wrote:wrote:I can run and post the full screen low and high settings for doom for completeness, if you think that's missing.
was there more than 1-2 fps between the cards?
wrote:Updated the 386 Doom tests according to your suggestions.
so it was 3fps between the worst and best VGA in 386DX40 Doom, or ~30%, seems significant for attaining that unplayable/playable balance at the time, thank you.
Not sure how I missed this juicy thread.
wrote:Will try to "smoke" the ET4000 VL/PCI at some point soon.
I'm a little lost amongst all delicious pr0n posted herein. Did you benchmark the ET4000 PCI vs. ET4000 VLB on an identical system? What was the outcome?
wrote:I'm a little lost amongst all delicious pr0n posted herein.
Thank you Sir, your comment is well appreciated.
wrote:Did you benchmark the ET4000 PCI vs. ET4000 VLB on an identical system? What was the outcome?
Soon.
Monkeying around with some "impressive" underperforers at the moment. Wanted to post some results and findings already, but cannot get myself to move on. Will run some VLB/PCI tests right after.
wrote:Chips 65554 4Mb
How have your experiences been with this card? I saw on the video card compatibility list that this may be the only GPU that's VESA 2.0 natively and can handle the PowerVR version of Tomb Raider well. Is that a 4 MB card? Also, is that an integrated printer port?
I spend my days fighting with clunky software so I can afford to spend my evenings fighting with clunky hardware.
The video card is ok-ish. An average performer. It has 4Mb of memory. Travelling at the moment and cannot remember what was the other port, but you are probaby right - an LPT one.
The main issue with it is that it messes up picture size and aspect ratio on some LCDs.
Never tried it with pvr tomb raider, so cannot comment really - sorry.
The original post was about building exemplary retro PCs, but since then i spent some time playing with hardware from the other side of the spectrum - the underperformers.
Accumulated enough data and decided to update this thread with it.
286 from 1990-1992
HARDWARE
NOTES
The motherboard is actually not that bad - handles IDE and FPU really well, but hates video cards - slows them down to a crawl.
Coretest for SCSI freaks-out when the CPU runs at 25MHz.
The Harris CPU does not overclock at all - anything above the default 20MHz is a no go.
IDE is much faster than the SCSI.
THE CULPRITS
https://petershipkov.com/temp/retro_pc_images … motherboard.jpg (featuring the Harris CPU)
https://petershipkov.com/temp/retro_pc_images … _2_et4000ax.jpg
https://petershipkov.com/temp/retro_pc_images … s/286_2_ide.jpg
https://petershipkov.com/temp/retro_pc_images … 6_2_adaptec.jpg
https://petershipkov.com/temp/retro_pc_images … ts/286_2_io.jpg
286 at 25MHz
https://www.petershipkov.com/temp/retro_pc_im … _25_2_stats.png
286 at 30MHz
https://www.petershipkov.com/temp/retro_pc_im … _30_2_stats.jpg
386 from 1993-1994
HARDWARE
NOTES
This motherboard is a top underperformer.
Initially i thought it may be unoptimized BIOS, but trying 30 other BIOS-es strongly indicated that the problem is the hardware itself. From past experience - if the hardware is ok, some BIOSes can improve its performance at least partially. Not this time around. There is a real log jam somewhere in that mobo.
SpeedSys memory metrics indicate no cache, which is strange. Tried different sets of chips to eliminate the chance of faulty ones - no bueno. This seems to be the main issue of the board.
THE CULPRITS
https://petershipkov.com/temp/retro_pc_images … motherboard.jpg
https://www.petershipkov.com/temp/retro_pc_im … rruslogic_1.jpg
https://petershipkov.com/temp/retro_pc_images … s/386_2_ide.jpg
https://petershipkov.com/temp/retro_pc_images … 6_2_adaptec.jpg
https://www.petershipkov.com/temp/retro_pc_im … _40_2_stats.png
https://www.petershipkov.com/temp/retro_pc_im … _2_speedsys.png
486DX2 from 1993-1994
HARDWARE
NOTES
Another impressive underperformer.
Kind of strange, because exactly the same chipset is used successfully by Asus and other manufacturers in very fast motherboards.
Visual inspection and comparison with such motherboards confirms that the hardware components match. So it must be the BIOS then.
Tried over 40 different BIOS-es. Started with ones from other SIS 461 based boards, but at the end the one that finally "clicked" was an UMC BIOS. Unexpected i admit, but the mobo woke up. VGA and IDE performance was where i expected them to be. The only trouble was an apparent disagreement between the BIOS and the memory controller about the amount of installed RAM. So, they settled on 1MB only.
The system does not boot at 40MHz FSB.
Cannot find manual for the CL combo board. Its default IDE performance is the same as the GoldStar Prime 2.
Also, there are memory expansion slots - any idea what chips should go there ?
THE CULPRITS
https://petershipkov.com/temp/retro_pc_images … motherboard.jpg
https://www.petershipkov.com/temp/retro_pc_im … 6dx2_et4000.jpg
https://petershipkov.com/temp/retro_pc_images … dx2_adaptec.jpg
https://petershipkov.com/temp/retro_pc_images … x2_goldstar.jpg
https://www.petershipkov.com/temp/retro_pc_im … 86dx2_stats.png
https://www.petershipkov.com/temp/retro_pc_im … x2_speedsys.png
BENCHMARKS
RENDERING
D'aww! I <3 motherboards that are just big enough for their ISA slots. 😄
There's a pretty significant difference between those Adaptec SCSI cards. Huh. Would not have expected that. And I'm still trying to get over how fast the Trident 8900 is.
I haven't ever really been concerned with benchmarks before, but this is kind of making me want to run some comparisons on the hardware I have lying around. I've got a 386SX/25 and a DX/40 that I've been trying to decide which to pair with an Orchid Fahrenheit 1280 and a Trident 8900D. I'm a little more concerned with Win 3.11 performance on the DX and assumed the 1280 would be faster than a Trident. But now I'm not so sure...
Now I'm also wondering how my three VLB graphics cards compare... my DX2/66 has an ATI Graphics Pro Turbo (mach64) because that's my old setup from back in the day, but I also have a Cirrus and Tseng ET4000/w32i. I could be missing out.
Another vote, BTW, for the 4000 VLB vs PCI. I don't have a PCI 486, and I'm curious how VLB and PCI compare.
Yeah, the compact mobos are cool in their own way.
I really like the Trident VLB VGA, even S3 does a bit better in Doom, which is the test that actully matters, i am still using the Trident one as the main video card in the rig.
Run some tests if you have the time - any info like tht is well appreciated.
In case my questions got lost in the long post above:
- i cannot find any info about the Cirrus Logic combo jumper settings. Any hints ?
- what memory should go into its available slots ?
Any idea why the 386 mobo seems to be cache-less ?
It is not BIOS settings, or faulty chips for sure.
Ok, will post VLB vs PCI tests at some point soon.
Too bad i fried a ET4000w32 VLB adapter during some of my "underperformers" jurney. It was a dark day. 🙁
Thanks.
It seems my VLB Tseng ET4000/w32i bit the dust as well. I just get an "unsupported resolution" on my workbench LCD, and my OSSC shows a couple red boxes on the screen, but that's it. sigh...
Wish I could help with the questions. That combo card is quite peculiar, and handy given the low number of VLB slots available on 486 boards. The SIP slots look like RAM, alright -- presumably video RAM. But so many pins! I've never seen SIP memory that long.
I decided to test my ISA / VLB kit. This is all on a 486 DX2/66 motherboard with VLB, using Landmark 6.0 (char/ms), PC Player Benchmark (fps), and 3DBench2 (fps). I bring you, the ISA Trident trio, ISA and VLB Orchids, and ATI Mach64 VLB:
Trident 8900CL (512K) - LM: 1164 chr/ms; PCP: 2.3fps; 3DB2: 16.4fps
Trident 8900D (1M) - LM: 4349 chr/ms; PCP: 4.1fps; 3DB2: 32.8fps
Trident 9000C (512K) - LM: 2678 chr/ms; PCP: 3.2fps; 3DB2: 26.4fps
Orchid Fahrenheit 1280 (1M) - LM: 1755 chr/ms; PCP: wouldn't run; 3DB2: 16.3fps
Orchid Kelvin 64 VLB (Cirrus GD-5434, 2MB I believe) - LM: 8330 chr/ms; PCP: only runs in 320x200 (10.8fps); 3DB2: 44.7fps
Tseng ET4000/w32i VLB (2M) - forfeit on account of not working anymore, apparently. suck.
ATI Graphics Pro Turbo VLB (Mach64, 4MB) - LM: 9274 chr/ms; PCP: 4.5fps; 3DB2: 45.7fps
Most surprising to me is, again, how fast that 8900D really is. What a sleeper. Who knew? The PCPlayer benchmark is clearly CPU limited, but you do get to see where the cards hold back performance a little. VLB is faster (around 2x in this case), but not 4x despite the 8MHz to 33MHz bus clock speed jump. Interesting.
I'm rather happy with these results, as it turns out the builds I wanted for nostalgic reasons turned out to be the best performers. Nice!
Put your numbers in charts for better readability. Thanks for making the effort. It will be awesome if we see some doom stuff as well.
Btw, what is the mobo you tested on ? Brand, model, chipset, etc.
I will try to test all the isa, vlb and pci cards on my asus pvi-486sp3 mobo and merge your results as well.
TBH, I don't know how you make those graphs. I've seen that style before -- is it an app or site that you use? I could toss them in Excel or something, but it wouldn't be as pretty. 😀
The board is a Chaintech 4ULD with a UMC chipset (UM8498F).
It is Excel. There are some templates in the "view" bar when you select a chart.
Thanks for the clarification about the Mobo.
Recently i wanted to verify how two motherboards based on the same chipset compare when one of them is "burdened" with VLB and the other one is not.
So, i picked Asus PVI-486SP3 and LuckyStar-486EF and ran the usual set of tests.
Both of them are from the same period 1995-1996 and rely on SIS 496/497.
Along with that i wanted to see how they handle the highest class 486 CPU at the time (AMD-X5-133ADZ 160MHz) as well as Pentium Overdrive P24T.
Both boards can run the POD at 100MHz. It performed well in the simple DOS tests, but failed at the 3D rendering ones, so i decided to keep it at the default 83MHz. Trying to stick here to my rule of working with completely stable systems.
BIOS settings set to the lowest available timings.
Under such conditions the LuckyStar mobo was quite picky about the quality of the installed RAM.
LuckyStar-486EF featuring the POD83 CPU.
http://www.petershipkov.com/temp/retro_pc_ima … motherboard.jpg
LuckyStar-486EF with AMD-X5-133ADZ 160MHz
http://www.petershipkov.com/temp/retro_pc_ima … s_dx5_stats.png
http://www.petershipkov.com/temp/retro_pc_ima … x5_speedsys.png
LuckyStar-486EF with Intel Pentium P24T 83MHz
http://www.petershipkov.com/temp/retro_pc_ima … s_pod_stats.png
http://www.petershipkov.com/temp/retro_pc_ima … od_speedsys.png
ASUS PVI-486SP3 rev 1.22 featuring the AMD 486DX5 CPU.
http://www.petershipkov.com/temp/retro_pc_ima … motherboard.jpg
ASUS PVI-486SP3 with AMD-X5-133ADZ 160MHz
http://www.petershipkov.com/temp/retro_pc_ima … 86dx5_stats.png
http://www.petershipkov.com/temp/retro_pc_ima … x5_speedsys.png
ASUS PVI-486SP3 with Intel Pentium P24T 83MHz
http://www.petershipkov.com/temp/retro_pc_ima … i_pod_stats.png
http://www.petershipkov.com/temp/retro_pc_ima … od_speedsys.png
Used Matrox Millennium MGA-2064W-R3 8Mb - a top dog back in 1996.
http://www.petershipkov.com/temp/retro_pc_ima … x5_matrox_1.jpg
score: LuckyStar (1) : Asus (0)
score: LuckyStar (1) : Asus (1)
score: LuckyStar (2) : Asus (1)
score: LuckyStar (2) : Asus (2)
score: LuckyStar (2) : Asus (3)
score: LuckyStar (2) : Asus (4)
score: LuckyStar (2) : Asus (4). This one is a toss.
score: LuckyStar (2) : Asus (5)
score: LuckyStar (2) : Asus (6)
Brief summary:
Final result is 6 to 2 in favor of Asus PVI-486SP3.
I expected the PCI only LuckyStar mobo to perform somewhat better than the PCI/VLB ASUS-PVI one, but apparently PCI and VLB can coexist just fine without any added dead weight.
Quite surprised by the achieved 20+ FPS by PVI+POD in the Quake test.
The synthetic tests gave better CPU scores to the LuckyStar motherboard, but the 3D rendering ones tell very different story.
Now lets compare the PCI/VLB/ISA ASUS-PVI board against ASUS VLI-486SV2GX4 rev 2.0 - a VLB/ISA one. As usual, i am capping hardware at year 1996.
Used the same AMD DX5 CPU clocked at 160MHz. BIOS timings set to the lowest available values (for best performance).
Used modified BIOS from Evolution of a Socket3 System to a POD @100MHz. It enables the motherboard to function well with L2 cache in WB mode.
Initially i didn't notice performance difference between 256/512/1024 Kb of cache, but closer inspection proved me wrong, so i maxed out the cache to 1024Kb for the contest.
Performance of VLB video cards is much more volatile compared to PCI ones from that period, so instead of picking a single card and running all the tests with it, i compared couple of them on both mobos for more complete picture.
The Weitek video card required +1 WS to cope with the system timings.
http://www.petershipkov.com/temp/retro_pc_ima … motherboard.jpg
http://www.petershipkov.com/temp/retro_pc_ima … i_dx5_stats.png
http://www.petershipkov.com/temp/retro_pc_ima … x5_speedsys.png
http://www.petershipkov.com/temp/retro_pc_ima … ake_wintune.png
http://www.petershipkov.com/temp/retro_pc_ima … pbench_doom.png
Brief summary:
ASUS VLI-486SV2GX4 is clearly one of the best "clean" VLB motherboards out there - stable, fast, compatible.
The two systems are more or less equal at handling DOS/Windows graphics, with slight advantage for VLI.
PVI shows advantage in the 3D rendering department.
Also its integrated IDE controller seems to be one of the best in its class. It this case - 3 times faster than the external IDE controller i used on the VLI. Hope to find time soon and experiment with VLB SCSI/IDE controllers to see if i can outdo the PVI's one.
Now i am curious about how VLB-less ALI/UMC based motherboards from the same 1995-1996 period stack-up against the tested ones here.
Hope to find some answers soon.
Story continues on page 3.
Spent few hours over the weekend to find out how different generations and types of video cards compare against each other.
Time period is capped at 1996.
Used unaccelerated DOS performance tests only as a common denominator because many of the older cards don't support accelerated GUI features, so no point of comparing them against the next gen ones.
Started running the tests with CPU at 160MHz CPU and FSB at 40MHz, but some of the ISA/VLB cards were too unstable.
Decided to step down to 133MHz CPU and 33MHz FSB to give all "players" a fair chance. Obviously, the results don't show peak performance of the more potent "participants", but the relative performance difference between all tested video cards. Feel free to compare the numbers below with the peak performance tests in my first post in this thread.
All tests were performed on Asus PVI-486SP3 rev 1.22 motherboard. For those who have doubts about the motherboard's ISA/VLB/PCI performance - take a look at my previous post.
Notes:
- Couple of people asked for comparison between ET4000/W32 ISA/VLB/PCI on the same motherboard - there you go.
- the UMC ISA adapter runs at 50MHz (core speed), but produces visual artefacts. At 40MHz picture is clear, but perf is low.
orange outline denotes best performance in particular time period
solid orange bar denotes best performance across the board
https://petershipkov.com/temp/retro_pc_images … isa_vlb_pci.png
Typing the names of the tested video cards here, so they can be searched:
__________ 1996 PCI __________
Tseng Labs ET6000
Trident ProVidia 9685
S3 Trio64V2/DX
PC-Chips 65554
Matrox Millennium MGA-2064W-R3
Cirrus Logic GD-5480
__________ 1994 VLB __________
Tseng Labs ET4000/W32i
S3 805
Trident TGUI9440AGI
Weitek Power 9000
Cirrus Logic GD-5428
__________ 1993 PCI __________
Tseng Labs ET4000/W32P 2Mb
__________ 1993 ISA __________
Cirrus Logic GD-5426
Cirrus Logic GD-5420
S3 928
Tseng Labs ET4000/W32i
Trident TVGA9000C
Trident TVGA8900CL-C
Realtek RTG3105
WDC90C30-LR
UM85C408AF
__________ 1990 ISA __________
Western Digital WD90C00-JK
OTI-037
Headland GC208-PC
Ahead V5000-50PC-B
Cirrus Logic GD-510/520
Trident TVGA8800CS
Tseng Labs ET4000AX
Brief summary:
Trident TGUI9440AGI is the best option for unaccelerated DOS graphics (gaming). In reality the performance difference between the VLB/PCI cards is marginal. They are all in the ballpark.
Interesting reading for later, thanks 😀
Nice! Yeah, you can really see the difference moving from ISA to a local bus for graphics. Not a surprise, but this clearly shows how significant that change really was.
Notice how ahead of the game ET4000AX was when released in the late 1980'ies. No wonder it made such a name for itself.
Competitors quickly closed the gap and bypassed the ETs, but their charm stayed throughut the 90ies.
At the same time Trident was looked down so much because of their widespread cheapo products, but they actually offered and some of the best performers. Looking at that TGUI boy - it is quite impressive.
Updated the ASUS PVI-486SP3 vs LuckyStar-486EF post with info about ASUS VLI-486SV2DX4 to make it a 3-way comparision: PCI - PCI/VLB - VLB
Decided to share some findings about quite interesting motherboard - the ASUS 386/33-64k.
The hardcoded 64Kb of DIP22 SRAM, max 8Mb of RAM and barebone BIOS (no timings, etc.) hint at an early stage 386 hardware, offering somewhat subpar performance.
UPDATE: After stumbling upon some random information online, i just realized that this motherboard is what Asus started their business with back in 1989 - the one they developed "blindly" without having access to an Intel's 386 CPU prototype. Or, at least that's what the legend says.
I was annoyed by the lack of options in the BIOS, so tried bunch of other ones that are rich on cache/ram timings.
Results were not great - performance was same or worse, compared to the default BIOS.
Okay, fine. Lets see if it overclocks then.
I didn't expect much, given the already low entry point, but for my surprise this boy turned out to be unexpectedly capable in that department.
It works completelly reliably in DOS and Windows 3.1 at 50MHz. Well, with one exception - it failed all 3D rendering tests. None of the 3DCC apps can get to UI, let alone render something - a big red-flag in my books.
There was a bit of complication with finding the right video card for the 50MHz FSB. After combing through bunch of them, it was an ET4000AX from 1998 that did the job. The rest were unstable, or didn't light up at all.
I started with 60ns RAM, but noticed that 70ns is fine too.
Ok, what about 55MHz ?
The motherboard takes it with ease (along with the CPU and VGA), but the IDE controller gives up.
Tried bunch of other IDE/SCSI adapters - they all hang during their turn at the end of POST.
Too bad ...
Screen does not light-up at 60MHz.
http://www.petershipkov.com/temp/retro_pc_ima … motherboard.jpg
http://www.petershipkov.com/temp/retro_pc_ima … _2_et4000ax.jpg
For some reason i was not able to capture screenshots of NSSI.
Quick comparison with one of the fastest 386 motherboards i know of - the PC-Chips M321.
As you can see, the system ticking with 50MHz is enough to overcome its inherent slowness and position itself ahead of M321 in most tests.
Even the ASUS motherboard lags in DOOM full screen + high detail and WinTune2, the DOOM test is not really relevant for a 386 system, so it kind of does not matter and the Windows GUI performance while slower than the M321 is still within the performance ballpark.
Conclusion:
Despite the relativelly low base performance of this motherboard, it transforms itself into one of the fastest boards i have seen to date, when running at 50MHz.
The 8MB RAM limit won't be a problem for games, but will be for memory demanding productivity applications and for Windows 3.x.
While the failures with 3D rendering hint at a problem in there, i think a great 386 gaming system can be built around it.
I came late and saw a topic. I liked it very much 😀
can you take a photo of your memory? brand and how many nanoseconds? which is located on 286-VLSI
you also disabled "parity error check" ? turn it on - failed?
