As I mentioned earlier, I recently upgraded my Pentium II with Windows 98 to run on an SSD. The initial post was written the night of the upgrade, which was rather late. This post goes into more detail, including benchmarks.
The upgrade took it from running a Maxtor model 91360D8, with a capacity of a approximately 13.6 GB, to a SanDisk SSD Plus with 120 GB of capacity. The Maxtor was still functioning well, albeit being loud by today's standards. The SanDisk was new from Micro Center.
In order to connect up the SATA drive to the IDE connectors on the motherboard, I needed either an adapter, or a PCI SATA card. I went with the former, first trying the Kingwin SATA to IDE Bridge Board, Model ADP-06. Unfortunately, this proved problematic. The main issue was that given how the IDE and power ports were connected, it was not possible either to get the SSD properly secured in the 2.5-inch to 3.5-inch adapter, nor to fit the IDE-> SATA adapter in the case. The adapter, when connected to both the SSD and the IDE/power cables, stuck out too far from the vertical hard drive enclosure to put the left side of the case on. Being unable to properly connect the SSD to the 2.5-to-3.5 inch adapter was also problematic; tape did not perform as well as hoped, and ultimately it was sufficiently difficult to get everything connected that I gave up on this adapter.
However, I was able to verify that the adapter did function properly via connecting it up to my Kingwin external SATA/2.5-inch-IDE/3.5-inch-IDE USB drive connector. So the problem was not the electronics of the board, but rather the dimensions. If I'd had a full ATX tower, I may have been able to get it to work, but with my small 440 BX system, it just wasn't happening.
So after doing a bit of research, I ordered Marvell IDE-SATA07 adapter, using the Marvell 88SA8052 chip. From what
I'd read, this adapter had an excellent compatibility record, and it also had much smaller dimensions that would work well with my case. The downside is it's not very common, but I was able to track one down online.
This adapter did work with my case, but it also had some caveats. The first is that it does not have a plastic "fence" around the IDE pins, and as it turns out, at least with my IDE cables, it was very difficult to remove it from the IDE connector. So much so that when I did while testing, it bent a few of the pins on one end. The good news is, I was able to bend them back enough to continue working - but if you use this adapter and have to remove it from an IDE connector, be extra careful while doing so.
The second was that the location of the missing pin on the IDE connector was on the opposite side from where it was on my Maxtor hard drive. Given other cable constraints, reversing this was not feasible; as such I had to rotate the IDE cable 180 degrees. This was a tight squeeze in the cramped case, but was feasible.
I also ran into some old-hardware issues. One of the power connectors I attempted to use appears to no longer work; I was able to confirm another was via the Maxtor, and am now using that one.
I used Macrium Reflect Free to clone the HDD to the SSD, and was thus able to successfully move my 17-year-old install of Windows 98 to the SSD, as well as getting a virtual machine image and keeping it on the original drive. With a bit of attention to the help documentation in Reflect, the Kingwin external adapter, and a traditional 2.5-inch enclosure for the SSD, this was surpsingly simple to pull off from my main desktop, and was pretty quick via USB 3.0. The Maxtor may be no speed champion, but it's not that shabby when the reads are 100% sequential.
All in all, it was nowhere near as simple as swapping in an SSD in laptop. However, once I had an adapter that fit the case, it was not too bad, and I suspect it would be quicker the second time were I to upgrade another Pentium II.
And now for the exciting part - benchmarks! As mentioned in the first article, quite a few programs now start much more quickly, including Windows. Time is not especially precise as almost all the benchmarks were taken via the timer on my Symbian smartphone rather than benching within Windows 98, but the overall trend is very clear.
The software chosen was based on what was already on the HDD and didn't require a CD.
Windows, boot to Netscape loaded: 148 seconds (HYDD) to 104 seconds (SSD)
Opera 10.60: About 7.1 seconds with both HDD and SSD; no noticeable improvement
A custom Java program I wrote (using programmatic startup time measurements): 6.2 seconds (HDD) to 3.8 seconds (SSD)
Excel 2000: 2.2 seconds (HDD) to less than 0.6 seconds (SSD)
Word 2000: 1.1 seconds (HDD) to less than 0.6 seconds (SSD)
VLC: 5 seconds (HDD) to essentially instant (SSD)
IE6: 4 seconds (HDD) to 1.8 seconds (SSD)
Visio: 4-5 seconds (HDD) to 0.9 seconds (SSD)
Microsoft PictureIT: 2.5 seconds (HDD) to 1.5 seconds (SSD)
Media Player Classic: 5 seconds (HDD) to 1 second (SSD)
Adobe Acrobat 4: 3 seconds (HDD) to 1.5 seconds (SSD)
Sim Tower: 5 seconds (HDD) to 4.6 seconds (SSD)
Everest Home Edition: 6.5 seconds (HDD) to 6.6 seconds (SSD)
Corel Print Studio: 4 seconds (HDD) to < 2 seconds (SSD)
Punch Super Home Suite: 7.5 seconds (HDD) to 5.4 seconds (SSD)
7-Zip: < 1.5 seconds (HDD) to 1.2 seconds (SSD)
Merriam-Webster Dictionary: 5 seconds (HDD) to 1.5 seconds (SSD)
Chessmaster 4000: 2.7 seconds (HDD) to 2.9 seconds (SSD)
Microsoft Works: 3 seconds (HDD) to 1.2 seconds (SSD)
Windows Media Player: 4.5 seconds (HDD) to < 1 second (SSD)
The average of the changes (without weighing for how long each one takes, since it's unlikely that you'd use this particular mix of software in a session) is that launching software now takes 44% less time with an SSD compared to the hard drive. And thats including the ones that didn't see any improvement or were marginally slower (likely due to background tasks or the imprecision of my timer).
And really... that's pretty fantastic. I knew the results would vary, as some software will be disk I/O intensive while loading, while other will not load much from disk, but will be CPU or memory-allocation intensive. But to see an average of 44% improvement when I'm still limited to a 450 MHz Pentium II, and also limited to the IDE interface and Queue Depth 1, really speaks well to the SSD's abilities. It's far from an ideal work environment for the SSD, and it still makes a very noticeable difference, both anecdotally and in benchmarks.
It is true that the Maxtor is not the strongest competitor. If I put my Toshiba 7200 RPM drive from my desktop in, what would it bench? It would be an interesting point of comparison; it likely could do a good amount better than the Maxtor. My suspicion is the SanDisk would still wind up on top by double-digits over the Toshiba due simply to random seek times, which would not have improved all that much between the Maxtor and the Toshiba; that's the area where SSDs really make their impact. I'd be curious to see the comparison, but not enough to take the time to bench a modern 7200 RPM with the Pentium II, as I was with the SSD.
It's also notable that these benchmarks are all with a 17-year-old install of Windows 98, which likely is rather horrifying to more than a few people. I do have the Windows 98 install disc, and could do a fresh install on the SanDisk; I would be curious how much faster that would be. However, I no longer have the install disks for a lot of the software on the current 17-year-old install, and that's a lot of the appeal of system i the first place. So while I could put a fresh install on for benchmarks, I wouldn't want to keep using the new install until I had at least most of Sim Tower, the Custom Entertainment Packs, Chessmaster 4000, etc. available to install. A few, such as Sim Ant and Sim Copter, I did make ISOs for, but not all. And given the iffy network card and old DVD drive (yes - DVD!) on the system, ISOs are not as easy as with new systems. I've found that the system will ready Taiyo Yuden CD-Rs without issue, but is much more miss than hit with other brands I've tried.
So for now, the system will move on with its 17-year-old Windows 98 install on SSD. Slower than with a fresh install, almost certainly. But still, much faster than with its old HDD, and that was the goal of the project.
Return to Blog Index