Anonymous Coward wrote:

There is a reason to go RAMBUS on the PIII. From what I recall back then if you wanted 133MHz FSB with an intel chipset, you pretty much had to use the i820 or i840. The i815 which came out later was gimped because you couldn't upgrade beyond 512mb. i840 can go to 2GB.

I have heard that the memory bandwidth on a PIII with RAMBUS is actually pretty damn good. I think the bigger issue at the time was that it was @#$%ing expensive. If I were going to build the ultimate PIII system of 1999, I say i840. I think I could learn to love it like my P60 VLB system.

I recently got my hands on a Dell Precision 220 (Intel i820) 2x 1ghz PIII
and a Precision 640 (Intel i840) 2x 833mhz Xeon

You are very correct about the memory bandwidth is fast...my VIA694T chipset dual 1ghz PIII system crawls in comparison
I'm really suprisedhow speedy the i820 system is....can't say the same about the i840 system...the Xeon cpu's are just plain PIII's in a different package

swaaye wrote:

The catch is that PC133 already saturates a Pentium 3's FSB. They have the same bandwidth. 64-bit 133 MHz SDR FSB = 64-bit 133 M […]
Show full quote

The catch is that PC133 already saturates a Pentium 3's FSB. They have the same bandwidth. 64-bit 133 MHz SDR FSB = 64-bit 133 MHz SDR SDRAM.

There were also some DDR P3 chipsets from VIA. They were rather pointless again because of how PC133 was already saturating the FSB. They were also VIA products so best to stay away if there were other options. Wow were some of the Apollo Pro mobos I ran into just pure garbage..

The only reason I can see it being beneficial to have more memory bandwidth than FSB bandwidth is if there is something else in the system using that bandwidth. AGP perhaps. But I don't think I've ever seen a major benefit in any reviews. Even today with Core 2 CPUs, you don't get much by going outside of a 1:1 ratio.

2GB of PC800 RDRAM would've cost an unreal amount of money. I actually doubt that it even existed when RDRAM P3s first showed up. There were shortages of PC800 to the point that Dell loaded machines up with PC700 and made systems slower than old 440BX. The pricing is probably still ugly today. I remember pricing out some RDRAM in 2005 for a Pentium 4 running on i850 and it was still not worth upgrading it from the 256MB it came with. Super expensive yet because it never got cheaper, due to the lack of volume once Intel effectively dropped RDRAM when i845 DDR came out.

i840 makes more sense than i820 though because it's a multi-proc chipset and the CPUs have to share the RAM.

both the i820 and i840 are multi cpu chipset...my Dell systems are existing proof 😀

I wonder if Vogons user MartinC (the author of the O.P.) would like to do one test on his "dream machine"? I am testing exactly the same mobo (a Gigabyte slot 1 440BX, model GA-6BXC, revision 2.0), with the same CPU (1 GHz, 100 FSB). I am also using PC133 SDRAM. It's CL2, and the BIOS settings are set to CL2. The mobo seems stable, but not running 3DMark 2001 s.e. It always fails at the beginning of the "nature test". If I remove the 1 Ghz CPU, and replace it with a slower CPU (eg P3 600 Mhz), this test passes. I notice in the downloadable .pdf manual, where it explains how to set the mobo's CPU speed, it gives examples up to 950 Mhz, but not any faster. BTW, I'm using a Ti4200 AGP graphics card, with the nVidia 45.xx driver. And Windows 98 s.e.

prophase_j wrote:

Can you tell me what your AGP aperture is set to? Also, in your BIOS, what do you have "PNP OS Installed" set to?

MartinC wrote:

The AGP aperture size disapears from the BIOS page once I start scrolling down, no idea why. It's possible it started doing this when I updated the BIOS, can't remember.

That's interesting. I flashed the mobo's BIOS to the very latest beta BIOS, and I also see this "phenomenon" - that is, when you go to the "Chipset features" BIOS settings page, you see -

AGP Aperture Size (MB) : 64 abled

Please notice the "garbled mistake" at the end. When you position the cursor over this "corrupted line", you then see this -

Memory Hole At 15M-16M : Disabled

I wonder if this mobo is not 100% happy running this 1 Ghz 1.7 volt CPU? All other parts of this test system have been tested, and are OK. However, I cannot discount the possibility that the CPU is defective. The top part of its plastic casing is broken (damaged in transit), and consequently I have to keep the upper heatsink section pressed against the CPU using an elastic band.

I notice on the mobo (and in the manual), there is a "turbo mode" jumper. Currently, it's set to "normal". I wonder if this 1 Ghz CPU needs some kind of "boost", and setting this jumper to "turbo" might help?

I reran 3DMark 2001 s.e. with a P3 600Mhz CPU, and it keeps crashing on the jeep test, with heavily corrupted graphics. It's possible the Gainward Ti4200 is defective. I removed it, and replaced it with an Abit Ti4200 using exactly the same 45.xx nVidia driver. 3DMark 2001 s.e. now works. (In fact, I ran 2001 s.e., then f.e., then 2000, then 99 Max, then 2001 s.e. again - No crashes.) Also, I reflashed the BIOS with the last official "F3" BIOS. Now, when I go to the BIOS "advanced chipset" page, I just see this -

Memory Hole At 15M-16M : Disabled

There is no "garbled/corrupted" line which "half mentions" the AGP aperture setting - it's not there at all now. Also, one other BIOS alteration I did was to set the SDRAM memory timing setting from CL2 to "Auto". After the 5 lots of 3DMark tests above worked OK with the P3 600 Mhz CPU, I removed this CPU and replaced it with the P3 1 Ghz CPU. I then reran those 5 3DMark tests: 2001 s.e., then f.e., then 2000, then 99 Max, then 2001 s.e. again. No crashes. I needed to switch the machine off for a bit, and when I switched on power again, I reran 4 3DMark tests yet again, and this time I recorded their scores:

3DMark 2001 s.e. = 6150
3DMark 2001 f.e. = 6092
3DMark 2000 = 6550
3DMark 99 Max = 7202, 13741

Other benchmarks:

Sandra (2002 Pro) CPU = 2652, 1344
Sandra multimedia = 5292, 6186
Sandra memory bandwidth = 710, 425

Pcpbench mode 103 (800x600x8) LFB = 74.8

Unfortunately, it seems likely that my early/initial tests were being impeded by a faulty Gainward Ti4200 graphics card. (BTW, the partially damaged 1 Ghz slot 1 CPU seems to be working fine.)

Inside the BIOS set up area, I set the SDRAM memory timing setting from "Auto", to "CL2". I then reran Sandra memory bandwidth:

Sandra memory bandwidth = 559, 429 (Was 710, 425 - see above)

This is odd. I press the F5 refresh button, and I get 468, 429. That's even more odd. I reboot the machine, and change the SDRAM memory timing BIOS setting from "CL2", back to "Auto" again. Sandra memory bandwidth = 761, 429. That's better!

Question: I would like to increase the mobo's FSB from 100 Mhz, to 112Mhz. Do I need to set the mobo's "turbo mode" jumper to "enabled"? (Please see the image in my previous post above. Thanks.)

I just got this mobo (Gigabyte GA-6BXC) working with a Powerleap adapter, running a 1.4 Ghz 100 FSB Celeron. At first, when I tried to use the PL, the BIOS POST message just froze. I then googled for: GA-6BXC powerleap, and found a reference to a special Powerleap Gigabyte BIOS called mr_6bxc_beta. I googled for that, and found this web link -

http://www.powerleap.com/downloads/bios/Gigab … r_6bxc_beta.exe

Amazingly, this download still works! (Curiously, I was unable to find this BIOS file by navigating the powerleap.com website.) It's interesting to note that the powerleap.com website has a special BIOS for the Asus P2B board. I've tried the PL adapter on a P2B board, but I didn't use (or seem to need) a special Powerleap Asus BIOS file.

I'll run some benchies on the Gigabyte mobo + PL soon..

About the "garbled/corrupted" BIOS option: I've run into this as well when I was testing a GA-6BXC not long ago. Seems to be a rather harmless bug in this BIOS version.

Another thing about this BIOS: There doesn't seem to be an option to turn ACPI support off, which leads to Win9x hogging an extra IRQ, making it quite impossible to squeeze a Terratec EWS64XL and a GUS PnP into the system and having them work at the same time. So I dumped the board and used an Epox BX3 instead. 😉

I've run some benchies on the Gigabyte board, with the Powerleap adapter + 1.4 Ghz Celeron. I've put the Asus P2B + "1.4 PL" scores next to them, for comparison. In the Asus P2B board, I used 1 stick of Infineon PC133 CL3 256MB SDRAM. In the Gigabyte board, I used 1 stick of "no name/brand" PC133 CL? 256MB SDRAM. Both boards used the same MSI nVidia Ti4200 graphics card (64 MB, 128-bit, NV25, 4x AGP). I set the Gigabyte's BIOS memory timing setting to CL2, and the board appears to be stable. The 3DMark results show that the Asus board is faster. I will change the Gigabyte's stick of SDRAM, and use the stick I used for the Asus board - which was an Infineon stick. However, it's only CL3.

Sandra memory bandwidth = 663, 430 (Was 761, 429 using the old Gigabyte BIOS, without the PL adapter+special BIOS)
Sandra CPU = 3858, 1883 (Asus P2B scores were 3850, 1879)
Sandra Multimedia = 7444, 9031 (Asus: 7440, 9026)

3Dmark '99 = 8916, 18363 (Asus: 9225, 18726)
3Dmark 2000 = 7811 (Asus 8039)
3Dmark 2001 fe = 7190 (Asus 7483)
3Dmark 2001 se = 7276 (Asus 7541)

Pcpbench LFB mode 103 = 92.9 (Asus 92.2)

Hmm, based on release dates, software support, and deprecation, I figure something like so:

81 pc
83 xt
84 jr+speech adapter
84 tandy
84 at covox speech thing
87 286 10 paradise plus svga cms innovation adlib mt-32 fb-01 covox speech thing
88 386 25 video seven v-ram svga cms innovation adlib mt-32 fb-01 covox speech thing
88 tandy sl
89 486 25 video seven v-ram svga sb1.0+cms innovation sound master cm-32l fb-01 covox speech thing
90 486 33 et4000 svga sb 1.5+cms innovation sound master cm-32l fb-01 disney sound source
91 486 50 et4000 svga sb 1.5+cms innovation sound master cm-32l fb-01 disney sound source sc-55
92 486 66 et4000 svga sbpro sb16 gus cm-32l fb-01 disney sound source sc-55
93 pentium 66 s3 928 pci svga sbpro sb16 gus cm-32l sc-55 mkii
94 pentium 100 et4000/w32p svga sbpro sb16 gusmax cm-32l sc-88
95 pentium pro 200 millenium sb16+db50xg guspnp cm-32l sc-88
96 pentium pro 200 mystique voodoo sb16+db50xg guspnp cm-32l sc-88st
97 pii 300 riva 128 voodoo2 sli sb16+db50xg guspnp vortex cm-32l sc-88st pro
98 pii 450 riva tnt voodoo2 sli sb16+db50xg guspnp vortex2 cm-32l sc-880
99 athlon 600 tnt2 ultra voodoo3 3000 pci sb16+db50xg guspnp vortex2 cm-32l sc-8850
00 athlon 800 geforce 256 ddr voodoo3 3000 pci sblive
00 athlon 1000 geforce2 gts voodoo3 3000 pci sblive
00 athlon 1200 geforce2 ultra voodoo5 5500 pci sblive
01 athlon 1333 geforce3 voodoo5 5500 pci audigy
01 athlon 1900+ geforce3 ti500 voodoo5 5500 pci audigy
02 piv 2200 geforce4 ti4600 voodoo5 5500 pci audigy
02 piv 2800 9700 pro voodoo5 5500 pci audigy
03 athlon 64 3200+ 9800 xt voodoo5 5500 pci audigy
04 athlon 64 3400+ 6800 ultra voodoo5 5500 pci audigy
04 athlon 64 fx-55(~4000+) 6800 ultra extreme voodoo5 5500 pci audigy

Trickiest part being sound card arrangements.

edit: s/ark2000pv/millenium & mystique/ trade some performance for higher res & bit depth

edit: s/paradise/video seven v-ram(headland hd208)/ The first vram based card, 2.885MB/s vs 1.139MB/s paradise plus(pvga1a) in graphics modes. For 320x200, that's 46 vs 18 fps. IBM VGA is 592k/s, & et3k 588k/s.

edit: et4000/w32 was late, coming out in 93 like the w32i.

edit: w32i not pci, go with one of the first pci cards. mode 13 only needs 4.48mb/s to match fps to refresh and 93 heavy hitters like doom were cpu bound, so choice between staying /w a 486 for better video or pentium for faster cpu, cpu gets it.

I used Memtest 86 to try and figure out which SDRAM stick(s) and also which mobo BIOS settings gave the fastest memory performance. I quickly realised that the "no name/brand" SDRAM stick was not very fast - the best speed I could get was 278. I then removed this stick, and replaced it with 2 single sided Infineon sticks. They are 128MB each, CL2. With the mobo's BIOS settings maxed out, Memtest reported the memory speed as 323. I thought that was good enough to rerun all the test again -

Sandra memory bandwidth = 721, 430 (Was 663, 430, using the other "no name/brand" SDRAM stick)

3Dmark '99 = 9293, 18667 (Was 8916, 18363; Asus: 9225, 18726)
3Dmark 2000 = 8287 (Was 7811; Asus 8039)
3Dmark 2001 fe = 7487 (Was 7190; Asus 7483)
3Dmark 2001 se = 7564 (Was 7276; Asus 7541)

Pcpbench LFB mode 103 = 93.1 (Was 92.9; Asus 92.2)

The Asus scores have been marginally beaten. But! when I tested the Asus board, I was not using CL2 SDRAM. It was CL3. Also, I don't remember fussing over the P2B's BIOS settings all that much. The good news is that using "maxed out" BIOS advanced chipset settings, with CL2 SDRAM sticks, and using a powerleap adapter + a 1.4 Ghz Celeron, all tests worked without a single problem.

I decided to try some OC'ing on this board. I did the following -

I set the mobo's FSB jumper to 112 FSB
I did not alter the mobo's "turbo mode" jumper. That jumper is still set to "default".
I did not alter the mobo's BIOS advanced chipset settings. These are still "maxed out". Eg CL2, etc.
I did not alter the powerleap's voltage jumper. That jumper is still set to "default".

Memtest passes, and also says that the memory speed is 361 MB/s, which I think is fast.

Sandra memory bandwidth = 855, 480 (Was 721, 430)
Sandra CPU = 4308, 2103 (Was 3858, 1883)
Sandra Multimedia = 8313, 10085 (Was 7444, 9031)

3Dmark '99 = 10410, 21202 (Was 9293, 18667)
3Dmark 2000 = 9248 (Was 8287; Asus 112FSB = 9078)
3Dmark 2001 fe = 8162 (Was 7487)
3Dmark 2001 se = 8263 (Was 7564)

Pcpbench LFB mode 103 = 104.3 (Was 93.1)

I decided to test another slot 1 Intel 440 BX mobo. This one is by MSI, and it's called MS6119. The PCB revision number is 1.2 BX. I've got a vague feeling this particular board is an OEM board. (I could be wrong of course.) It's a "plug and play" board - the CPU speed and FSB values are not set using dip-switches on the mobo. Instead, these values are selected inside the BIOS set up area. (There is a jumper on the mobo to set the FSB - to either 66 or 100.) The BIOS POST screen says that the BIOS is "W6119MJ V2.9 120100". The BIOS POST datestamp says 08/03/2000. In an attempt to keep to the spirit of the O.P., I have put a 1 Ghz slot 1 CPU in to this 440 BX board. The BIOS POST displays info about this CPU as: "Pentium III-MMX CPU at :02E Mhz".

I wanted to test a slightly different graphics card, and so I chose a Gainward Ti4200 NV28 card. Also, I wanted to test a different nVidia driver, and so I chose an old one: 29.xx. I am using 2 sticks of CL2 SDRAM, with a total memory value of 256MB. I "maxed out" the advanced chipset BIOS settings.

3DMark 2001 s.e. = 6581
3DMark 2001 f.e. = 6550
3DMark 2000 = 7399
3DMark 99 Max = 8039, 14239

These scores seem good to me. After these tests, I went to the BIOS set up area, and changed the FSB from 100 to 103. The BIOS POST message then displayed info about the speed of the 1 Ghz CPU as 34E Mhz.

3DMark 2001 s.e. = 6707 (Was 6581)

Then, I tried to increase the FSB from 103 to 112, but the BIOS POST screen became corrupted, and so I did not continue with that OC'ing configuration. BTW, Sandra says that this 1.7 volt CPU is: 1.66V.

5u3 wrote:

Another thing about this BIOS: There doesn't seem to be an option to turn ACPI support off, which leads to Win9x hogging an extra IRQ, making it quite impossible to squeeze a Terratec EWS64XL and a GUS PnP into the system and having them work at the same time. So I dumped the board and used an Epox BX3 instead. 😉

I just got the chance to mess about with one of these soundcards (EWS64XL) for the first time yesterday. I noticed written on the PCB, near the ISA slot connector section, it says "RTFM". Well, I did glance through it and look at the pictures. I wish my German was a bit better, but I did understand the overall gist of things. I understood the bits where it said "Windows 98" and "CD ROM", etc. 😉

I decided to test the ISA slot on my Abit KT7A board. After all, that is the primary reason why I am interested in these KT133A based "retro rocket" mobos. I plugged in the soundcard (with the front bay unit attached to it), and switched on power. To cut a long story short, I noticed that Window 98 was very unstable. (Please note: I did get it working in the end. Edit - Actually I didn't.) The problem was the allocation of resources. The EWS application software includes a utility called Control Panel. Inside this Control Panel app, there is a SET UP button. That allows you to set the card's resources. I set them to IRQ = 5 and 7, and DMA = 0 and 1. Previously, one of the IRQs was set to 11. (And my graphics card was "relagated" to using IRQ 10.)

Inside the mobo's BIOS, I set these options:

PNP OS Installed = Yes
Force Update ESCD = Enabled

Resources Controlled By = Manual
IRQ Resources:
IRQ-5 assigned to = Legacy ISA
IRQ-7 assigned to = Legacy ISA
(all other options = PCI/ISA PnP)

DMA Resources:
DMA-0 assigned to = Legacy ISA
DMA-1 assigned to = Legacy ISA
(all other options = PCI/ISA PnP)

The system appears stable. Inside Win98, I run the "DOS command prompt box" and test Doom. I can hear funky tunes, and also SoundBlaster effects. I notice inside the MS System Information utility, I've only got IRQ 9 free. IRQ 10 has got something called "SCI IRQ used by ACPI bus" assigned to it. I also have IRQs 3, 4 and 6 free, but that's because I have disabled both COM ports, and also disabled the FDD.

Unfortunately, when I was typing this message above, the MS System Information utility froze. I don't think the system is stable after all. The problems seem to occur after I run the "DOS command prompt box" inside Win98 (to test a DOS game), and then exit back to the win98 desktop. One other thing I noticed - inside dxdiag, when I test the sound effects, I usually hear a clicking sound (in the right headphone) just before the birds begin to twitter.

Sounds like a difficult quest you took. Hope you've succeed, I would love to have that as an alternate computer.

Hey guys!!! I have not booted up this rig in a year, it's at my mum's (mom) house 1500KM away but I did a few more finishing touches to it, I can say it's probably a perfect reflection of what would have been a dream machine in the late 90's

I changed the case to something more reminiscent of the era and the software selection has been updated also.

Christmas is near so when I go up I will try to get some new pics to add