SAS card for G5Quad

[V.Yakob]

Success complete now! 😃

Got it running at full speed and booting now!

I also decided to try using 3041E in my Dual Core G5.
Did I understand the procedure correctly?

I need to take the firmware from post #96

sasflash -ufirmware 1064_b3_fw.fw
sasflash -ubios 1064_b3_bios.rom

Do I understand correctly that you are doing this in order to erase the error from past firmware in post #50 that were not suitable for this board?

sasflash -o -e 5

What is fcode in post #106? 🤔 Do I need to use it after 1064_b3_bios.rom?

sasflash -ubios sas1064e_b3_fcode.rom

The last step is to turn on 3Gbps support according to the instructions in post #95?

 

[flyproductions]

I need to take the firmware from post #96

If your card works but just doesn't boot (most likely), you can leave the firmware untouched.

sasflash -ufirmware 1064_b3_fw.fw
sasflash -ubios 1064_b3_bios.rom

Sorry, but i think i gave some accidently wrong advice in this point. -ufirmware and -ubios are the commands to dump the respective files from the card.

The correct commands for flashing are -f for firmware and -b for the bios (which includes the fCode)

sasflash -o -e 5

This needs to be done to erase any bootservice not working in the G5. I only got my Quad booting with the card with no other boot service than fCode present.

What is fcode in post #106? 🤔 Do I need to use it after 1064_b3_bios.rom?

sasflash -ubios sas1064e_b3_fcode.rom

fCode is, what the G5 needs to recognize the card as bootable before the OS loads. Yes, this is the file to be flashed. Would suggest to truncate the name to fcode.rom for DOS-compatibility.

So, as mentioned above, command should be: sasflash -b fcode.rom

But first you shoul try, if the card doesn't just boot! There seem to be some of them allready haveing the fCode-bios on them.

The last step is to turn on 3Gbps support according to the instructions in post #95?

As all three parts, firmware, bios and configuration are in three different non volatile storage areas, you can do the steps in any order you like.

 

[V.Yakob]

It works! I have collected everything in one archive to make it easier for other forum members to cope with this task.

Receive information about the installed card
sasflash -list

Upgrade the Firmware and BIOS
sasflash -o -f 3041ERB3.fw -b mptsas.rom

Install Fcode
sasflash -o -b fcode.rom

 

[Tratkazir_the_1st]

@V.Yakob
I suppose you can combine last 2 commands together, -b mptsas.rom -b fcode.rom. If this wouldn't work - 1 by 1 .

 

[V.Yakob]

Yes, this is not Photoshop! 😁

Yes, I got similar results in my tests.
Used 2 x Netac 128 Gb + deployed OS.


I also tested a partition of two SSDs separately OS.

RAID0 (Striped)


RAID1 (Mirror)


Power 💪🏻 Mac!

P.S. probably, if I add more disks to the array, the speed will increase. I'll check it later.

 

[flyproductions]

P.S. probably, if I add more disks to the array, the speed will increase. I'll check it later.

Not sure. I'm afraid, you'll run into the limitations of 4-lane PCIe 1.0.

 

[Amethyst1]

P.S. probably, if I add more disks to the array, the speed will increase. I'll check it later.

Not sure. I'm afraid, you'll run into the limitations of 4-lane PCIe 1.0.

As has been shown, 800 MB/s is the limit.

 

[V.Yakob]

Not sure. I'm afraid, you'll run into the limitations of 4-lane PCIe 1.0.

3 SSD Netac -- 256 + 128 + 128




It is reported here that the PCI-Express 1.0 4x bus limit is 1 GB\s.
Minus different errors, Netac is not the fastest drives, 3041E board errors, I think 800-850 MB\s can be obtained.

 

[flyproductions]

As has been shown, 800 MB/s is the limit.

Ahhh, sorry, sure, my bad! 🤪

As long as the AHCI-M.2 blade reaches up to 800, sitting on a four-lane-adapter, a striped array of SSDs sureley should too!

 

[flyproductions]

It is reported here that the PCI-Express 1.0 4x bus limit is 1 GB\s.
Minus different errors, Netac is not the fastest drives, 3041E board errors, I think 800-850 MB\s can be obtained.

Yes, this is "reported" here too! 😉

A 256 GB AHCI M.2 SSD reaches over 800 MB/s with a 4-lane-card. If not limited by the bus, this SSD can max out at about 1.200 - 1.300.

 

[tfelder]

PCI to PCIe Bridge in PowerMac G4 MDD​

Just to let you know that yes, the 3041E SAS card works on a PCI bridge adapter. However, transfer speed are not much better than on an IDE-SATA adapter or with the SIL3112 SATA cards ~50-110MB with a RAID0 array consisting of 2xSATA3 SSDs.

Note: this is not the magical-pixie-dust unobtainium StarTech PCI-X to 4x PCIe Adapter but the cheap single-PCIe lane 1x adapter. Someone still needs to try this with the StarTech.

 

[socom_22]

So might easily be one of the fastest storages ever to boot a Powermac into Mac OSX

1.000 MB/s, bootable in a Quad G5 should be the mark to hit!

It is very possible to boot a Power Mac G5 Quad from a storage system capable of 1,000+ MB/s. In fact, I have already done it and am posting from such a system right now!

This is using Fibre Channel. I have never been able to achieve transfer speeds like this with SAS (and still be bootable) on a Quad.

 

[LightBulbFun]

thats awesome! I never realised a Fibre Channel card would be bootable!

 

[flyproductions]

It is very possible to boot a Power Mac G5 Quad from a storage system capable of 1,000+ MB/s. In fact, I have already done it and am posting from such a system right now!

Very nice! 👍

This is using Fibre Channel. I have never been able to achieve transfer speeds like this with SAS (and still be bootable) on a Quad.

Can you give some details to your setup? Which card do you use? Sure it has to be some raid-configuration as even the M.2 SATA SSD only reaches about 850 and NVMe-drives just don't run in Powermacs.

 

[ervus]

Isn't this NVMe?

https://forums.macrumors.com/threads/post-your-ssd-cf-sata-pata-powerpc-benchmark-results.2063361/post-31743970

 

[V.Yakob]

Isn't this NVMe?

https://forums.macrumors.com/threads/post-your-ssd-cf-sata-pata-powerpc-benchmark-results.2063361/post-31743970

Booting Mac OS X in this design doesn't work, right? It's not interesting. 😅

 

[flyproductions]

Isn't this NVMe?

No. It's M.2 SATA. The exact same blade which i have in my Quad. R/W around 850. Seen by the system as "generic AHCI". But unable to boot.

 

[flyproductions]

Booting Mac OS X in this design doesn't work, right?

As far as it is AHCI, the answer is no.

 

[socom_22]

thats awesome! I never realised a Fibre Channel card would be bootable!

It's one of the coolest "secret" capabilities of PPC Macs in my opinion!

Very nice! 👍


Can you give some details to your setup? Which card do you use? Sure it has to be some raid-configuration as even the M.2 SATA SSD only reaches about 850 and NVMe-drives just don't run in Powermacs.

My results unfortunately were not straightforward at all to achieve. During the pandemic I spent a month or two to reverse-engineer a majority of the Apple/LSI Logic fibre channel card. I am currently using 2 of them which I heavily modified with the goal of maximizing the bandwidth of the available PCIe lanes in the Quad.

Benchmark results are one thing. Real-world transfer speeds are usually much less once CPU/file system overhead/losses are taken into account. So far, my Quad setup has proven to be capable of delivering (reliable) real-world sustained file transfers at up to 1.01 GB/s in Finder. All while booted from the actual storage itself.

30 seconds later, very consistent and minimal fluctuation between 950 - 1035 MB/s! The reason the read speed is shown to be less than the write speed is because that extra ~200 MB/s is being transferred from network shares via the 2x Gigabit Ethernet.

I just wanted to share my results and to show what practical copy speeds are possible with the Quad. I am sure it can be pushed even further than this.

I'm not going to get into the details about exactly what I did just yet because quite honestly, the process is not even reproducible yet. I decided I wanted to create an identical pair of cards in case the first two stop working, however the second pair I made did not work at all like the first pair did. There appears to be subtle variations in either the board design itself or the FCode expansion ROM (or both) which makes this whole setup so much more inconsistent to re-create.

I gave up on it over a year ago because my current setup works perfectly, but may revisit it again someday to figure that all out. If I'm able to do that, I'll definitely create a dedicated forum post on here to explain how it all was done.

Until then, my focus right now is on PPC G4 upgrades using 7448 and 7457 chips. Keep an eye out on the chip swap thread for updates on that. I have a couple of pretty neat PPC projects I've been working on that I'll be posting about soon!

 

[flyproductions]

I am currently using 2 of them which I heavily modified with the goal of maximizing the bandwidth of the available PCIe lanes in the Quad.

Great job!

...but, if its only for the PCIe-bandwidth, shouldn't this be relatively easy possible, using two of the SAS-cards with two SATA SSDs connected to each of them and then running these four drives in Raid 0?

Also would be interesting to know, which kind of media u used. I don't know anything about fibre channel. Do they run standard SSDs or are there some special drives?

 

[flyproductions]

Until then, my focus right now is on PPC G4 upgrades using 7448 and 7457 chips.

Ahh, interesting too! 👍

I have a G4 Cube, running some Dual 7447a @ 1.6 GHz. Would really like to swap for two 7448s @ 2 or 2.1 ...but am afraid, neither getting the parts nor swapping them to the card is exactly what can be called an easy task. 🤔

 

[socom_22]

...but, if its only for the PCIe-bandwidth, shouldn't this be relatively easy possible, using two of the SAS-cards with two SATA SSDs connected to each of them and then running these four drives in Raid 0?

That was the goal when modifying the cards, yes. Not necessarily the overall goal for the system. If the only thing you're looking to do is to create the fastest possible boot disk, then you're absolutely right:

I would take 3 of those flashed SAS cards, populate the two x4 PCIe slots and the one x8 slot. Then connect 4 SATA III SSDs to each controller for a total of 12. Then use Disk Utility to RAID0 the 12 SSDs together. The end result should be one of the fastest possible boot disks in a Quad (assuming all that works properly).

The problem with this approach is that it isn't very practical. It leaves no available PCIe slots for equally fast storage to transfer to. As I showed in my previous post, I want to be able to copy files at over 1 GB/s (after overhead). This means you actually need 2 storage systems of roughly equal bandwidth capability (and at least one of them bootable).

I may experiment with these LSI SAS cards a bit with my own setup to see how they perform.

 

[flyproductions]

That was the goal when modifying the cards, yes. Not necessarily the overall goal for the system. If the only thing you're looking to do is to create the fastest possible boot disk, then you're absolutely right:

Yes, this was my goal. But in the end it's all more like a feasibility study. In real world the raid's disadvantages in case of random read/write transactions of small portions of data (which is the most common usecase on a system drive) vs. a single drive make using any raid configuration as a boot drive kind of pointless.

I would take 3 of those flashed SAS cards, populate the two x4 PCIe slots and the one x8 slot.

Not possible in my configuration as the second slot is blocked by the GPU. And the remaining one is used by the M.2 SATA drive, which delivers about 800 R/W with much higher random transaction performance compared to the raid...but is sadly not bootable.

 

[LightBulbFun]

one way to maybe work around the PCIe limits is to use a PCIe switch card,

I have seen this done with MacPro5,1's to get Gen 3 drives running at full speed, no reason the same wont work in a PCIe G5

"just" get a Switched PCIe x16 to 4xM.2 card, the PCIe switch will take the Gen 1 x16 PCIe slot and pump it into a full speed speed PCIe x4 Gen 3 slot

and also I *think* all 4 slots on the card should be able to talk to each other at full speed, even if the host slot is not full speed, although I am not sure if its something I have seen tested

so even if its one of the x8 or x4 slots, the AHCI or SAS or Fibre channel cards that are installed on the PCIe switch card should be able to talk to each other at full Gen 2 or Gen 3 speed

would be quite fun to see someone do something like that

 

[flyproductions]

I have seen this done with MacPro5,1's to get Gen 3 drives running at full speed, no reason the same wont work in a PCIe G5

Yes! This is, what two 970 EVO plus in Raid Zero come up with in my 5,1:

Nice speeds! But note the massive drop in random write transactions, falling far behind even a single 980 PRO running on the same card.

would be quite fun to see someone do something like that

Sure, i could swap the switch card to the G5. And i'm somehow sure, the SAMSUNG AHCI SATA blade would run to it's full potential of about 1200 MB/s instead of 800 on the cheap adapter. ...staying not bootable of course. But i just don't feel like ripping everything apart just for the sake of science at the moment. 😉