That means, if I would make a "9" driver or pre-Leopard driver for NVMe I would need to restrict every transfer to 128K or even lower - and after that I would need to double-buffer everything into a physically contingous well-aligned internal buffer. That would slow down the entire transfer and any advantage of NVMe would disappear.
That's exactly what I thought ! ...... NOT, lol.
It's like you walk around thinking your all that and then SATAman, obviously a member of the remaining Annunaki, spouts out all this secret divine knowledge. My God, man, I always read your posts at least 3 times. Light bulbs going off... BAM
Under Leopard a non-bootable NVMe drive in a 2005 G5 with PCIe slots can achieve over 700 MB/Sec on reads and writes.
OK, that is just plain sick, I paid a kings ransom for my Mercury Accelsior (https://eshop.macsales.com/item/OWC/SSDPHW2R960/) about 4 years ago for my main Mac Pro and thought that at about 800MB/sec, nothing was even close, and you are saying in a 2005 G5... 700 MB/Sec, what sorcery is this ? I appreciate all the cool information you have posted !
Lastly, for those on a budget, even though we have highlighted many storage technologies here, I was wondering if you can give us the "Sataman" guidelines of what you consider the two best bang for the bucks as far as a bootable OS 9 SSD solutions... one using the native PATA, and one as using a PCI card.
For the native PATA I think the best is the SATA <--> PATA converter, but only few are good. There was a topic earlier about it.
Also, there is a very bad batch of Kingston drives on eBay, I got a few from a seller from Slovenia, they kind-of work - but have nothing to do with genuine Kingston. This is a known problem, use Samsung (pr other known brand) instead. For older machines like Sawtooth 120GB drives are good enough.
The good SATA <--> PATA converters can handle master-slave as well, so you can use two.
For PCI-X card I think the current best is the 1V4, which is going to be de-throned by the updated 3124, with "9" support.
Some other good future candidates are the Marvell 6042-based boards (have 7042-based driver which was sold commercially to quite a few customers).
The "9" and FCode could happen.
Than the "Frodo" (Braodcom RAIDCORDE) which looks very simple and is very fast, sold for $20 and up and the chip is inside of some G5-s as I know, with open source "X" driver written by Larry Barras at the time he was at Apple.
I have two combo SATA/PATA cards with BIOS chip from VIA, it's a simple PCI card - but it has 3 SATA ports and one PATA.
It has a quite well-written documentation and I think, I can reuse much of the recent written code for 680-3112-3114 and hopefully it has no sleep problem with the Quicksilver.
The 680-3112-3114 code is largely done, at least the PowerPC part. The Intel part is important for the testing purpose, working on it but not today, today is SAS day.
The performance of 680-3112-3114 is clearly not up-to-date, is very far from 3124 - but that's the fastest I can do.
Once it is done I will go back to 3124 because it is clearly a better choice. All I miss is the "9" driver which can be done based on the new "X".
It could happen, the first approach of the "9" driver for 3124 won't have port multiplier support.
Sorry, the entire virus situation was a huge 3-moth setback for everything.