I've heard bus logic VL SCSI cards are better than the Adaptec ones. I guess soon you will be able to verify for us.

Anonymous Coward wrote:

I've heard bus logic VL SCSI cards are better than the Adaptec ones. I guess soon you will be able to verify for us.

I don't have the specifics on me but I replaced a Adaptec VLB SCSI controller with a Bus Logic version in my 66mhz 486 and it has performed better and given me less trouble.

Very nice 486! How did you tweak the CT2230?

@PeterLi: I recapped the amplifier section of it and also most of the sound-generator section. The result was a noticeable improvement in treble and bass - previously the card was quite muffled. ELGEN and Wincap capacitors really do not age well. Due to me being a doofus with the iron though, I've - ahem - "uniquely" recapped it (in other words, there're caps sticking out every which way, on both the component *and* solder sides!) Because of this I can't really say I've "repaired" it as... well... here, see for yourself:

Amazingly it works and sounds much better than it looks. (Thank goodness)

Well, it's been a while since I last posted. Work and real life can be quite annoying for me, leaving me with not much free time. Anyway, after acquiring an IBM PS/1 2168-BB1 in very poor condition (and subsequently fixing it up -- I'll document my exploits later if there is enough interest), I've realized that the 2168's chipset performs lower overall than the Alaris Cougar 486BL's OPTi chipset, so I've repurposed this machine to be my 'midrange 486-class'. I've thrown my Pentium Overdrive 83MHz into it and changed the VL bus video card to a Boca Voyager VL (Chips and Technologies F64300) with 2MB of RAM and two 44256 "cache" chips. The CHIPS F64300 is a positively brilliant VGA chipset - it's got excellent DOS/VGA speed plus some GD5429-level GUI acceleration to make Windows 95 more bearable. Unfortunately I did not acquire an Adaptec VL-bus scsi card as I had planned, so I will continue to press the BusLogic into service.

CACHECHK reports 10 for the 1~16KB range, 22 for the 32~256KB range and 48 for everything above 256KB. Curious that the OPTi would be so fast at caching but slow at regular RAM reads/writes. (The score with the onboard 486BL3 isn't much different - I suspect it's the OPTi chipset to blame for the mediocre RAM speed as I am using 60ns SIMMs with the memory timings set to their fastest speeds in the BIOS.)

SPEEDSYS scores:

Seems SPEEDSYS really likes the Pentium Overdrive, likely due to the matter I mentioned beforehand (over-emphasis on FPU efficiency). No, I don't believe those HDD scores at all either. What's in there is the same IBM DCAS 2GB SCSI unit, a 5400rpm unit with 8.5ms seek. Such seek times as reported would outperform the 10K Fujitsu SCSI unit in my Windows 98 rig! I suspect the BusLogic is 'cheating' on SPEEDSYS's benchmark somehow. (I hate things that 'cheat' on benchmarks, because they obfuscate objectivity in these tests.)

Uh oh, I noticed my cheap HP pavilion m703 monitor flickering with snapping sounds. Guess I'll be budgeting for another CRT... (Voodoo cards don't like flatpanels much in DOS, at least not under Commander Keen.)

Pictures of the rest of the gear (no, it's nowhere near as mangled looking as that poor SB16) to follow in my next message.

As promised here are the pics of the inside. The outside of the case is rather generic and drab in my opinion, so I've cut to the good stuff.

overview:

the pentium overdrive and 12/15ns cache chips:

the boca voyager VL card:

and finally just to show the CPUs:

The Alaris Cougar 486BL board is definitely unique among 386/486 VL boards. It's not often you'll find a motherboard that can accommodate a 386, 486 or a Pentium in one platform. (At least, not often outside of huge lumbering PS/2 towers.) While the 486BL chip is 486SX-compatible and has broadly equivalent performance, it's technically a 386 chip with IBM's extraordinary 16KB cache, clock tripler and other tweaks. This gives me a fair bit of flexibility for what I want to do: if I want to run older games that run too fast, I take the PODP5V83 out and run it on the 486BL. If I need an FPU (but not the Pentium's speed) I stick the Cyrix FasMath in. Or, if I need/want max performance, I stick the PODP5V83 in. (An Am5x86-133 would not be much faster than the onboard 486BL-100 and would generally only benefit me in benchmarks and programs that run dirty 286/386 code with no notion of Pentium features. Also the board likely does not run 3.3v to the CPU socket so I'd need an adapter for it, like a trinity works powerstacker.)

I'm curious about the Boca Voyager VL card. How did you benchmark it?

The Boca Voyager, I relied on SPEEDSYS. It's actually faster in DOS than the Tseng it displaced under that program. Under Windows it feels like a GD5429, at least by me.

I recently read an article on the Boca Voyager in an old PC Magazine. It didn't get very good reviews. I think the review was based mostly on windows driver speed. It was kind of vague on DOS performance though.

I wouldn't rely on speedsys scores for anything except maybe the memory graphs. I'm more interested in how it performs under real world situations. Btw, I noticed that the speedsys videoscore changes more with the motherboard than the videocard.

I can believe that. It doesn't exactly feel 'slow' in any case. You'll notice though that one of the SOJ sockets is a bit crooked, though that doesn't seem to affect its operation.

I'll have to do some doom timedemos to get something more reliable for speed benchmarks. Under Mode13h at least the Chips screams.

OK, here're the Doom timedemo scores, run with default options (default screen size) and no sound under plain vanilla DOS (no memory managers loaded):

with Pentium Overdrive PODP5V83 CPU: 1564 realtics (47.75 points), placing it about 2.5 points above the Am5x86-P75 baseline here: http://thandor.net/benchmark/32
with onboard IBM 486BL-100 CPU: 2491 realtics (29.98 points), placing it a mere 0.09 points below the ST 486DX2-80 baseline. (Recall the 486DX2-80 uses a 40MHz bus, whereas this is using 33MHz.) The score is the same with and without FPU, curiously.

Not bad for a motherboard which is technically a very hot-rodded 386DX bus.

Eep386 wrote:

OK, here're the Doom timedemo scores, run with default options (default screen size) and no sound under plain vanilla DOS (no me […]
Show full quote

OK, here're the Doom timedemo scores, run with default options (default screen size) and no sound under plain vanilla DOS (no memory managers loaded):

with Pentium Overdrive PODP5V83 CPU: 1564 realtics (47.75 points), placing it about 2.5 points above the Am5x86-P75 baseline here: http://thandor.net/benchmark/32
with onboard IBM 486BL-100 CPU: 2491 realtics (29.98 points), placing it a mere 0.09 points below the ST 486DX2-80 baseline. (Recall the 486DX2-80 uses a 40MHz bus, whereas this is using 33MHz.) The score is the same with and without FPU, curiously.

Not bad for a motherboard which is technically a very hot-rodded 386DX bus.

I thought it was more like a 386SX bus. 😜 It's a neat board anyway, as you can easily cover almost the entire early-mid 90s period of DOS gaming.

I wonder, does the Blue Lightning actually support the full 486 instruction set, or is it basically just a souped-up 386SX?

Cool modifications and interesting updates.

I thought it was more like a 386SX

It's an easy mistake to make. You're thinking of the IBM 486SLC. The IBM 486DLC (Blue lightning) is based on the 386DX. Both chips are identical internally, but have different numbers of data and address lines.

The highest load is usually experienced by TAG ram, which is used whenever the system uses cache. That's why people often put one grade faster chips there: 20ns if main cache is 25ns, 15ns if main cache is 20ns etc.

That is, your cache is only as fast as your TAG ram. I suggest you use 12ns chip for TAG and put the 15ns in the main cache. Not that you'll notice any difference, that is just illogical to use slower TAG.

RacoonRider wrote:

The highest load is usually experienced by TAG ram, which is used whenever the system uses cache. That's why people often put one grade faster chips there: 20ns if main cache is 25ns, 15ns if main cache is 20ns etc.

That is, your cache is only as fast as your TAG ram. I suggest you use 12ns chip for TAG and put the 15ns in the main cache. Not that you'll notice any difference, that is just illogical to use slower TAG.

I would use a faster TAG, but the tag chip is soldered to the board. Thanks, MSI USA/Alaris. 😒 Just as well, I hope the tag chip doesn't fail anytime soon.

I did notice though, that the board insists on good quality SRAM chips. This board kept crashing and glitching until I bought a bunch of AT&T 12ns chips.

Ok, more benchmarks. Both were run in pure vanilla DOS with no drivers loaded. All comparing against the reference systems listed in these benchmarks: http://thandor.net/benchmarks

With the Pentium Overdrive 83MHz:

Vidspeed L: 22140 bytes/millisecond (scores 1427 above the Stealth 64 2MB PCI/P100 baseline)
Syschk: 46189 char/sec
DOSNO286: 15835 (vs. 15764 for the reference Pentium Overdrive system)
X-Vesa 1.0.2:
- read 8bit: 2430K/sec 16b: 4850K/s 32b: 9660K/s 64b: 9440K/s 80b: 5150K/s
- write 8bit: 10600K/sec 16b: 21100K/s 32b: 41200K/s 64b: 41200K/s 80b: 34800K/s

With the onboard 100Mhz IBM 486BL/3 *no FPU*:

Vidspeed L: 22170 bytes/millisecond
Syschk doesn't run on the 486BL/3 (gives ERROR 22 at 19A8:4C73, I suspect faulty CPU detection)
DOSNO286: 980 (faster than the i486SX2/50 at least...)
X-Vesa 1.0.2:
- read 8bit: 2550K/sec 16b: 5100K/s 32b:10100K/s (no 64 or 80 bit scores, needs an FPU)
- write 8bit: 10700K/sec 16b: 21300K/s 32b: 42200K/s (ditto)

When we add the Cyrix FasMath 33MHz...:

Vidspeed L: same as above
Syschk still won't run (same result as above)
DOSNO286: 5517 (much better, though still pretty low... about 130 points above the Cx486DX-40GP, which isn't bad considering it's using a 387 for the FPU!)
X-Vesa 1.0.2: (sub 64-bit figures were the same as above, so only 64/80-bit figures posted)
- read 64b: 6230K/sec 80b: 6060K/s
- write 64b: 7160K/sec 80b: 6340K/s

All things considered, these scores all seem very satisfactory to me. The 486BL seems to have slightly better (~1%) I/O scores than the Pentium Overdrive, whilst the POD cleans house everywhere else. Obviously some benchmarks dramatically favor FPU throughput (DOSNO286 and SPEEDSYS being two examples), giving very different results when the CPU is changed. Hilariously Syschk *hates* the Buslogic and IBM DCAS combo, turning in a ridiculously low 652K/sec throughput, whereas SPEEDSYS gave those ridiculously _fast_ seeks. Syschk also hates the IBM 486BL... I suspect a bug in their CPU detection code.

I changed the Boca BIOS on the Chips VLB card to the Expertcolor DSP6430's BIOS (another F64300-based VLB card), and now Syschk doesn't crash on the 486BL anymore. In addition X-Vesa sees a lot more video modes (and they seem to work!) Unfortunately Dawn Patrol's SVGA support is still wonky on it, and SDD/UniVBE keeps dying trying to detect the card. VBETEST doesn't hang up on it anymore though, but a lot of VESA modes don't work right when double-buffering is used (I suspect CHIPS's VBE is less than perfect.) I find it odd that X-VESA works properly with it (which does much more intensive I/O) but not VBETEST.

I tried swapping the card with an STB Powergraph C33 (WD90C33-ZZ based) VLB card, but that turned out to be a lot slower than the Chips under Wolfenstein 3-D. Dawn Patrol doesn't like the WD card either in SVGA mode - I think that game should be submitted to the VGA compatibility list. SDD doesn't die with it though at least, but it only allows 4bpp modes with the utility.

After burning a new BIOS for the PowerGraph, SDD works much better, giving me a vastly wider variety of modes. Performance didn't change much, however.

OK! After changing some jumpers (that Total Hardware '99 called 'factory reserved') on the STB PowerGraph C33 VLB, its performance jumped nearly 2x. It's now about 75-80% the speed of the Chips, but with far fewer compatibility issues (UniVBE runs properly so long as I tell it not to use linear framebuffers.)

Since it's not really documented anywhere that I can see, here's what I've ascertained from the STB PowerGraph C33's JP2/1-3 settings:
JP2/1: Waitstate jumper 0?
JP2/2: Waitstate jumper 1?
JP2/3: Sync rate? (when set, it uses regular VGA sync rates, but when removed, it outputs at different frequencies - crappy monitors and flatpanels beware!)

Apologies for digging up this rotten old corpse of a thread, but I got a different Alaris Cougar II board and a different CHIPS F64300 VLB card since my last post to this ancient thread. They work just fine together this time, and the Chips is behaving itself. My last board, I probably just ESD'd or something.

This time around I am leaving it at the stock 75MHz / 25MHz x3 speed with the original IBM 486BL/3 CPU. I ain't gonna try to hotrod it this time. The stock 15ns cache chips work well so far, knock on wood.
I've since got faster boards for hotrodding so I don't need to torture this board.