VOGONS


LuckyStar LS486E rev.C2 and Cyrix 5x86@133

Topic actions

Reply 40 of 110, by rad

User metadata
Rank Newbie
Rank
Newbie

Thanks for that info! Unfortunately the BIOS options don't have such feature anywhere on the screens. I've tried to modify my BIOS dump from awdflash and then open it with modbin utility but unfortunately there is no such option available anywhere. I'm attaching the BIOS ROM here if you can dig further into it.

Award BIOS Editor refuses to open that file at all. I thought the dump was not OK, but I've tried to compare the BIOS with the dumps located here:
http://sannata.ru/bios/486PCI/
and my BIOS dump and the LS486EC2.zip are exactly the same files (hash checksum verified).

Also loading directly the BIOS BIN data from memory into modbin also didn't reveal additional options or hidden features. I wish I could use modbin6, but this is only for 6.00 BIOS versions...

Any ideas how to tweak this setting? Chipset should support it, at least what is written on page 65 in the datasheet for ISA Bus Interface.

BTW, setting the ISA speed to PCI/3 didn't help.

Reply 41 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I was able to open your BIOS with MODBIN, howver your suspicions are confirmed, there does not exist a hidden I/O Recovery Time option. I wonder if this is hard coded via some other means. You can still try alternate ISA sound cards to determine if any of them are stable. Alternately, some PCI sound cards are known to have better DOS support than others, lke the Vortex 2, which also has a header for an XR385. Best to establish stability and optimal settings with the hardware you currently have, then try to find some alternate means to achieve ISA sound.

You should be able to add PS/2 mouse capability to your motherboard, and 512 single-banked cache.

Plan your life wisely, you'll be dead before you know it.

Reply 42 of 110, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie

Ok after some more testing I have to say my board is really weird. In dos, I cannot run quake 1 stable w/o FPM memory (2x16mb). I get 18.2 fps in dos quake (133MHz enhancements enabled). Unfortunately I cannot boot into windows unless I run 2x60mhz. It will freeze a few seconds after showing the desktop.

I can get into windows stable running with enhancements on using EDO, but then dos quake isn't very stable (I get the occasional hang) and it seems the board refuses to detect my voodoo 2 card..

feipoa wrote:

I was able to open your BIOS with MODBIN, howver your suspicions are confirmed, there does not exist a hidden I/O Recovery Time option. I wonder if this is hard coded via some other means. You can still try alternate ISA sound cards to determine if any of them are stable. Alternately, some PCI sound cards are known to have better DOS support than others, lke the Vortex 2, which also has a header for an XR385. Best to establish stability and optimal settings with the hardware you currently have, then try to find some alternate means to achieve ISA sound.

You should be able to add PS/2 mouse capability to your motherboard, and 512 single-banked cache.

is there a guide posted someware?

Reply 43 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

kanecvr, to me your symptoms sound like your CPU isn't getting enough voltage for 133 MHz operation. If you are up to 3.85 V and these results still occur, it may be that your CPU isn't reliable at 133 MHz. Did you follow rad's work to add variable CPU voltage to your LS486E? It is on page two of this thread. It parallels this earlier work, Modifying your motherboard's voltage regulator for overclocking

Non-detection of a Voodoo2 card on a SiS496-based motherboard isn't a big problem because you can use a Voodoo3.

Information on adding native PS/2 mouse support can be found in this post, Native PS/2 mouse implementation for 386/486 boards using the keyboard controller

Plan your life wisely, you'll be dead before you know it.

Reply 44 of 110, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie

Yea, I noticed... At 3x40 the V2 is detected fine. It seems 120mhz is the chip's limit.

Thanks for the links. The thing is It seems the LS486E doesn't have a KB controller - it must be built into the chipset or something, so I don't think I can follow the guide you posted.

Odd thing is, the board has a 5-pin header (CN5) that is unpopulated - it looks suspiciously like a PS/2 header. I did solder pins onto it but I couldn't get a PS/2 mouse to work. It could be an IRDA header.

fTuKWbTl.jpg

By continuity checking I found that Pin 1 is +5v, pin 4 is GND, and pins 3 and 5 trace to the Winbond W837 I/O chip - does this mean it's an IR header?

Reply 45 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I appologise. I remember I had looked up the image for this motherboard a few weeks ago and recalled it having the standard DIP-40 keyboard controller. Either I remembered incorrectly or the Google Images displayed the incorrect image.

At any rate, even if the keyboard controller is built into the chipset, it is possible to track down the appropriate MOUSE CLOCK and MOUSE DATA pins. However, on this particular motherboard that VIA chip might be the keyboard controller. Is it? If it is, check the spec. sheet to see if it supports PS/2 mice.

Plan your life wisely, you'll be dead before you know it.

Reply 46 of 110, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie

The chip is a VT82C41N keyboard controller, but I can't find a datasheet for it. I was able to find datasheets for the VT82C42 and C42N and those support PS2 mice... just can't seem to find any info on the VT82C41N except for the fact that it is a 40 pin KBC

Reply 47 of 110, by rad

User metadata
Rank Newbie
Rank
Newbie

Well in order to use PS/2 port I think it should be enabled first in the BIOS options. When you open your BIOS via MODBIN you'll see that PS/2 mouse support is disabled and hidden from the menu options. Try to enable it there and reflash the BIOS to see if it'll start to work afterwards.
EDIT: what about that 4-pin connector that is located on the edge of the board just besides the battery cell? /EDIT

I was able to finally stabilize the board with CPU-PCI mem post buffer enabled and I was even able to set the DRAM speed to Slower instead of Slowest. For now those settings are the tightest possible at all. Every other faster setting refuses to boot after POST or is not stable at all. I've completed another round of MemTest for more than 8 hours as well as completed a full run of SuperPi of 16M which took 13h 9m. For now I've replaced the ISA card with ESS1938s PCI sound card, which for now is OK. I've an additional MPU-401 internal ISA card with Roland MPU-401 external module, and attached CM-32L and SC-55mkII modules and all of them work perfectly in the current setup. ESS soundcard has good compatibility in DOS and doesn't require any TSR programs at all, only one exe to initialize it at boot.

Here are some benchmark results at 2x66MHz operation with enhancements enabled:

Norton SysInfo 8.0: 423.3

PC-Config v8.20:
20350% (speed compared with IBM-PC);
84000, 30952 (Dhrystones, KWhetstones)
VideoRAM throughput (text): 10551KB/sec
Chars/sec via BIOS: 79530 Byte/sec
Chars/sec via DOS: 51870 Byte/sec
1lst Level-Cache throughput: 254500 KB/s
2lst Level-Cache throughput: 37100 KB/s
Memory throughput: 25300 KB/s

PCPBench (VGA mode): 24.8
PCPBench (SVGA mode): 12.6

LandMark 2.0:
Integer ALU (Mhz): 622.13
Floating-point FPU (Mhz): 1623.95
Video (char/sec): 21845.00

DOOM timedemo: 1344 realtics; 55.57 fps

Quake 1.06 timedemo:
18.8 fps full screen (health bar + ammo bar visible)
17.6 fps full screen (health bar only visible)
16.6 fps full screen (only 3D area)

3DBench2: 103.4

PC Bench 9.0:
CPUMark16: 171.03
16bit Protected Mode Large Mix: 48.20
Math Coprocessor: 58429.50

CPUIndex 2.3: 19

SuperPi (win 95):
32k: 33s
1M: 31m 35s
16M: 13h 09m 27s

Justin Benchmark (win 95):
Integer: 0.037
Float: 0.046
Text Processing: 1.63
I/O Processing: 0.741
Memory Access: 0.068
Execution Time: 2.522

VidSpeed 1.10 (640x480x8bit):
Copy DRAM to banked VGA: 24.70 million bytes/sec
Copy DRAM to linear framebuffer: 28.02 million bytes/sec

Few more attached screenshots from benchmark tools:

The attachment IMG_0979.JPG is no longer available
The attachment IMG_0980.JPG is no longer available
The attachment IMG_0985.JPG is no longer available
The attachment IMG_0986.JPG is no longer available
The attachment IMG_0987.JPG is no longer available

EDIT: I'm going to find out Voodoo3 2000 PCI card and can run GLquake afterwards.
Do you know from where I would be able to find 10/12 ns cache chips? And if I find 1024kbit chips to populate and replace current ones and upgrade the L2 cache to 512kB, should it be detected properly by the system? In the manual it says that the jumper can switch between 128 and 256kB, but doesn't mention anything about 512kB. I wonder if it'll be detected automatically, is it supported at all and if it is, what jumper configuration it'll need.

Reply 48 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

You results are most impressive - faster than the UMC-based board in some tests. The memory read speed is the single most result that stands out to me. Otherwise, our L1 and L2 speeds are the same. I noticed that I only scored 9.9 in PCP Bench SVGA mode. I was using a Matrox G200. Is the G200 slower in DOS compared to the Millennium II? We should really try to compare our systems using similar hardware. Unfortunately, I do not own a Millennium II. I have a Virge DX, Trio64V2/DX, Virge GX, Trio3D, Ati Rage128 VR, Ati Rage 128 Pro, Mystique 220, RIVA TNT, Matrox G200, Matrox G200 MMS, GeForce2 MX400, 3DLabs Permedia 2, 3DLabs Oxygen VX1, ATI Rage Pro, and some Diamond Stealth with an Alliance chipset. Do you have any of these to compare with?

Keep in mind that if you upgrade your cache to 512 KB, you may need to tweak your timings again (or you may not!). For the 10 ns chips, you may need to find some IC suppliers from Asia. Maybe there are some tips in this thread, Super Fast 486 Cache - 10 ns

It is suspected that the 10 ns and 12 ns cache chips may be counterfeit, however they seem to work regardless. It would be nice to actually measure their response times.

If you look at the manual for the Zida Tomato 4DPS, a board which has a similar layout to your LS486, there are 2 jumper blocks which set the cache size, JP20 and JP21. You may be able to draw some analogy between the 4DPS and your LS486.

Plan your life wisely, you'll be dead before you know it.

Reply 49 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I pulled out my IBM 5x86c-133 (Biostar MB-8433UUD) system and ran some comparative tests using a Virge GX against your Luckystar LS486E.

LS486E vs. 8433UUD

Quake
18.8 vs. 19.8

DOOM
55.6 vs. 61.1 (62.0 if using Matrox G200)

3dbench2

103.4 vs. 95.7

Pcpbench SVGA
12.6 vs. 12.5 (10.1 if using Matrox G200)

Pcpbench VGA
24.8 vs. 27.8

Speedsys
CPU Score: 75.5 vs. 75.5
L1: 193.6 vs. 190.4
L2: - vs. 73.4
RAM: 62.9 vs. 56.5 (or 51.3 if RAM Read WS=2 instead of 1ws. WS=1 is stable if RAM <=64MB)

Cachechk
L1: 274.9 vs. 273.8
L2: ~101.6 vs. 101.6
RAM Read: 85.8 vs. 69.5 (or 55.6 if RAM Read WS=2 instead of 1ws. WS=1 is stable if RAM <=64MB)
RAM Write: ~92.4 vs. 92.6

Landmark
CPU: 622 vs. 621
NPU: 1624 vs. 1618

Sysinfo
423 vs. 422

I do not really understand why your Quake and DOOM scores are that much lower than mine, yet your 3dbench2 score was quite a bit better than mine. I figured Quake would make more use of your super fast RAM read speeds, but my system came out ahead. Why? Do you have any of the PCI cards I listed so that we can compare our systems with the same graphics?

Note that when I use a Matrox Millennium G200, my scores are the same as the VirgeGX's scores except that the DOOM score increases to 62.0 fps and my PCPBench SVGA score drops to 10.1 fps.

Last edited by feipoa on 2016-04-06, 23:52. Edited 2 times in total.

Plan your life wisely, you'll be dead before you know it.

Reply 50 of 110, by rad

User metadata
Rank Newbie
Rank
Newbie

Well from the list of graphics cards the one that I have is S3 Virge/DX. I can place it in my system and repeat those tests again in the morning and post the results here.

I really don't have an explanation what could be the difference in Doom and Quake scores, but let me try with the DX card and then compare the results. It's very surprising for me about the difference in the memory read speed but it seems that there aren't a lot of programs that take benefit from it, eapecially under DOS. Do you know some more memory-dependant progs where we can see real world difference because of that memory speed difference?

Reply 51 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Yes, please do retest with the Virge DX. I think the Virge DX and GX are supposed to be similar in performance.

I do not know of any "real world programs" that will definately benefit from the faster RAM read times. I wish more of the 1995-1998 games had some timedemo. Something which moves a lot of information to and from main memory, perhaps the Direct3D game Battlezone? That game run okish, but I do not beleive there is a timedemo.

Plan your life wisely, you'll be dead before you know it.

Reply 52 of 110, by rad

User metadata
Rank Newbie
Rank
Newbie

Here are the results with Virge DX... pretty much the same, just a little bit slower than Millenium II:

LS486E (DX) vs. LS486E (MII) vs. 8433UUD

Quake 1.06
18.8 vs. 18.8 vs. 19.5

DOOM 1.9s timedemo3
54.8 vs. 55.6 vs. 61.1

3dbench2
102.6 vs. 103.4 vs. 95.6

Pcpbench SVGA
12.6 vs. 12.6 vs. 12.4

Pcpbench VGA
24.7 vs. 24.8 vs. 27.6

Cachechk
....
RAM Write: 92.9 vs. 92.9 vs. 92.4

VidSpeed
Copy DRAM to banked VGA: 18.43 vs. 24.70 vs. ??
Copy DRAM to linear framebuffer: 28.02 vs. 28.02 vs. ??

One thing that could maybe cause that is the L2 cache. I'm forced to put it to WT mode in 60/66 MHz operation otherwise the system is not stable. When I start the timedemo it gives different errors during loading and can't complete the test. Can you retest your scores with L2 cache set to WT? I don't think the amount of L2 cache could be culprit, but maybe WT/WB could?

EDIT: What version of Quake and DOOM you're using to test for those values?

Last edited by rad on 2016-01-26, 14:11. Edited 2 times in total.

Reply 54 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Interesting. I didn't realise at Millennium II was faster than a VirgeDX. I wonder if your VirgeDX is clocked slow?

In the past, I did not notice much, if any, DOS speed change when using L2: WB vs. WT. I will run it again to be sure. For using a 66 MHz FSB on the MB-8433UUD, I found that L2-WB was more stable than L2-WT.

Do you think removing one stick of RAM will increase my scores? Right now I have 1024K (WB) with 128 MB. If I use 64 MB, won't more cache be available, thereby increasing the cache hits?

Plan your life wisely, you'll be dead before you know it.

Reply 55 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I have confirmed that using only 64 MB instead of 128 MB did not improve Quake, DOOM, 3dbench2, or PCPBench socres with either L2: WB or WT. I left my Matrox G200 installed for these tests.

Setting L2 to write-through mode :
Quake: no change
DOOM: realtics went from 1210 (WB) to 1215 (WT)
PCPBench: no change
PCPBench VGA: framerate went from 27.6 (WB) to 26.8 (WT)
3Dbench2: framerate went from 95.1 (WB) to 94.7 (WT)

In short, the changes in Quake and DOOM scores with WT vs. WB do not explain your lower scores for these benchmarks. Be sure that sound is not setup in DOOM (run setup.exe). For Quake, try running Quale -nosound -nomouse

Plan your life wisely, you'll be dead before you know it.

Reply 56 of 110, by kixs

User metadata
Rank l33t
Rank
l33t

For one, more cache means better performance. You have 1024kb while rad has only 256kb. As previously posted it means at least 5% boost.

Requests here!

Reply 57 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I also have 128 MB of RAM. It takes 1024 KB to cache all that RAM in write-back mode. So are you saying that,

1024 KB WB cache + 128 MB RAM is faster than, say, 256 KB WB cache + 32 MB RAM? Does that make sense? In both cases, the entire quantity cache is needed to cache the stated quantity of RAM.

Could you verify your statement with some numbers/benchmarks for the scenerio where more cache performs better than less cache when all the cache is used to cache the max quantity of RAM? For the results you identified in the other thread, you are using way less than the cacheable limit of RAM.

Plan your life wisely, you'll be dead before you know it.

Reply 58 of 110, by kixs

User metadata
Rank l33t
Rank
l33t

Cacheable area is determined by the chipset and the TAG chip and not by cache size itself. That's why Pentium Pro with it's on-chip 256KB L2 can cache all 4GB memory.

Cache size is used as a fast "mini memory" and if the data is small enough and fits entirely in this cache it will run a lot faster. More cache, more data fits in it. But it all depends on the app. Some apps won't see any benefit, some will run "a lot" faster.

On the other hand... DOS fills memory from 0 to whatever. Windows fills it backwards - from what you have (128MB to 0). That's why it matters for Windows to cache all the memory. As for DOS you will rarely use more then 16 or 32MB. I only use the memory modules at hand and to be big enough for my tests. I might do some comparisons with smaller cache & more main mamory when I have some free time (spring/summer).

Requests here!

Reply 59 of 110, by feipoa

User metadata
Rank l33t++
Rank
l33t++

The TAG RAM stores the addresses of discrete address blocks in system memory and in the examples provided with the SiS 496 and UMC 8881, the TAG size is sufficient for the size of RAM installed. It is not an issue for these tests.

We are dealing with the UMC 8881 and SiS 496 chipsets, both of which use direct-mapped cache, meaning that each cache line is assigned to a set number of memory blocks, and multiple blocks share a single line. The greater the number of blocks (the more RAM you have) which are shared by a single cache line means the greater the frequency of dumping and reloading of data to the cache. This occurs when the information the CPU requests comes from multiple blocks that happened to be shared by the same cache line. Each cache line can only hold a single memory block at any given time. So the less memory your system has, the less frequency each cache line needs to share a block of memory (less thrashing). Less sharing means more cache hits. This is why having less memory for the same quantity of cache should theoretically increase your cache hit rate, however these DOS benchmarking programs require far too little RAM access for this to be noticeable. If you use 4 MB of RAM and have 1 MB of cache, then each 256 KB of cache is assigned to only 1 MB of RAM. The chances of a cache hit are greater than if you are using 128 MB of RAM, whereby each 256 KB of cache is assigned to 32 MB of RAM (more block sharing required). Unfortunately, direct mapped caching does not allow for the least frequenty used cache lines to be reassigned to other memory blocks like on more modern computers. I think set associate allows for remapping of cache lines to other block sets.

If my understanding on this concept is incorrect, please correct me.

The memory requirement of DOS Quake is 8 MB. An interesting test would be to run the Quake timedemo on a system with:

1024KB-WT & 8 MB, 16 MB, 32 MB, 64 MB, 128 MB, and 256 MB. If 8 or 16 MB don't show the best results, then we might need another program which is more demanding on RAM. Note that with direct-mapped cache, 1024 KB of cache in write-through mode can cache up to 256 MB with these motherboards.

256KB-WT & 8 MB, 16 MB, 32 MB, 64 MB. Note that with direct-mapped cache, 256 KB in write-through mode can cache up to 64 MB.

256KB-WT & 64 MB should provide the same results as 1024KB-WT & 256 MB, however 1024KB-WT & 8/16 MB should be faster than 256KB-WT & 8/16 MB.

Plan your life wisely, you'll be dead before you know it.