VOGONS


First post, by Return 0;

User metadata
Rank Newbie
Rank
Newbie

Hi, recently I found a motherboard as mentioned in the topic, I powered it up, and to my surprise, the performance is very low (Pentium 166MHz P54C 2.5x66 + 2x256KB SRAM cache + 32MB(2x16) 50ns EDO), despite having very low timings for RAM (Read - 2, Write - 1) and cache (0/0/0 read/write/tag hit) in BIOS. Additionally, L2 is set to WB, TAG to 7+1, and Page mode is on. Analyzing SPEEDSYS, I conclude that the issue lies with memory write performance. Here are the results:
L2 Read/Write/Move: 87/35/57 [MB/s]
RAM: 62/32/25 [MB/s]
CPU: 124pts
VGA : 35MB/s
PCPlayer (640x480) - 7.4fps
Quake (320x240) - 29fps
Does anyone have experience with this kind of hardware? The write performance is worse than on a 486... Any suggestions? maybe some mobo modifications?

Reply 2 of 11, by Return 0;

User metadata
Rank Newbie
Rank
Newbie

With cpu overclocked to 200Mhz
IMG-3620.jpg
IMG-3618.jpg

My P24T (OC 110) with HOT-433 to compare:
IMG-3466.jpg

Reply 3 of 11, by pyrogx

User metadata
Rank Member
Rank
Member

That's one of those UMC UM890 chipset based boards you have there. They are kinda rare and not much is known about the chipset either (no datasheets, nothing).
I have a similar UM890 board (Pine PT-733) and I remember that it performed quite decently - not on par with an Intel Triton but still fine for an early Pentium board.
However mine has the UM8891/92BF chipset while yours has UM8891/92AF. Difference seems to be the support for PB-Cache and EDO-DRAM timings which the AF variant seems to lack. Maybe that's why the performance is so low on your system.

Reply 4 of 11, by Return 0;

User metadata
Rank Newbie
Rank
Newbie

Ohhh, but HOT-433 have UMC chipset too but lower series 8881/8886 and have great performance... With POD@100 is even (fps) faster than this 200Mhz P54C. The same async SRAM cache (512KB 15ns + 64KB 10ns TAG). So it really adding some timmings but only to write? Read and move is quite good.

Reply 5 of 11, by pyrogx

User metadata
Rank Member
Rank
Member

486 is a different platform, so performance is not really comparable to a pentium system. OPTi also had good 486 chipsets, but their pentium chipsets were crap, all of them.

Reply 6 of 11, by Return 0;

User metadata
Rank Newbie
Rank
Newbie

It make sense. So I will be angry if I will buy it after 486 in 90' hah...

Reply 7 of 11, by dionb

User metadata
Rank l33t++
Rank
l33t++

I don't have any tips, but want to add that I experienced this too with this chipset, with memory performance on both Chicony CH-890A and Biostar MB-8500UUR (both with asynch cache) that was abominable, comparable to an EDO chipset with integrated VGA, or one running on 32b memory bus width (single 72p SIMM in chipsets that supported it) - so bad in fact I thought that this chipset only had a 32b memory bus. That is incorrect, others reported much better performance, on par with other early EDO Pentium chipsets.

Those were tests I did a looong time ago and I don't have the boards anymore, so can't troubleshoot.

Reply 8 of 11, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

My intuition also says that something isn’t quite right there. That graph is just too flat between cache and memory from what I usually see on my systems

But without experience with that chipset, It’s hard to say personally.

I do wonder what performance would be with a k6-3 on that system. That may be your fix. You can underclock it to 166 if you wanted. You’ll need an interposer. Pcb files for one in my sig, or grab one off fleabay

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 9 of 11, by Return 0;

User metadata
Rank Newbie
Rank
Newbie

Thank you for answer, it looks flat because of max Value. It cannot be 32bit because works only with pair of memory sticks. My brain can not handle that theoreticaly better chipset is worse... I understand if performance will be the same like on hot-433 but is much lower with almost 2x higher freq CPU. So I will start searching another socket 5 board.

Reply 10 of 11, by pyrogx

User metadata
Rank Member
Rank
Member

I found some Speedsys results from the board I have, with a P1-133, 32MB of RAM and 512k PB-Cache. Numbers are quite different:

The attachment SSTIMG01.JPG is no longer available

Reply 11 of 11, by Return 0;

User metadata
Rank Newbie
Rank
Newbie

Yes, memory writing result is much higher. One more thing, my board is running with standard async SRAM cache ICs (W24257AK-15) maybe that's a problem? I just look around and almost all socket5 boards are equiped in "M" or "FK" SRAM chips, like W24M25AK or UM61C256FK. Maybe board recognise standard SRAM chips and set up higher timmings?

Edit:
I was blind but now I see... Checked: Wrong way, without cache memory writing is still on the same low level. But I in my opinion it should be higher- propably like in previous screen from P24T (also 32b, also async cahce and also UMC chipset but socket3).

Edit 2:
Evidence! I flashed BIOS from Biostar MB-8500UUR and it is better:
image0.jpg
image1.jpg%22

But with this BIOS on-board IDE controller refuse to work, so I run speedsys from floppy.