATA speed):

Each port is capable of 3.0 Gbit/s SAS link rates and 3.0 Gbit/s SATA link rates.

 

[DearthnVader]

I'd suspect the former but keep in mind PCIe 1.0 ×4 is ≈1000 MB/s max, as the manual says (though it's inaccurate in terms of the SATA speed):

So PCI-E 1.0 1x is 250MB/s max theoretical throughput.

So assuming we have the right SAS drives( 250MB/s peak capable single SAS drives if they exist? ) connected in RAID 0 4 drive array then we should be able to reach somewhere near the Max theoretical peak throughput with this controller in a 2005 G5 with PCI-E x4 connections?

Or whatever peak speed SAS drives we can get x4.

I understand there is always overhead, that peak very by drive and block size and such, speeds don't ways reach the Max theoretical throughput of the PCI bus, and there is always file system, kernel, and driver overhead in the OS.

Just generally speaking from Macs, controller, PCI bus type, I have seen some pretty good benchmarks when compared to the theoretical throughput of the PCI bus, the best being the ATTO 64bit 68pin SCSI RAID controllers in Powermacs with 64bit 33Mhz PCI slots.

I seem to recall the ATTO cards could hit about 240MB/s in benchmarks when in RAID 0 with the right drives and enough of them. That's pretty good when you consider the theoretical throughput of 266MB/s.

Also, I seem to recall seeing benchmarks of the SiL3112 cards in 66mhz slots hitting 200MB/s+ in the B&W and Yikes.

The Apple Fiber Channel RAID PCI-X version should be a nice card if we can find one and the Fiber Channel enclosure. If we could find a 64Bit PCI-E to PCI bridge card. To use the PCI-E Apple Fiber Channel Raid controller in a 64bit 33mhz PCI slot in a Powermac we likely could also get very near the theoretical throughput of 266MB,s.

Tho if I read it right the Fiber Channel enclosure has 7 PATA bays????

Or a PM G5 with PCI-X native slots( 100Mhz or 133Mhz ).

If I read that correct that's 64bit PCI@100/133Mhz. So 266MB/s x 3 or 4( roughly. Or a max theoretical throughput of 750+MB/s or 1000+MB/s.

Assuming we can find "compatible" drives that peak at over 250MB/s in a 4 drive RAID 0 array.

If the Fiber Channel RAID enclosure has PATA we should be able to use PATA to SATA adapters. Tho the best speeds I've seen in my QS G4 when connect to the Fast internal IDE bus is around 90MB/s in QuickBench using a 500MB/s SATA 6G SSD.

 

[DearthnVader]

From what a read, 4 of these should do quite nicely with QuickBench in RAID 0.

https://www.ebay.com/itm/2040655854...UM04My8G0VOt1ydIYEyeyV6RiPOD|tkp:BFBM5O-AmZJh( not my auction! )

With the SAS card that is the topic of this thread?

 

[flyproductions]

I always though SAS stood for SCSI Attached Storage?

The Delock, flashed with Firmtek/Seritek-firmware, i once had, even appeared as Parallel SCSI in System Info

And, while marketed as SATA II, showed similar performance. This "masking" or however it could be called seems to be the only way to make those things bootable in Mac OS / Open Firmware.

 

[flyproductions]

So 3 Gigabits Per Second / 8 = 375MB/s per channel?

Yes, but, beeing somehow "translated" to SATA I, only 150 minus overhead are left. And that's what we see.

 

[Amethyst1]

The Delock, flashed with Firmtek/Seritek-firmware, i once had, even appeared as Parallel SCSI in System Info

...but that doesn't turn it into a SCSI card.

If we could find a 64Bit PCI-E to PCI bridge card. To use the PCI-E Apple Fiber Channel Raid controller in a 64bit 33mhz PCI slot in a Powermac we likely could also get very near the theoretical throughput of 266MB,s.

This should do the trick 'cause PCI-X can fall back to 64-bit 33/66 MHz PCI speeds can't it?

 

[ifrit05]

This should do the trick 'cause PCI-X can fall back to 64-bit 33/66 MHz PCI speeds can't it?

Yes, in theory, but sadly it has been discontinued for some time.

 

[flyproductions]

So PCI-E 1.0 1x is 250MB/s max theoretical throughput.

The best, i got to see with single lane SATA II/III cards (AHCI, not bootable) was this Digitus

The Marvell, i still have, gets even a bit higher. But that's only for read.

Both with some OCZ SSD, i still use in the cube.

Also, I seem to recall seeing benchmarks of the SiL3112 cards in 66mhz slots hitting 200MB/s+ in the B&W and Yikes.

Just rechecked the Sawtooth which has one of them in one of the 33MHz slots. And i got little over 60 with an SSD. So will more likely be max 130 than 200.

If I read that correct that's 64bit PCI@100/133Mhz. So 266MB/s x 3 or 4( roughly. Or a max theoretical throughput of 750+MB/s or 1000+MB/s.

Yes, i have seen some Youtube Vids with a guy, owning one of those zillion "fastest Quads in the World", who hit something like 700, if i remember correctly. He went some other way, involving two Sonnet Tempo SSD, each with two disks on it and got all of them together in a RAID. Not bootable of course, as it's AHCI.

But you get just that or maybe 50 to 80 MB/s less, if you put just one SSD on each or take two Accelsiors, costing a quarter of the Sonnets, as all of that is bottlenecked by the two lane connectors of the cards.

Assuming we can find "compatible" drives that peak at over 250MB/s in a 4 drive RAID 0 array.

Maybe we can. But, considering the dramatic weakness of these RAIDs in random small block transactions, is this really worth all the hassle? I would not boot of a RAID again, even if it were bootable, just because of that. And, if booting is not a requirement, IMHO one is nice off with those 400MB/s, the Accelsior offers plug and play with one SSD and without this abysmal small block performance.

Yes, i must admit, i fell a bit in love with this thing! 😎

 

[flyproductions]

...but that doesn't turn it into a SCSI card.

To that, System Info has some other opinion. 😅

 

[flyproductions]

...and, i think this has been allready posted, but just to be complete, how the LSI SAS contoller is seen by the System:

 

[flyproductions]

Tried to get this thing to boot again, but it just does not.

But it looks like @Tratkazir_the_1st has at least a different firmware if not a different card. His shows up as Product: SAS304E, Revision: Firmware 1.26.0.0, Fcode 1.00.49. While mine shows Product: SASWT4I (even with SAS3041E at a label on the card), Revision 1.26.0.0 with the (somehow important) Fcode part missing.

So, if he doesn't bother and has a tool to dump his firmware, maybe he can post it. So i could try to flash mine to get it booting too.

At a closer look, the card seems to have two EEPROMS on it. So it might even be some kind of "dual bios".

Edit: Further Google research led me to this side, where some AIX-firmware for LSI SAS controllers can be found, which has the Fcode 1.00.49 part in it. Also i have SASFlash for DOS right now. So i should be basically ready to flash this thing "for mac" myself.

 

[Tratkazir_the_1st]

@flyproductions
I can try to dig in my downloads & sent to you something about fcode ROM, if you wish. Not guaranteed it would be what are you talking about. (You can try to dig in Broadcom support site for firmwares but this is a bit boring "pleasure" as for me ).

 

[flyproductions]

@flyproductions
I can try to dig in my downloads & sent to you something about fcode ROM, if you wish.

Thanks! But i have some other idea, if you still have the card and a bit of experience in DOS-flashing stuff.

Attached you find an archive containing the DOS flasher "sasflash" as well as it's manual. But you do not need to read that. Cause all you need to know to dump your cards firmware is very simple. It's just one command:

sasflash -ufirmware (filename).fw

About 300k of free space are needed on the disk. Would be very kind. I found some firmware, containing the Fcode-part. But the rest just doesn't seem to match. At least it refused to flash to my card.

 

[flyproductions]

@flyproductions
I can try to dig in my downloads & sent to you something about fcode ROM, if you wish.

Kind of partial success! I was able to extract the Fcode-part from the file, i downloaded on this IBM-site and successfully flash it to the card. So now i get the same as you under the SAS-tab in System Info. Even with the same Fcode-version 1.00.49. But it still doesn't boot. Otherwise it's still working. So the Fcode itself just seems to be for another card.

But at least i now know that it is not the firmware, but the BIOS-file, what needs to be flashed! So command to dump the needed data is

sasflash -ubios (filename).rom

Size should be something like 200k.

 

[flyproductions]

So, some progress (not really).

I know now that the Fcode is ok, cause
a) 1064E is explicitely mentioned and this is what i have
b) sasflash flashes it at all in the first place. This software is very restrictive to the things one tries to flash and explicitely states "adapter compatible"

But the dammned thing still does not boot!

So i dumped the bios just to see, that it has multiple images in it of which the Fcode was the second. So i put that in front (see attached file) and reflashed. Flashed fine again, but it still doesn't work.

So now i'm a bit out of ideas. Only thing i can imagine ist, that there is also something different in the firmware. Maybe some kind of "link" to the bios. So may well be both of the dumps from a working card are needed.

But, as i didn't find any working firmware image for the adapter at all, i stop for now. 😟