Ozzuneoj wrote on 2024-05-09, 19:16:

I'm not trying to derail the thread or diminish the value what is being discussed, but I think as long as SATA drives exist there won't be much of a reason to go with anything else for retro PCs. Even on modern systems there is hardly any perceptible difference between a decent SATA drive and an NVMe in average consumer workloads, despite the big difference in benchmark numbers.

Sata generally performs best for all PIO and UDMA retro computer uses, I agree, especially if you can enable block mode transfers. Seems like there is a gap for the ATA-2 / ATA-3 period, that is late socket 3 through early slot 1, when WDMA was the best transfer mode. Even though it's listed in the Jmicron and Marvell specs, I have not been able to make WDMA work over a PATA-SATA bridge, even though PIO and UDMA work fine. So there's a spot where legacy, CF, and SD storage can make a good case for itself.

This discussion also got me looking into Sata Express. Looks like an interesting effort that failed to get market penetration and made sata a legacy standard.

I'm still amazed that the linux boot loader was able to get the NVMe to be presented as ATA over PCI, even if it is extremely slow. I feel like there's still a piece to that puzzle that I don't understand.

I installed Windows 7 32 bits on NVME PCIE 3.0 x4 without cache on an Asrock 4Core1600P35WIFI+ motherboard by adding an oprom to the bios, the NVME appears as NVME IDE.
On PCIE 1.1 x4 the loss is very noticeable with a maximum read of 845MBs and a maximum write of 462MBs.
Still, it is far superior to the motherboard's SATA 2 controller with a SATA SSD.
I actually installed Win7 before on a SATA SSD and then installed the Microsoft Hotfix for NVME on Win7, after that the NVME SSD worked fine, so I cloned from the SATA SSD to the NVME and surprise surprise, it booted.
So it seems that each PCIE version is close to doubling the bandwidth of the previous one.

Edit
Instructions here: https://winraid.level1techs.com/t/experimenta … 32528/61?page=4

Hoping wrote on 2024-06-25, 15:09:

I installed Windows 7 32 bits on NVME PCIE 3.0 x4 without cache on an Asrock 4Core1600P35WIFI+ motherboard by adding an oprom to […]
Show full quote

I installed Windows 7 32 bits on NVME PCIE 3.0 x4 without cache on an Asrock 4Core1600P35WIFI+ motherboard by adding an oprom to the bios, the NVME appears as NVME IDE.
On PCIE 1.1 x4 the loss is very noticeable with a maximum read of 845MBs and a maximum write of 462MBs.
Still, it is far superior to the motherboard's SATA 2 controller with a SATA SSD.
I actually installed Win7 before on a SATA SSD and then installed the Microsoft Hotfix for NVME on Win7, after that the NVME SSD worked fine, so I cloned from the SATA SSD to the NVME and surprise surprise, it booted.
So it seems that each PCIE version is close to doubling the bandwidth of the previous one.

Edit
Instructions here: https://winraid.level1techs.com/t/experimenta … 32528/61?page=4

Thanks for sharing. The read throughput is significantly faster than a good quality SATA SSD on a decent PCIE SATA controller. Write performance is pretty close.

In terms of real-word performance,I suspect that two-lane SATA PCIE controller with an Asmedia, JMicron or Marvell controller with a Samsung 860 or 870 TLC based SATA SSD would be comparable and easier to set up. These SATA controllers typically have options ROMs that allow them to be bootable and have a fringe benefit of allow one to connect more than one drive.

On PCIE 2.0 or faster, I would definitely choose NVME for performance.

Hoping wrote on 2024-06-25, 15:09:

I installed Windows 7 32 bits on NVME PCIE 3.0 x4 without cache on an Asrock 4Core1600P35WIFI+ motherboard by adding an oprom to […]
Show full quote

I installed Windows 7 32 bits on NVME PCIE 3.0 x4 without cache on an Asrock 4Core1600P35WIFI+ motherboard by adding an oprom to the bios, the NVME appears as NVME IDE.
On PCIE 1.1 x4 the loss is very noticeable with a maximum read of 845MBs and a maximum write of 462MBs.
Still, it is far superior to the motherboard's SATA 2 controller with a SATA SSD.
I actually installed Win7 before on a SATA SSD and then installed the Microsoft Hotfix for NVME on Win7, after that the NVME SSD worked fine, so I cloned from the SATA SSD to the NVME and surprise surprise, it booted.
So it seems that each PCIE version is close to doubling the bandwidth of the previous one.

Edit
Instructions here: https://winraid.level1techs.com/t/experimenta … 32528/61?page=4

How far back can you take this? Can you port this process back to a Pre PCI-e computer using a PCI-PCIe adapter?

I have not created or modified anything, the credit goes to the Winraid forum user who modified the option ROM and shared the method to integrate it into the BIOS.
This system has the problem that it links the NVME that is used with its DEVID to the motherboard BIOS, I imagine that if the NVME fails to change it you have to modify the BIOS again. I kept a chip with the original BIOS inside the case of that computer, just in case.
Well, I don't have one of those PCIE to PCI adapters, but I don't see why it shouldn't work, the process they show in the Winraid forum seems to give good results with AMI BIOSes in general and some success with Award BIOSes. I also think I've read of someone using a PCI card to boot the NVME option ROM. I think that method is similar to XTIDE and would be the one I would most like to try, but I don't have a PCI card to try it on.
It also looks like you can even install Windows XP on an NVME. I think there are drivers for NVME and XP made by the community.
Note that I used a simple and cheap NVME with no cache, I guess a better unit might give better results.
I think I had slower write speeds before because this NVME gets overheated very fast, I glued a big heatsink to it, but it still gets to fairly high temperatures, and today I did the test just after powering on the computer, so it was cold.