Pentium 1? The CPU will hold things back, you don't have to worry about storage performance 🤣

I've had lots DMA issues when I reviewed ally my Socket 7 and Super 7 boards. Some hard drives would allow enabling DMA, others didn't. So try a few. CF cards allowed enabling DMA most often I found. SATA to IDE controllers was also a mixed bag. The top board for storage is the GA-5AX, with a ATA/66 controller and working DMA.

philscomputerlab wrote:

ATA 66/100 sounds great, but on a Pentium III I don't see the point. Maybe for cloning disks, or copying stuff, but in terms of loading Windows or games, I doubt it does anything.

I used to have a few 440BX P3 machines in the family, and I saw Win2k boot noticeably faster when using an ATA66 or ATA100 controller. However, it's so long ago that I can't remember which, if any of these, were well controlled comparisons or if they all had multiple variables involved.
Anyway, it's not very important unless you like to boot Windows a lot. Which with Win98, might actually be a consideration I guess. 😀

But what a modern HDD does bring to the table is brutally fast access time and small file performance.

I agree that's true in final effect, but in raw terms (ignoring the density), I feel compelled to say I think modern drives may generally seek a bit more slowly than they did around 10-15 years ago. When you factor in the short stroking effect though, they still end up being faster.
I experimented once with short stroking a drive, and the benefit definitely leveled off when taken to an extreme. So if you short stroke a modern drive to match the full capacity of a 10 year old model, it will be dramatically faster. But if you test both those drives in a relatively tiny disk range like 8GB, there might be very little difference between them at that level.
I think the head seeking on modern drives has been nerfed a bit because people complain about head chattering nowadays. Judging from Amazon and NewEgg reviews, the few remaining performance oriented drives get a lot of RMAs from customers who think the seek noises are a defect. Most drives are kept slow enough to be inaudible nowadays.

When I was working with Slot 1 and also S370 systems, the S370 has double the IDE performance, could not "feel" it. And yes, when I work on a project, re-staring Windows happens over and over and over. 🤣 Here the solution I use works really well. Very fast boot times, held back by the processor (the faster the CPU, the faster the machine boots).

I was hoping the SIL3512 you were investigating (if I recall correctly) would solve the DMA prob - oh well.

Another possibility I suppose would be a late model PCI SCSI card with a SCSI to SATA bridge. Or SCSI-IDE if you have an acceptable drive around there.

On my p3 rig I currently use an Adaptec U160 + a SCSI-IDE adapter for boot drives (separate drive for win98, XP) with a SIL3112 for 2x 1.5TB WD Green archive drives (mostly disc images). Having all these drives adds up power and noise wise - so silencing the boot drives by making them SSD would be a big win.

SCSI to SATA? Cool, didn't even know that 😀

Don't like SSD's as they are a real pain to recreate lost data from.
And I don't like drive solutions that would potentially render the rest of the machine a giant bottleneck.

For CF cards. It seems like they are not the best for all machines.
Real platter drives are the most versatile and most compatible solution.

For real harddrives, I tend to go for ATA-100/ATA-133 controllers.
Brand of drives that I like the best, are Seagate PATA that I mod capacity size, using sea tools.
SATA drives are pampered with a SATA to PATA converter.
The type of converter that clips onto the hard drive it self and not on the MOBO or a PCI card.

On my 98 rigs I do like Phil and use modern SATA with an IDE to SATA adaptor. It allows DMA, and pretty much saturates the motherboards onboard IDE interface. When it comes to 98 the difference between ATA 33 through 133 never really comes into play when using a modern HDD as the onboard cache and vastly superior access time seems to render it moot. The trouble of using a PCI card for a faster interface than what onboard is far more trouble than its worth.

On my pure DOS rigs I use DOM's. Never had a single issue with them. While I used to adore PC-DOS 7/2000, This past year I have switched to MS-DOS 7.1 and find that a partitioned 8GB DOM (two 4GB partitions), is perfect for my builds. You can get them in the common female connector and stick them straight onto the mobo's IDE socket farily cheap, or pay a bit more and get the male version and run two of them on a single IDE cable. I have yet to find a reason to need more than 8GB in a DOS rig, so I just use one directly plugged into IDE 1. This leaves IDE 2 open for an IDE CD/DVD drive and a CF drive (I use the 3 1/2" front bay type as well). You never want to run a CD/DVD drive and HDD on the same channel if you can avoid it.

I would not bother with using a SSD or CF as a 98 drive as the swap file thrashing will kill them way to fast. But you can always use one if you put the swap file on an actual HDD. You will see pretty much maxed out performance and get decent life out of the SSD or CF. I've toyed with CF on 98 in the past and found that while it was super fast, the whole system would pause from time to time due to the slow write speeds and the swap file doing its thing. Reading was fantastic though.

A down and dirty method to speed up a 98 rig is using two good, modern SATA drives. The first one partitioned to 120GB (leave some headroom for the 128GB limit), and install the OS and whatever games, apps, etc you want on it. The second one with a 2 to 4 GB partition then a second partition of 116 to 118 GB for ISO, and other junk. Install one drive on IDE channel 1 and the other on channel 2. Put the swap file into the 2 to 4 GB partition and run as normal. You would be amazed at how fast 98 can be with a setup like that, even if its using an onboard ATA 33 controller.

Just tossing out some options that I've done and found to be very effective.

EDIT

A side bonus of keeping the swap file in its own partition is it REALLY keeps the fragmentation rate down by a massive amount. Defragging becomes something that rarely has to be done with this method.

Even MLC and TLC SSDs have been shown to be able to handle hundreds of terabytes of writes so I wouldn't worry about longevity in most cases, or having the swap file on them. I have a few drives around that have been running Vista and 7 desktops for years with swap and have < 10 TB write totals.

Just put SWAP along with ie and system temp files in the RAM (using a RAM disk) on Win9x systems and don't worry about fragmentation and risks of wearing anymore 😀