VOGONS


First post, by Marco

User metadata
Rank Member
Rank
Member

Hi all,

I just let some benchmarks ran on my retroPC. Some interesting numbers as I think although I have the feeling that the results do sometimes differ based on the n° of benchmark-passes.

Base system:
- 386SX/27,5
- 13,75MHz ISA
- GD5428 Video
- 8MB RAM
- CF and 7200/16MB HDD

Have fun

Attachments

Last edited by Marco on 2024-03-13, 18:55. Edited 1 time in total.

1) VLSI SCAMP 311 | 386SX25@30 | 16MB | CL-GD5434 | CT2830| SCC-1 | MT32 | Fast-SCSI AHA 1542CF + BlueSCSI v2/15k U320
2) SIS486 | 486DX/2 66(@80) | 32MB | TGUI9440 | SG NX Pro 16 | LAPC-I

Reply 1 of 5, by douglar

User metadata
Rank Oldbie
Rank
Oldbie

I think what you are seeing there is a combination of a couple things:

  • The Windows 95 storage driver is capable of doing much higher throughput on writes
  • The Windows 3.11 storage driver does faster FAT access, maybe because it is not fat 32 ?
  • Winbench suffers in Windows 95 because it is likely needs more than 8MB to avoid swapping Edit: see next post
  • 3DBench DOS does not run well in a dos window, probably because of the VGA virtualization, and Windows 95 has better virtualization than Windows 3.11

It would be interesting to know why Win95 has such an advantage in storage throughput when it isn't likely to be multi-sector transfers. Perhaps it is because it is using 32bit protected mode commands for its PIO operations.

Last edited by douglar on 2024-03-13, 18:55. Edited 2 times in total.

Reply 2 of 5, by douglar

User metadata
Rank Oldbie
Rank
Oldbie

It looks like Winbench has a lot of variability based on the OS. I wouldn't be surprised to find out that it is heavily dependent on a select group of GUI calls, which could make the results vary a lot with the video driver, too.

https://www.os2world.com/wiki/index.php/BENCH … indows_and_OS/2

Reply 3 of 5, by Marco

User metadata
Rank Member
Rank
Member

interesting OS/2 link.

my remarks here:

File transfer results:
- these are really a ?? to me. I cant see a scheme here.
- all OSs run on same FAT16.
- Win95 shows no compatibility issues
- WfW3.11 allowed 32bit file access but ony BIOS access for drive access. Thats maybe due to my XT-IDE Bios.
Any idea what to improve here?

Download speed results:
- added.
- weird with WfW. The driver seems to allow much closer HW-usage than on W95 maybe?
- also quite slow on DOS. Maybe here I would need packet driver optimizations (buffers ???)

GPU results:
- as stated to fuzzy, hard to compare. As I always used exactly the same tools it might be related also to Win32s@WfW usage vs nativeW95 capabilities
- WinTach seems to be most "stable" but maybe even then hard to compare (see bulletpoint above)

1) VLSI SCAMP 311 | 386SX25@30 | 16MB | CL-GD5434 | CT2830| SCC-1 | MT32 | Fast-SCSI AHA 1542CF + BlueSCSI v2/15k U320
2) SIS486 | 486DX/2 66(@80) | 32MB | TGUI9440 | SG NX Pro 16 | LAPC-I

Reply 4 of 5, by Marco

User metadata
Rank Member
Rank
Member

I also now tried with The microhouse driver. It will work. With some limitations:
- strangely only 32bit disk access for my large hdd. My 408mb hdd will be set to 16BIT
- the file access will be limited to 16BIT now for all drives. Before it was 32bit for all drives.

What does this mean for the benchmarks:
- 15mb file now 41s instead of 52s
- small files now 68s instead of 39s

Easy decision to go back to std hdd driver w/o 32bit disk access.
Or maybe better someone has an idea on how to reach 32bit for all options 😃

1) VLSI SCAMP 311 | 386SX25@30 | 16MB | CL-GD5434 | CT2830| SCC-1 | MT32 | Fast-SCSI AHA 1542CF + BlueSCSI v2/15k U320
2) SIS486 | 486DX/2 66(@80) | 32MB | TGUI9440 | SG NX Pro 16 | LAPC-I

Reply 5 of 5, by kingcake

User metadata
Rank Oldbie
Rank
Oldbie
douglar wrote on 2024-03-13, 17:34:

It would be interesting to know why Win95 has such an advantage in storage throughput when it isn't likely to be multi-sector transfers. Perhaps it is because it is using 32bit protected mode commands for its PIO operations.

Yes. VCACHE algorithm was also improved over WFW. WFW only uses VCACHE if in 32-bit disk access mode. Otherwise it relies on DOS caching like Smartdrv. I don't know if this benchmark can disable caching or not.