Marco wrote on 2024-03-04, 21:04:

Hi all, […]
Show full quote

Hi all,

I just finished the benchmark set for my new ide DOM transcend 8gb on my 386. Its white paper states up to 50-70mb/s.

Of course access times are close to 0 but here’s the strange thing:

- random or seq read or writes won’t pass 2600kbit/s
- my 7200rpm ide maxtor reaches here always 3600kbit/s at least on seq.

I tried already changing ide ports, running solely etc but always same results. Tested with various progs. Both disks are Fat16 formatted.

Do you have any idea why this can be? Hopefully not that sector aligning thing but I cannot imagine that here.

Thanks a lot

2600kbs = 335 KB/s yes? That does seem pretty slow. You should probably get between 1300-1600 KB/s

What are you using to get past the 512MB limit? If you are using an option ROM, do you have the RAM shadowing turned on for the ROM's address?

p.s. What tool are you using to measure the speed?

Sorry, I relaced it by KByte/s. My fault.

Yes shadowing is enabled for my xtide rom. With shadowing disabled performance won’t pass 1300kbyte/s.

Tested programs:
- speedsys
- nu si8
- diskmeter
- 4_speed

Btw the 2600kbyte are also the maximum for buffered read speed on the DOM.

What is indeee much faster is the speedsys “linear verify speed”: 172000kbyte/s DOM vs 52000kbyte/s HDD. Whatever this number means as it is far beyond isa capabilities.

Marco wrote on 2024-03-04, 21:29:

Sorry, I relaced it by KByte/s. My fault. […]
Show full quote

Sorry, I relaced it by KByte/s. My fault.

Yes shadowing is enabled for my xtide rom. With shadowing disabled performance won’t pass 1300kbyte/s.

Tested programs:
- speedsys
- nu si8
- diskmeter
- 4_speed

Btw the 2600kbyte are also the maximum for buffered read speed on the DOM

This is running on an ISA bus, yes? How fast is it running? Default is 8MHz.

If your ISA bus is running at 8MHz and you have a 33 MHz 386, you are going to be doing great if you are getting over 2000 KB/s from your IDE drive.

https://forum.vcfed.org/index.php?threads/isa … 60/#post-692332

-- Edit, fixed MB to KB

It’s at 13,75 MHz

Marco wrote on 2024-03-04, 21:37:

It’s at 13,75 MHz

I usually expect get 1300KB/s -1600KB/s on an 8MHz bus, so 2600KB/s on a 13,75 MHz bus sounds about right to me.

And speedsys is reporting 3600KB/s on that same system with the Maxtor IDE? That's awesome. Can you post a screen shot?

Hi, will do tomorrow. Diskmeter and 4_speed are confirming the numbers.

If anyone has an idea about why the dom is slower I’d also appreciate hints.

The only point I would see is the hdd cache that’s not available on the DOM. Maybe that can make a difference. But I doubt that as well as the no-cache-performance of the DOM is also much higher anyway than 2600kbyte/s 😀

Perhaps the DOM just can read 512 bytes at once.
No multi-sector transfers.

I have a SD card based DOM that has this issue.
It is an ideal companion to my Tekram DC-680T VLB caching controller.

Ahaaa. That might be a point.

douglar wrote on 2024-03-04, 21:46:

Marco wrote on 2024-03-04, 21:37:

It’s at 13,75 MHz

I usually expect get 1300KB/s -1600KB/s on an 8MHz bus, so 2600KB/s on a 13,75 MHz bus sounds about right to me.

And speedsys is reporting 3600KB/s on that same system with the Maxtor IDE? That's awesome. Can you post a screen shot?

On older screenshot with an older drive already close

Marco wrote on 2024-03-04, 21:29:

What is indeee much faster is the speedsys “linear verify speed”: 172000kbyte/s DOM vs 52000kbyte/s HDD. Whatever this number means as it is far beyond isa capabilities.

The verify command doesn't push data over the bus. It just reads the storage internally and reports "Success" or "Fail". It's kind of a need idea in theory. It would tell you how fast the device can read without pushing data over the bus. But in some cases, I suspect that the IDE firmware cheats and just reports "Success" as soon as it sees the verify command to try to get a better benchmark score. In those cases, the only thing you bench marking is how fast your CPU can push ATA commands back and forth to your IDE device.

Ah thanks also

Marco wrote on 2024-03-04, 21:59:

Ah thanks also

Congrats on a top ISA IDE benchmark score.

Post your 386 Speedsys results here

Here the latest screenshot

I tested the same with a ide2cf adapter and the results are more or less the same as with the DOM. Maybe both don’t support that multi sector transfer

Marco wrote on 2024-03-05, 14:30:

I tested the same with a ide2cf adapter and the results are more or less the same as with the DOM. Maybe both don’t support that multi sector transfer

You are using the exact same XtIde Universal BIOS build, yes?

Maybe you can get XUB to tell you what it is seeing.

https://www.xtideuniversalbios.org/
Boot menu drive information
The Boot Menu can display a little bit of information about the drive:

• Capacity This shows the drive capacity. This is the same as reported by the drive unless you have specified CHS or LBA manually. Capacity is read from INT 13h AH=08h for drives not handled by XTIDE Universal BIOS.
• Addr. This shows the current addressing mode:
  - NORMAL is used for drives with 1024 or less cylinders (504 MiB / 528 MB and smaller drives). NORMAL is the fastest mode since no address translations are required.
  - LARGE is used for drives with 1025...8192 cylinders. LARGE addressing mode L-CHS parameters are generated with Revised Enhanced CHS calculation algorithm. LARGE addressing mode can use LBA if drive supports it.
  - LBA is used for drives with 8193 or more cylinders and LBA support. L-CHS parameters are generated with Assisted LBA algorithm.
• Block Shows the maximum number of sectors supported in a single multi-sector transfer. The larger the better. 1 means that block mode is disabled or not supported. CF cards usually supports block mode commands but do not allow blocks larger than 1 sector.

Sorry I just have the 8kb rom burned and cannot change to the large rom anymore. I read that only the large rom version can show up the information.

But anyway thanks

Hi, I know this isn't exactly a solution to the problem, but it might be a workaround:

By using ab HDD cache, you can spare the DOM more often.
So there will be less likely a situation in which it gets being overburdened.

On the other hand, though, a HDD cache like SmartDrive usually costs precious conventional memory.

Luckily, the Helix Multimedia Cloaking package has one SmartDrive alternative included that doesn't take up much conventional memory (it uses Extended Memory).

The cloaking manager and compatible utilities once were sold but sometimes also bundled for free with other software, like mice driver software.

If that's not being an option, there is still Central Point PC-Tools (v7 etc), which comes with a tiny HDD cache that's still smaller than SmartDrive.
Maybe it will do good enough.

Edit: This is just an idea, of course. No one is forced to try this.

It just came to mind, because the IDE HDDs of my youth/childhood usually featured a "sector cache".

The cache was about 64 KB in size, sometimes smaller.
It was intended to make sure that the HDD didn't require multiple revolutions to read a whole sector.

In addition, these HDDs also had a debug port on the PCB.
By attaching a serial cable (TTL levels, not RS-232 levels !), it was possible to read the data of the HDD's on-board computer.

Example:
https://stason.org/TULARC/pc/hard-drives-hdd/ … -SL-IDE-AT.html

By contrast, CF cards do have a very little cache (1KB, maybe 2KB), if at all.
It's more of an i/o buffer, though, as with a serial port.

Thanks a lot for your detailed input.

Two points here:
1. from major experiences back in the days I would only recommend smartdrv over all alternatives. Mem isn’t a problem currently

2. in particular at XUB you can read that on slow pcs with fast hdd incl large hdd caches the smartdrv could even reduce performance. The explanation is quite clear but this one I’d like to benchmark.

Rgs

Conclusion: after reinstalling a lot I will switch back to my modern hdd. There is really little to non performance improvements because of the very low DOM/CF access times on a 386sx. Installations take same time as the cpu is bottlenecking expanding files etc. Furthermore having no noise at all takes away the whole retro feeling for me. Anyway a good try. I will stick to m ide hdd with 16mb cache onboard. Lessons learned and evenings spent 😀

Next will be benchmarking smartdrv on/off there.