Error 0x7CF wrote on 2021-02-06, 23:17:

It doesn't have the same problem with just the M.2 to PCIe in a native PCIe slot in core 2 quad system does it? I'd figure that might be a really nice upgrade if the chipset supports PCIe, though definitely overkill for any core 2 quad system.

one day with a different motherboard I will use it like that but the board I am currently using is hiding the only 1x slot under the gpu at the moment haha. But I assume it would do around 120mb/sec on a 1x slot this particular 128mb m.2 isn’t the fastest one around.

I now have a 8-bit xt-ide and a 64mb cf I can do experiments with btw. But I don’t think anyone is expecting anything impressive from that 🤣.

SodaSuccubus wrote on 2021-02-07, 00:13:

I don't remember the disk performance itself feeling *too* terrible compared to a spinning drive of the time, but I have had multiple brand name CF cards get murdered by all the page file swapping and writing Win 9X does.

Like, astonishingly fast. Within a week of heavy gaming. I'd stick to using CF cards with DOS.

My plan for the page file is a gigabyte I ram on this pentium 1 motherboard. My friend hasn’t gotten it to me yet. I will add it to the OP when I get some experience with it. Though a good alternative would be a modern sata hard drive on these old computers the high end magnetic drives are actually fast enough for page file and doesn’t have the wearing out problem it can be along side your ssd on the add in card. Ssd would still probably be faster but the page file not wearing out the ssd is valuable.

Edited OP to include I Ram drive

Looky what I just picked up.

How fast should this one be?

Sphere478 wrote on 2021-02-07, 03:45:

Looky what I just picked up.

How fast should this one be?

My wild guess is maybe 10MB/second read speed , if you are lucky, probably less. That guess is based on the speed of the Sandisk CF cards that I used at the time and that were around that capacity.

If the thing was faster, they would have bragged about it.

I am actually curious, so please benchmark it if you can. Testing in CF USB 2.0 card reader should give a good ballpark figure .

Sphere478 wrote on 2021-02-07, 03:45:

Looky what I just picked up.

How fast should this one be?

The answer depends. There's a lot of variables--

Probably worth while to see if your system is configured optimally. There are many ways to do this, but this is the way I have been doing it.

1) Run speedsys, do the fast tests and save the report

2) Look in the report under "ATA/ATAPI Device Information"
You will see something like "Maximum Transfer Mode : PIO 4, DMA 2, UDMA 4 (UltraDMA/66)"
That's the list of transfer modes that your CF device says it supports

3) The line after that will read something like: "Selected DMA Transfer Mode : DMA 2"
That indicates what speed was negotiated between your device and your controller.
The negotiated speed is going to be the upper limit of your throughput and latency.
The speed negotiated is usually the fastest possible, but sometimes no.
If you are stuck at DMA 2 when your device wants to go faster, you need to use a cable or connector that grounds pin 28 (yea ole' 80 pin IDE)

4) In DOS, your IO speed depends on the real mode IO routines in your BIOS or BIOS extensions if you have any. It is not uncommon to find out that the BIOS on most motherboards does not support DMA transfer modes or multi word transfers without UDMA.SYS loaded.
Many third party EIDE controllers have their own BIOS. Controllers with BIOS extensions usually support the faster transfer rates
Windows 3.1 and later uses protected mode I/O, but up through Windows 98se, you still needed to manually go in and enable DMA access modes in the drive properties.

Here are some sample benchmarks that I did on a Barton 2500+ Althon / Nforce 2 chip set. The bios on that board supports UDMA transfer rates.
I did these benchmarks in DOS.
The same benchmarks come out about 10% faster under Windows 98se with DMA enabled.
In an AT device with properly coded firmware, the AT verify command should cause the device to read a sector without transmitting any data, showing a number close to the internal read speed. Some firmware seems to cheat and respond "good" without reading the sector.

So all that said, here are the results for an 7-8 year old WD silicon drive CF that I bought used on eBay. If you are doing better than that, you are doing OK. This drive is the base minimum of what you should see.

1                  Maximum Transfer Mode         : PIO 4, DMA 2
2                  Selected DMA Transfer Mode    : DMA 2
3                  Random access time     : 0.27 ms
4                  Linear verify speed    : 10250 KB/s  
5                  Min/Max verify speed   : 10232 KB/s / 10283 KB/s
6                  Linear read speed      : 5895 KB/s
7                  Min/Max read speed     : 5895 KB/s / 5895 KB/s
8                  Linear write speed     : 3654 KB/s
9                  Min/Max write speed    : 3525 KB/s / 3797 KB/s
10                  Buffered read speed    : 5947 KB/s

Here are the results from a 2GB DaneElec CF that I bought used on ebay. This drive is pretty close to the best I see from a CF.

1                  Maximum Transfer Mode         : PIO 4, DMA 2, UDMA 5 (ATA-100)
2                  Selected DMA Transfer Mode    : UDMA 5
3                  Random access time     : 0.42 ms
4                  Linear verify speed    : 483877 KB/s  (suspected cheating firmware)
5                  Min/Max verify speed   : 479833 KB/s / 487956 KB/s  
6                  Linear read speed      : 17749 KB/s
7                  Min/Max read speed     : 17695 KB/s / 17771 KB/s
8                  Linear write speed     : 13080 KB/s
9                  Min/Max write speed    : 11157 KB/s / 13274 KB/s
10                  Buffered read speed    : 18779 KB/s

Here are the results from a 2GB DaneElec FC that I bough new on eBay. I suspect that it is counterfeit or something else is horribly wrong with it. Firmware looks nothing like the other drive.

1                  Maximum Transfer Mode         : PIO 4, DMA 2, UDMA 6 (ATA-133)
2                  Selected DMA Transfer Mode    : UDMA 6                  
3                  Random access time     : 31.37 ms  (Yuck!)
4                  Linear verify speed    : 342088 KB/s  (suspected cheating firmware)
5                  Min/Max verify speed   : 342050 KB/s / 342125 KB/s  
6                  Linear read speed      : 1012 KB/s  (horrible considering the system that it is in)
7                  Min/Max read speed     : 1012 KB/s / 1012 KB/s
8                  Linear write speed     : 4560 KB/s
9                  Min/Max write speed    : 4545 KB/s / 4579 KB/s
10                  Buffered read speed    : 1547 KB/s  

Here is what I get from a $20 mSata device attached to the same computer with a PATA bridge. It's just hands down faster. Boots windows 98se in about 13 seconds, compared to about 20 for the best CF.

1                  Maximum Transfer Mode         : PIO 4, DMA 2, UDMA 6 (ATA-133)
2                  Selected DMA Transfer Mode    : UDMA 6
3                  Random access time     : 0.12 ms
4                  Linear verify speed    : 255234 KB/s  (This looks legit.  This drive likely can read & verify the data internally this fast)
5                  Min/Max verify speed   : 245037 KB/s / 256037 KB/s
6                  Linear read speed      : 74641 KB/s
7                  Min/Max read speed     : 74396 KB/s / 74723 KB/s
8                  Linear write speed     : 87308 KB/s
9                  Min/Max write speed    : 83390 KB/s / 87699 KB/s
10                  Buffered read speed    : 69470 KB/s

There's a lot more in this thread--

Small capacity SSD PATA/SATA benchmarks

Here's a picture of the 32gb msata device that does so well in the bench marks.

The chip on sled is a ubiquitous jmicron -JM20330 Sata/Pata bridge.

Seems like mSata is getting phased out by M2 , so theres some pretty good deals on that form factor these days.

Edit: fixed typo

Used mSATA drives are really nice because they're physically small enough to fit into the laptop ide adapters that have the same dimensions as the 2.5 hard drives for retro laptops that use caddies and storage-small enough that older OSes won't fall apart at the sight of their capacity. They're also usually quite cheap and quick too. For me, certainly a better option than CF cards. In some cases I even prefer them over the DOMs I've come to like.

I usually get em off the popular online auction website for 20something a pop for 32GBish drives. Presumably drives from those old HP Stream laptops with too little storage for any real OS.

Warlord wrote:

ya cf cards don't have dram cache.

So does SandForce controller and quite a lot of other PATA/SATA based SSDs.

The Serpent Rider wrote on 2021-02-10, 18:09:

Warlord wrote:

ya cf cards don't have dram cache.

So does SandForce controller and quite a lot of other PATA/SATA based SSDs.

On a retro pc you aren't likely to notice. All DRAM cache does is batch write instructions so that it can spread it around to each DRAM chip evenly and ensure more consistent performance. Without DRAM cache a large file copy can eventually slow down to HDD or even less than HDD speeds, but we are still talking about 40 or 50 mb/s typically.

It used to be possible to buy SSDs that essentially had no controllers. That was a real problem. They were basically big flash drives and would freeze and lockup all the time during random/reads and writes.

That is why SD and CF card solutions are not ideal for windows 98, IMO. Just the windows 98 page file can definitely thrash the flash storage without a controller to manage it.

CF cards use controllers, which usually shared with IDE DOMs of the same time period.

The Serpent Rider wrote on 2021-02-10, 18:09:

Warlord wrote:

ya cf cards don't have dram cache.

So does SandForce controller and quite a lot of other PATA/SATA based SSDs.

Speedsys reports that some of my CF's have a 1KB ram cache. Didn't have a strong correlation with performance though.

It also reports that the 40pin DOM's that I tested have a 1KB cache.

Looking at the bench marks, seems like there's some sort of buffer on all of them.

1DriveName         Cache   Form  BufferedRead   LinearRead
2HyperDisk 4GB      1 KB   DOM      31,251        30,488
3HyperDisk 8GB      1 KB   DOM      65,472        59,092
4DaneElec 2GB(bad)  1 KB   CF        1,546         1,012
5DaneElec 2GB(good) 1 KB   CF       18,779        17,749
6DaneElec 8GB       N/A    CF       25,723        17,560
7Lexar 1066x 16GB   1 KB   CF       19,410        18,614
8TOPRAM  2GB        N/A    CF       27,324        24,776
9TRANSCEND  2GB     1 KB   CF       37,315        33,190

Speedsys reports that some of my CF's have a 1KB ram cache. Didn't have a strong correlation with performance though.

Some controllers have internal cache, but I doubt it can be probed with common utilities, so reported 1 KB is just a dud.

I was planning on buying a few of these to try out. The previous comment about the ATAPI standards not matchign up with older pentium boards, should that lead me to believe these wont work?

https://www.ebay.ca/itm/PATA-IDE-TO-SATA-Conv … F8/233772943327

I'm running an Aptiva 2176 and I'd like to upgrade to SSD's just for the noise, and because I dont' know how long I can trust a 25 year old HDD. My bios will only support up to 8.4 GB per hard drive, and that's okay. I'm running 2 10's in there now with the extra space just urecognized. They run fine as 8 gb drives. I can find 16 gb SSD's online that I think would be perfect, even if I'm losing 50% of the storage, and I would like to put a newer Sata optical drive in the computer. Do you think any of this would be successful. I also didn't think about the page file issue. Is there any way around that in software or will Windows just constantly use up the life of the SSD ? I would probably be fine to just keep running the 2 HDD i have, I'm just tweaking for academic purposes.

Actual SATA SSDs can also work on anything i486 and higher. I'm not done with my research but it seems if you have a higher-powered hard disk controller - like my PTI-255W VLB controller in my 486 Desktop - you can use a SATA to PATA adapter with a modern SSD using a DDO and get full capacity, or you can even use a mSATA to 44pin adapter with a 44 to 40 pin adapter and utilize mSATA drives on an IDE system fairly easily. I have not figured out precisely why this works on my desktop but not other systems like my NEC Versa Laptops - which includes a Pentium 75 - but I think it has to do with the capabilities and open-compatibility of the chipset and circuitry. I think it may also have to do with the conversion circuitry. My desktops greatest achievement was 256GB mSATA on the PTI-255W using a mSATA to 44 pin enclosure attached to a 40 pin to 44 pin adapter. Between that and the DVD-RW drive I have in that machine, FreeDOS - with ALL features -t ook only 10 minutes to install, and about 5 seconds to boot on that system - the POST took longer to complete than booting did.

The Serpent Rider wrote on 2021-02-10, 18:27:

CF cards use controllers, which usually shared with IDE DOMs of the same time period.

At least in 2011, CF cards had pretty abysmal random read/write performance. Maybe new ones are better, but I can't see why they would need to be considering their intended purpose. The 4KB read/writes are what is important, as I remember it.

https://www.tomshardware.com/reviews/sdxc-sdh … -i,2940-10.html

I am not educated in CF cards, but I would imagine the controller doesn't do much more than manage the IDE bus.

At least in 2011, CF cards had pretty abysmal random read/write performance.

First of all, that's SD cards review. Secondly, that's totally in line with IDE DOM dated around 2012-2013 - Re: Small capacity SSD PATA/SATA benchmarks
Although SLC NAND provides somewhat better results with reading.

september0451 wrote on 2021-02-10, 19:37:

I was planning on buying a few of these to try out. The previous comment about the ATAPI standards not matchign up with older pentium boards, should that lead me to believe these wont work?

https://www.ebay.ca/itm/PATA-IDE-TO-SATA-Conv … F8/233772943327

I'm running an Aptiva 2176 and I'd like to upgrade to SSD's just for the noise, and because I dont' know how long I can trust a 25 year old HDD. My bios will only support up to 8.4 GB per hard drive, and that's okay. I'm running 2 10's in there now with the extra space just urecognized. They run fine as 8 gb drives. I can find 16 gb SSD's online that I think would be perfect, even if I'm losing 50% of the storage, and I would like to put a newer Sata optical drive in the computer. Do you think any of this would be successful. I also didn't think about the page file issue. Is there any way around that in software or will Windows just constantly use up the life of the SSD ? I would probably be fine to just keep running the 2 HDD i have, I'm just tweaking for academic purposes.

I personally would never bother with IDE to SATA adapters using undisclosed bridge chips . You may get lucky or you may not, but there are enough other variables that may cause issues, that "saving" 10$ to get a cheap "mystery" adapter is just not worth it, IMHO .

The Serpent Rider wrote on 2021-02-10, 21:28:

At least in 2011, CF cards had pretty abysmal random read/write performance.

First of all, that's SD cards review. Secondly, that's totally in line with IDE DOM dated around 2012-2013 - Re: Small capacity SSD PATA/SATA benchmarks
Although SLC NAND provides somewhat better results with reading.

Your right, but I also don't see why any CF card would perform better.

Low cost SSDs from 2012 and 2013 were terrible, but that is kind of the point I am trying to make. I owned a cheap SSD back in the day. It would lock up all the time and it wasn't fun.