When you're using modern storage media you run into overhead and other bottlenecks. It's well known that IDE (especially if you can use VLB) is quicker if you're using a modern device. To get a better picture you should re-run those benchmarks using period correct harddrives, then you'd probably see a difference in favour of SCSI.

Marco wrote on 2024-03-27, 21:25:

Comments are really very welcome

Have you tried to burn a CD on that 386SX?
I could burn with 2x speed without problems, however, my 20 MHz clocked system was just a tiny bit too slow for 4x.
Adaptec 1542B, PlexWriter SCSI

No I don’t have a scsi writer. Sorry.

@Vetz:
I hope I understand. I thought scsi is indeed known for cpu offloading independent from the hdd connected. Ide stays ide with all pros and cons regarding I/O and cpu usage on non udma systems regardless of the drive connected.
Same for scsi. I was then on the wrong track. At least I have confidence now 😀

BlueSCSI v2 is good for ~8MB/s, Adaptec 1542CF is also much faster than 1-2MB/s seen here. Something tells me Windows Adaptec driver is not utilizing DMA (ISA DMA is very problematic with >16MB of ram), or a very slow SD card was used?

SD: I use a SanDisk Extreme PRO SDXC UHS-I. Up to date one. Promising up to 200mb. Sequential write on my standard Ryzen PC is very good on that card.
RAM: I use 16MB
Win driver: win95r2 default. No further settings there to be adjusted. Just parameters I can add manually under driver config. Pls also note that worse performance under dos which is not win driver related.

Anything very keen to find possible config gaps 😀

Marco wrote on 2024-03-28, 10:27:

SD: I use a SanDisk Extreme PRO SDXC UHS-I. Up to date one. Promising up to 200mb. Sequential write on my standard Ryzen PC is very good on that card.

Sequential write is important for recording streaming data (videos mostly), and useful on devices that have gigabytes (or more) of RAM to buffer writes during any hiccups . It's useless metric for reading, especially for random access which might actually be slower due to various write optimizations.
For such use cases you want a card with low latency, both average and worst case, and it's only recently that we got SD cards promising that kind of performance. These are marked A1 and A2, though A1 should be enough in this case I think.

That being said for slow enough CPU even DMA is not going to help because it's actually preventing it from accessing RAM. That's still performance improvement over PIO pooling but not visible as lower CPU usage. To actually see benefits you'd need a cached system - either a 486 CPU or a decent 386 mobo with chipset that doesn't HOLD the CPU if the data it wants is in the on-board cache. Both will still stall if RAM access is required obviously.

For a very fast 486 or Pentium, or better CPU, you want PCI to bus-master the transfer. ISA and slow DMA engine will be slower than PIO (although there at least you will see lower CPU usage when using DMA).

Why don't you use a 15k scsi hard disk? i think the result will be different

Thanks again.
I might not have been clear: I only tested seq write on the current PC. Pls also remember that I did try to simulate seq writes when copying large files under both OS. I am very open for all suggestions but somehow sure the SD card isn’t the bottleneck. But I will also investigate here. I tested some weeks ago a std much older CF card and results where much better especially on Access time than with
My current scsi solution. Hm

AlessandroB wrote on 2024-03-28, 11:14:

Why don't you use a 15k scsi hard disk? i think the result will be different

Thanks. I dont want these high sounding spinning drives to be honest. I would also need an (active) adapter chain to get back to scsi 50pin.

Again: my expectation is/was that a flash card solution will more than compete with high performance drives as I could proof with my IDE2CF adapter compared to my maxed out maxtor ide hdd.

But 35ms access time in win95 … that’s just ridiculous and I still can’t believe all is as it should be. Or maybe it’s the complex scsi adapter which includes much more logic than a ide2cf.

8-bit SCSI is slow and way overrated.

Marco wrote on 2024-03-28, 11:39:

Thanks. I dont want these high sounding spinning drives to be honest. I would also need an (active) adapter chain to get back to […]
Show full quote

AlessandroB wrote on 2024-03-28, 11:14:

Why don't you use a 15k scsi hard disk? i think the result will be different

Thanks. I dont want these high sounding spinning drives to be honest. I would also need an (active) adapter chain to get back to scsi 50pin.

Again: my expectation is/was that a flash card solution will more than compete with high performance drives as I could proof with my IDE2CF adapter compared to my maxed out maxtor ide hdd.

But 35ms access time in win95 … that’s just ridiculous and I still can’t believe all is as it should be. Or maybe it’s the complex scsi adapter which includes much more logic than a ide2cf.

15k drive are absolute silent, like modern ide drive, no worry about that. You not need an active adapter, just mechanical.

Imho passive ones are ok for 68 to 50 but not from 80 to 50.

They always look like that:
https://www.ebay.de/itm/125579176184?mkcid=16 … emis&media=COPY

Edit: this might be due to the integrated power supply of sca2 connections

@Marco , do you have a PCI SCSI adapter ? Testing with that on a newer machine would provide a baseline (and sanity check the Bluescsi v2+ SD card combo).

Marco wrote on 2024-03-28, 14:49:

Imho passive ones are ok for 68 to 50 but not from 80 to 50. […]
Show full quote

Imho passive ones are ok for 68 to 50 but not from 80 to 50.

They always look like that:
https://www.ebay.de/itm/125579176184?mkcid=16 … emis&media=COPY

Edit: this might be due to the integrated power supply of sca2 connections

The only difference between 68 & 80 is that the 80 has power & ID selection in the connector integrated. The rest of the signals are identical.

Marco wrote on 2024-03-28, 14:49:

Imho passive ones are ok for 68 to 50 but not from 80 to 50. […]
Show full quote

Imho passive ones are ok for 68 to 50 but not from 80 to 50.

They always look like that:
https://www.ebay.de/itm/125579176184?mkcid=16 … emis&media=COPY

Edit: this might be due to the integrated power supply of sca2 connections

integrated power supply does not mean "active" connect a 15k to adaptec 1542 is easy and chep just signal routing using adapter.

Agree to you. I learned that now.
Unfortunately no other pci based scsi in place to test performance. Maybe I will go for a 15k drive to test 😀

Or !

get yourself an acard ide cdrom to scsi adapter, a 68 to 50 with high byte termination converter, and firmware mod the acard, then you can test both ide & scsi with the same hard drive !

weedeewee wrote on 2024-03-28, 17:21:

Or !

get yourself an acard ide cdrom to scsi adapter, a 68 to 50 with high byte termination converter, and firmware mod the acard, then you can test both ide & scsi with the same hard drive !

Or order a 20$ Adaptec SCSI card keep things just a bit simpler. 😉

Seriously, that idea is interesting, but the the added complexity actually might introduce variables impacting performance in unpredictable ways.

IMHO, unless OP want to specifically test with the same HDD (his choice, of course), I believe it would be more interesting to first try to get a best case scenario for both IDE and SCSI performance an HDD isn't really necessary for that.

At the end of the day if a 15K drive is either faster or not than the Bluescsi v2 + SD card is useful datapoint.
Whether Bluescsi v2 + SD card is actual faster or not on a PCI SCSI card is another useful datapoint.

Some results from an ISA only (probably 8MHz ISA) Pentium machine a 1542CF and a 10K drive.

Will this work? Adaptec 1542CF and Cheetah 10k with SCA-2 80-pin hot-swap connector