Evolution of a Socket3 System to a POD @100MHz

Showcase your Retro PC / Build logs.

Re: Evolution of a Socket3 System to a POD @100MHz

Postby PC-Engineer » 2019-12-02 @ 19:23

@feipoa
I have some additional questions to your 486 Comparison and the 3D Games Comparision

  • In your 3D tests you used for the 40MHz FSB CPUs a PCI speed of 27,5MHz (post #56) and 40MHz (post #71), was the PCI @40MHz stable with complete testing config (Graphics + 3Dfx + IDE + ...)?
  • In the 486 comp. all 40MHz FSB tests were tested with 20MHz PCI speed. Hurts to see like the 20% overclocked CPUs perform often slower than the originals in the DOS applications. Why did you chose the M919 for your tests (3D tests too) and not in general the MB8433 or one of your faster SiS496 boards - 66MHz support?
  • I compared the (in general low) values for the PCPbench (640x400x8 LFB) with my results with the HOT433 (same chipset, other graphics card). Seems, that the chipset has an issue with the LFB. Could you find the rootcause for that?
  • Which 486 PCI board is your personal favorite and why?

I am very interested in the backgrounds.
User avatar
PC-Engineer
Newbie
 
Posts: 47
Joined: 2019-7-07 @ 08:26

Re: Evolution of a Socket3 System to a POD @100MHz

Postby feipoa » 2019-12-03 @ 10:19

PC-Engineer wrote:Seems, that my diagram suggest a ranking. It is difficult to reed between the lines. The intention was to list the absolute results for direct compairison of individual values from other users.

I think the majority of the people who see this thread will skim it, look at the charts, and move on. Since the CPUs on the chart are ordered from lowest sum of results to highest sum, people could draw the wrong conclusion and rank the CPUs in this fashion. Perhaps if the CPUs weren't listed in descending order, but were more random, it would be less likely that someone could misinterpret the results. I also appreciate the value in the raw data, so I usually include normalised and non-normalised data. The normalised chart is a lot more representative of total performance for these four benchmark condition; however, it is still not a fair overall comparison because results from PCPbench are tallied twice, allowing for PCPBench to hold twice the importance of Quake or DOOM.

Interesting that PCPBench is 10% faster with the i486DX-120 compared to the cx5x86-120. My previous results indicate only a 1-2% lead for the i486DX-120 at 640x480x8.

PC-Engineer wrote:CPU scaling via overclocking: 33MHz - 40MHz FSB (+20%)
intel DX2: +19,5%
intel P24T: +19,5%
intel DX4: +19,4%
Cyrix 5x86: +19,2%
AMD DX4: +16,6%
AMD Am5x86: +13,5%

The scaling issue with the Am5x86 appears to be due to your DOOM results. I wonder if you jotted down the value correctly? I got closer to a 20% increase in my previous tests. Your results show only a 4% inrease in DOOM from 133 MHz to 160 MHz.

PC-Engineer wrote:I got a 2842A and and replaced the 2842VL. The Jumper of the 2842A was allways set. Do you have any indication that the jumper has a negative influence?

It has been so long since I setup this system, but I don't think I did much analysis on this.

PC-Engineer wrote:I thought this would be a wait-state jumper, but without a performance drop (only measured in transparent mode) by toggeling. I beleave it must have another function (interrupt?). Couldn't found a hint for it until now. And no, i didn't consider the synchronize mode in the comparison. Do you have an expection?

Could be the IRQ 2/9 jumper. Some cards have two jumpers, one for IRQ, the other for 0 WS.

When using a Trio64 on this motherboard with DOOM, transparent yielded a 6.3% improvement (61.7 fps vs. 65.6 fps). With the Mach64, there was a 12.1% improvement (58.44 fps vs. 65.50 fps). I use an S3 Vision968 and must use synchronise as the system won't boot with transparent. For comparison, synchronise yields ony 58.4 fps. If the Trio64 came in 4 MB, I'd probably use that card on this system.

Are you able to measure the difference in synchronise vs. transparent with your graphics card in this board?

PC-Engineer wrote: In your 3D tests you used for the 40MHz FSB CPUs a PCI speed of 27,5MHz (post #56) and 40MHz (post #71), was the PCI @40MHz stable with complete testing config (Graphics + 3Dfx + IDE + ...)?

Could you paste vogons links to the post? I'm not sure how to view post #56 and #71. For the Ultimate 486 Benchmark Comparison, when the FSB was set to 40 MHz, the PCI bus was automatically set to 26.7 MHz. Unfortunately, the PC Chips m919 sets the PCI bus automatically. I didn't realise this when I started benchmarking. It is the only 486 board I've experienced which does this. I used a different motherboard for 486 CPUs in the Ultimate 686 Benchmark Comparison so that the PCI bus was kept at 40 MHz. Also, the latest benchmarks using the Voodoo on a socket 3 have the PCI bus at 40 MHz: viewtopic.php?p=697548#p697548 The hardware used in that thread is listed here: viewtopic.php?p=690851#p690851

For the Biostar MB-8433UUD and the hardware selected, I did not experience any issue with the PCI bus at 40 MHz. I have noticed that many PCI-based socket 3 systems do not function well with a PCI network card, the MB-8433UUD being the exception.

PC-Engineer wrote:In the 486 comp. all 40MHz FSB tests were tested with 20MHz PCI speed. Hurts to see like the 20% overclocked CPUs perform often slower than the originals in the DOS applications. Why did you chose the M919 for your tests (3D tests too) and not in general the MB8433 or one of your faster SiS496 boards - 66MHz support?

I figured this out too late. There is no indication in the BIOS that the BIOS is scaling down the PCI bus when a 40 MHz FSB is set. I realised something was odd with the result and discovered the issue after hooking an oscilliscope up to the PCI CLK signal. On my list is to redo the 486 benchmarks with the MB-8433 and a more fine-tuned selection of benchmarks.

I dodn't have a SiS 496 board which can do 66 MHz reliably. Apparently the Lucky Star LS486 can do 66 MHz reliably, but I don't have one. The MB-8433UUD can do 66 MHz reliably.

PC-Engineer wrote:I compared the (in general low) values for the PCPbench (640x400x8 LFB) with my results with the HOT433 (same chipset, other graphics card). Seems, that the chipset has an issue with the LFB. Could you find the rootcause for that?

I've had about a dozen HOT-433 boards over the years and have always run into some sort of problem with them. I still have two original versions of HOT-433 and two 4.0 boards, but don't care for them at all. The fact that both revisions of this board have a non-functional PS/2 mouse header is a bad sign from the onset.

PC-Engineer wrote:Which 486 PCI board is your personal favorite and why?

MB-8433. It can be modded to accept 1024K. Has a working PS/2 mouse port. Can boot Windows 2000 without any issue. I could not get the HOT-433 to boot W2K using a PCI SCSI host adapter card. The system would lock-up between HDD access and CD-ROM access. There are some other nice UMC 8881-based boards with 1024K, but they don't have a PS/2 mouse header. For SiS 496 boards, I like the Chaintech-486SPM MSI MS-4144. I have the DTK PKM-0033S setup in a system with a Cyrix 5x86-133/4x; it is a SiS 496 board, but it comes with those darn leaky rechargeable batteries, so by now, most boards for sale will be ruined. I've convered the board to using a native PS/2 mouse and coin cell battery.

A more general answer is that I like any PCI 486 board which has 1024K cache capability and PS/2 capability. In many cases the KBC needs to be modified for PS/2 capability. I do not care much for boards which have PCI+VLB because of bus bridging which reduces performance.

I do not care for boards which only support single-banked L2 cache.
User avatar
feipoa
l33t++
 
Posts: 6746
Joined: 2011-3-07 @ 13:54
Location: Canada

Re: Evolution of a Socket3 System to a POD @100MHz

Postby PC-Engineer » 2019-12-06 @ 22:58

I found the root cause for the worse values of the Am5x86 in Doom (and Quake). If i set JP26 from ND to D then Doom jumps from 57,5 to 64,3 fps (Quake from 17,2 to 17,5). So the performance increase (for the four Benchmarks) from 133MHz to 160MHz jumps from +13,5 to +18,5, which is a reasonable scaling now.
If i keep JP26 in ND position then, apart from PCP bench Mode 100, the synchronize mode is clear faster than the transparent mode. This effect is only observable with this CPU and only @160MHz. Additional i got two (not reproducable) crashes while the short benchmark session with JP26 in D position and transparent mode, one crash took place while leaving the editor in Norton. With the ET4000W32 i got similar benchmark values and no crashes (while this short usage), but this card din't post with JP26 in ND and transparent mode. Toggeling this jumper in synchr. mode showed no influence to the benchmark values. Maybe toggeling JP26 could be a solution for your problems with your S3 968 in transparent mode.

I updated my previous posts and the diagrams with the new findings.

Additionally i did benchmarks for the comparison with trnasparent and synchronize mode:
Synchr. vs Transp.JPG
Synchronize Mode vs. Transparent Mode POD@100MHz and Am5x86@160MHz


Doom: ----------------- POD +7,4% / Am5x86 +4,8%
Quake: ----------------- POD +2,3% / Am5x86 +0%
PCPBench SVGA: ------- POD +7,6% / Am5x86 +7,9%

feipoa wrote:I've had about a dozen HOT-433 boards over the years and have always run into some sort of problem with them. I still have two original versions of HOT-433 and two 4.0 boards, but don't care for them at all. The fact that both revisions of this board have a non-functional PS/2 mouse header is a bad sign from the onset.

Understand, but i didn't try the PS2 port until now on my HOT433 and i have no need to install Windows 2000 :happy: But appart from this two issues this is a decent board with a (online) support page at the supplier. The 486SPM is very decent too, but my board needs help to support PS2 - i read your tutorial.
I was wondering that you got your MB-8433 running stable with 66MHz and the cache hack because of EMC and signal quality of your wiring - well done!
In general i cannot understand why many user favorite the M919.

Thanks a lot for the explanations!

PS: i had bad experiances with single banked cache too, especially the chinese ISSI chips seems to be alergical against single banked mode (also with 33MHz)
User avatar
PC-Engineer
Newbie
 
Posts: 47
Joined: 2019-7-07 @ 08:26

Re: Evolution of a Socket3 System to a POD @100MHz

Postby feipoa » 2019-12-07 @ 03:07

I've made the cache hack jumperable now, so I can more easily switch between 1024K and 512K double-banked modes. For 66 MHz FSB, I went back to using 256K because memory read was more reliable with only 1 WS. For 1024K, I had to use 2 WS for the level of stability I desire. What I would like to do is fine a way to do your dirty TAG modification so that I can use 256K in write-back mode and cache all the 64 MB of memory. But I don't have the UMC 8881 pinouts.

It is not just W2K that the HOT-433 has trouble with. Every now and then I would see the same symptoms on boot with NT4. I'm guessing PCI bus mastering wasn't properly implemented for host controller cards. How long have you been using the HOT-433? I thought it was a good board in the beginning too, but not after years of testing.

Single banked cache requires more wait states than double-banked for 40 MHz. Some older motherboard manuals identify this in their "auto" mode for memory in the BIOS.

Looks like your percent increase for transparent mode falls in like with my results of between 6-12%. Unfortunately, I have a hand written note in my manual that I must set the VESA clock delay (JP26) to "Delay" rather than "No Delay". And based on your tests, you mentioned that "Delay" and "Transparent" caused some irregular crashes. I could always test it again though.
User avatar
feipoa
l33t++
 
Posts: 6746
Joined: 2011-3-07 @ 13:54
Location: Canada

Previous

Return to System Specs

Who is online

Users browsing this forum: No registered users and 6 guests