VOGONS


Reply 20 of 56, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

Currently investigating possibilities for two boards, D101GGC and P4M890M, both 775. The 890 apparently has 98 drivers for SATA. Audio for DOS looks iffy unless one of the older ones of these work... https://drivers.eu/Chipset/VIA/Audio ... note, older Via P4 chipsets may have more luck
Then the ATI board has realtek which might need some hoop jumping .. https://msfn.org/board/topic/178295-audio-dri … comment-1161697
However, I have a CT4810 that should work out if those don't. I had/have ALS4000s somewhere which I remember playing DN3D and Quake fine on in DOS and some Glide stuff, but those require a hunt.

Onboard graphics on the D101 might be good enough, as the ati x200 up are based on 9600 cores and have win98 drivers and good enough retro applications unless you wanna whine about table fog etc that 80% of 3D cards from the era didn't find it worth the bother of supporting either. Not sure about the Unichrome on the Via though, even if it's "fully compatible" with older Savage... that doesn't really get you great compatibility 3d wise, some stuff will be okay. It has PCIe not AGP, but in the stash there are further x??? series ATI cards, x600, x1300 that may pan out better, or I can whack a PCI GF6200 in it, which won't be spectacular, but will function.

CPU, not much P4 selection in 775, got a 2.6 and a PentiumD 945, which is maybe a little much, and Celerons breeding like cockroaches... think there's a 3.2 in a Dell somewhere. (Dell has a lot of wifeys old stuff on and needs to remain fairly unmolested "for reasons"... maybe could get away with swapping CPU out)

Though also maybe I should consider the MS7005, which needs it's heatsink retention crap fixing, s478.. AC97 sound so might end up screwing with http://www.georgpotthast.de/dossound/ or similar solutions, if manufacturer drivers not available (realtek?) AGP, for which I might not have anything very good immediately available, maybe a 9250 or a GF3... gf4x00s are hiding on me.

Edit: The MSI P4M890M I am going to remove from consideration upon review of data included in the thread elcrys posted below. It's younger sibling, the MSI P4M900M which is gonna be highly similar, appears to respond poorly to throttling. Whether this is MSI or VIA's fault is not yet clear, other VIA P4Mxxx based boards should be tested.

Last edited by BitWrangler on 2022-01-22, 18:45. Edited 1 time in total.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 21 of 56, by TrashPanda

User metadata
Rank l33t
Rank
l33t

775 isnt a bad option if you can find the right 98se drivers, I mean its getting towards the point where a lot of the 98 drivers will be sketchy but should work well enough if you dont push boundaries too much 😉

The x700/x1300 (since they share similar GPU cores) PCIe GPUs should work well enough under 98se since their PCIe drivers should be fairly kind to 98se's lack of non dodgy PCIe awareness if you stick to late but not final 98se catalyst drivers.

Might want to stick to a non HT/dual core P4 tho, if you have a 775 Celery that is single core it would be better for a 98se rig.

Reply 22 of 56, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

775 isnt a bad option if you can find the right 98se drivers, I mean its getting towards the point where a lot of the 98 drivers will be sketchy but should work well enough if you dont push boundaries

LGA775 had abundant amount of AGP options with Intel and VIA chipsets

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 23 of 56, by TrashPanda

User metadata
Rank l33t
Rank
l33t
The Serpent Rider wrote on 2022-01-21, 20:06:

775 isnt a bad option if you can find the right 98se drivers, I mean its getting towards the point where a lot of the 98 drivers will be sketchy but should work well enough if you dont push boundaries

LGA775 had abundant amount of AGP options with Intel and VIA chipsets

I own quite a few Asrock solutions with the AGP/PCIe DDR1/2 hybrid boards, I buy them up on eBay when I see them cheap enough*, thankfully Asrock still has all their 98se drivers up. (*Mostly revisions/models I dont have)

The main issue with 775 AGP and Pentium 4 is there is only a small gap of time where a good amount of Celeron/P4 models were still supported in the BIOS and AGP was still being included on boards, I find that a ton of Celeron/P4 model support was dropped when Core 2 hit the scene and AGP was replaced by PCIe. (Mostly the slower/low FSB Single core models were dropped then later all the slower HT models, not all boards did this but a lot did especially with newer BIOS updates)

It was also around this time that 98se drivers started to get dropped too, so while you can find boards finding 98se drivers can be a bit harder especially for Intel boards and more obscure brands like Foxconn.

Reply 24 of 56, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

I am a little mired in analysis paralysis...

On the one hand, there's the pushing the boundaries approach and working out all the kinks in getting a later 775 board like the P4M890 working smooth, on the other hand there's the "common sense" of using what's maybe a better board for 98 with better native support, one of the socket 478 systems, then borrowing additional hands, on the third hand there's not wanting to "rob" components at the top end that I'd rather use in C2D systems and on the fourth hand, at the lower/older end, using components I'd rather use for Tualatin or Athlon. On the fifth hand, wasn't part of the justification being that P4s are cheap so I should be doing this with "nothing" parts to illustrate the possibilities of budget retro in 2022.

Anyhooooooo, CPUs kicking around include a 1.7 willy, ewww, a 1.8 woody, meh, a Celeron 330 hmm, and I think I know where a 2.4 Northwood is. Now I'm gonna sound like Doc Hudson to Lightning McQueen when I say, "You gotta go faster to go slower." meaning I wanna probably bus overclock and have a CPU running around 3 ghz (Mainly because I don't have any high FSB CPUs to bring down) because looking at pixel_workbench's initial L1 disable results on a 2.8 (First post, expand the first quote) I think I wanna have 533,667 or 800 possibilities to hit more in the DX2/66 range. In theory, CPU shouldn't be getting real hot when L1 crippled even if overclocked quite a bit, but I will be seeking a stable and well cooled overclock anyway for windows.... IF it turns out that the ACPI fiddling throttle as well as L1 disable allows fine tuning of 486 level speeds and a little below, bearing in mind leileilol's experience is that level 6 is probably as much as it takes, then maybe it stays at whatever FSB allows that. It would seem that maybe FSB on minimum, L2 disable, and throttle on 6 would maybe give Pentium II kinda performance.

--------------------------------------------------------------------------------

Thread with info on ISA-less sound options... Does a Tualatin build have to give up decent MIDI support?

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 25 of 56, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

Oldsflash...

It's like a newsflash, but for recently rediscovered retro stuff 🤣 Some Prescott CPU may support a lower 14x multiplier in some motherboards. In some cases this may be forced. It's a power protection feature, when 115W chips are stuffed into 95W boards. http://www.xtremesystems.org/forums/showthrea … ltiplier-UNLOCK!

Also, some early production Northwood may have been downward unlocked down to 8x multi. I don't have good references for that yet, mostly finding "that one guy" forum posts on it. This appears at the moment to only be "A" models, so designated because they were at the speed grades of previously released willamette cores, so 1.8A, 2.0A... ... I vaguely remember chatter about this back in the day, something along the lines of were some early production left unlocked, or was the locking method unreliable and only x/10 CPUs were actually unlocked. Kinda difficult to figure that out sometimes with Intel as they kind of build up several months of production before official release and "open the floodgates" meaning you get a multi month spread of early CPUs "new" on the market at the same time.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 26 of 56, by TrashPanda

User metadata
Rank l33t
Rank
l33t

PRESSHOTT !!!!

not the CPU to use if you want to run a high overclock 🤣, I remember having a ton of fun with one a few years back seeing how far I could push it.

It got quite toasty even with a good cooler on it.

Reply 27 of 56, by elcrys

User metadata
Rank Newbie
Rank
Newbie
mockingbird wrote on 2022-01-20, 16:50:

Seeing as how the multiplier-unlocked engineering samples and extreme editions are unobtanium, a VIA + P4 combination seems the best choice for CPU throttling.

With throttle, you get 16 different speed reduction options with VIA chipsets, plus the ability via switch to not toggle the L1 (so 32 combinations, actually), so setmul is of little value for the P4 compared to throttle.

I have tested MOSLO and THROTTLE, but CpuSpd beats them both:
CpuSpd - A Hardware Based CPU Speed Control Utility for DOS/Win9X Retro Gaming

For my P4 Northwood on i845 it can throttle via ACPI in 8 steps, then via ODCM in another 8 steps and it can disable both caches (not possible in BIOS). I was also hoping that I will be able to use Speedstep, but no luck. My complete setup:

BCM BC845DL
Intel Mobile Pentium 4 2,3 GHz SL789 @ 1,6 GHz
512 MB DDR
nVidia GeForce4 MX440 128MB (128bit) AGP DVI XpertVision - VBE 3.0 compatible for changing refresh rate in DOS
3Dfx Voodoo 2 12MB PowerColor
NEC D720101GJ USB 2.0 PCI
ESS AudioDrive ES1869F ISA with wavetable ES981P
Transcend SSD510K 64GB Industrial SSD
IDE CD-ROM

Reply 28 of 56, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

Awesome, we muddle along thinking "all these speed utils do more or less the same thing". So I musta not really taken notice of that. It seems to be one that combines multiple techniques, so a one stop shop for everything you can do, fantastic!

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 29 of 56, by pixel_workbench

User metadata
Rank Member
Rank
Member

Turns out that CPUSPD can disable caches on a P4 completely. Below are some results from a quick test I did:

P4 2.53GHz Northwood 533MHz FSB, Aopen MX46 (Sis 651), DDR-333, Radeon 9550
Test results are for 3dbench 1.0c / PCPbench / Doom (fps)

Normal:
636.6 / 481.7 / 154.3

Using "cpuspd cd":
24.7 / 6.9 / 10.5

I expected the 533MHz FSB to make a bigger impact with the caches disabled, this is only marginally better than my P4-2.8 on 400MHz FSB result. But at least we now have a reliable way to disable the caches.

My Videos | Website
P2 400 unlocked / Asus P3B-F / Voodoo3 3k / MX300 + YMF718

Reply 30 of 56, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

Thanks, hard to know whether the 300mhz clock drop held it back, or the 133Mhz bus jump didn't do much.
______________________________________________________________

Found some handy Pentium 4 chipset references...
https://flylib.com/books/en/4.57.1.47/1/
https://flylib.com/books/en/4.57.1.48/1/

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 31 of 56, by mockingbird

User metadata
Rank Oldbie
Rank
Oldbie
elcrys wrote on 2022-01-22, 18:23:

I have tested MOSLO and THROTTLE, but CpuSpd beats them both:
CpuSpd - A Hardware Based CPU Speed Control Utility for DOS/Win9X Retro Gaming

For my P4 Northwood on i845 it can throttle via ACPI in 8 steps, then via ODCM in another 8 steps and it can disable both caches (not possible in BIOS). I was also hoping that I will be able to use Speedstep, but no luck. My complete setup:

Fantastic, thanks for the info. I am contemplating a very similar system, except with a Yamaha PCI and SBLink... What's the advantage of the Mobile P4 as opposed to an ordinary one?

mslrlv.png
(Decommissioned:)
7ivtic.png

Reply 32 of 56, by greasemonkey90s

User metadata
Rank Member
Rank
Member

hmmm i thought of using p4-m but they default to a 12 multi because of speedstep or whatever. unless theres a dos utility to control speedstep maybe? how useable though i have to test.

possiblie using sstep 1.0 on this link.

http://falcosoft.hu/dos_softwares.html

Reply 33 of 56, by mockingbird

User metadata
Rank Oldbie
Rank
Oldbie
greasemonkey90s wrote on 2022-01-23, 02:38:

hmmm i thought of using p4-m but they default to a 12 multi because of speedstep or whatever. unless theres a dos utility to control speedstep maybe? how useable though i have to test.

possiblie using sstep 1.0 on this link.

http://falcosoft.hu/dos_softwares.html

I read on this forum that the "...speedstep pin is not connected in desktop boards", so you're stuck with the low state on desktop boards (P4-M speedstep only has two states, low and high, supposedly). This is not a microcode issue, both the desktop and mobile 400Mhz P4 northwoods use CPUID 0F29h.

mslrlv.png
(Decommissioned:)
7ivtic.png

Reply 34 of 56, by elcrys

User metadata
Rank Newbie
Rank
Newbie

Unfortunately this falconsoft's SSTEP 1.0 doesn't work either. As mockingbird wrote, it's probably some physical incompatibility with desktop boards for these early speedstep CPUs.

In any case the main advantage of Mobile P4 is power consumption. This SL789 I have is quite exotic - it's a hybrid between Mobile P4 (high TDP, heatspreader) and P4-M (low TDP, no heatspreader). This one has heatspreader, low TDP (35W in high mode) and defaults to 1.6 GHz in "battery" mode with only 1.2V. I think this CPU was originally (and exclusively) used in some Toshiba laptops.

Anyway, with full copper heatsink with low speed Noctua fan, I'm getting only slightly over 30 °C in Prime95 with the performance of high-end P3s. This is more performance than I would ever need for this mainly pure DOS/early W98 games build, although the CPU can be overclocked via FSB to approx. 2.6 GHz with the same low voltage.

Reply 36 of 56, by mockingbird

User metadata
Rank Oldbie
Rank
Oldbie
elcrys wrote on 2022-01-23, 11:40:

Anyway, with full copper heatsink with low speed Noctua fan, I'm getting only slightly over 30 °C in Prime95 with the performance of high-end P3s. This is more performance than I would ever need for this mainly pure DOS/early W98 games build, although the CPU can be overclocked via FSB to approx. 2.6 GHz with the same low voltage.

Yes, I suppose... But wouldn't the P4-M 2.0 be a better choice with the 20x multiplier (20 x 133 = 2660)... What FSB are you running at to get to 2.6? 110? How does that work with Vanilla Brookdale (i845), is the FSB independent of the AGP/PCI divider?

mslrlv.png
(Decommissioned:)
7ivtic.png

Reply 37 of 56, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++
Kahenraz wrote on 2022-01-23, 19:54:

I remember some manufacturers managing to fit a full size socketed Pentium 4 into their laptops for a time. I think it was Dell?

Several did I think because the mobile was a while appearing in quantity. At that point they were "only" 60W chips I think though.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 38 of 56, by elcrys

User metadata
Rank Newbie
Rank
Newbie
mockingbird wrote on 2022-01-23, 19:57:

Yes, I suppose... But wouldn't the P4-M 2.0 be a better choice with the 20x multiplier (20 x 133 = 2660)... What FSB are you running at to get to 2.6? 110? How does that work with Vanilla Brookdale (i845), is the FSB independent of the AGP/PCI divider?

As was discussed, with P4-M 2.0 you are stuck with 12x multiplier (low/battery state) on a desktop board.
For my CPU it's 16x and I used SetFSB for OC. BCM BC845DL has clock generator CY28323PVC and it has proper dividers for AGP/PCI up to 200 MHz in steps 66/100/133/166/200. FSB on 166 MHz seemed stable which equals to 2.6 GHz @ 1.2V. Luckily the only thing you can change in BIOS regarding OC is memory divider (here I set 1:1), so memory OC wasn't a problem even for 2 - 2 - 2 - 6 timings.

Reply 39 of 56, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

SiS had a tendency to implement every possible feature, hoping to broaden their chipset markets, so mayyyybe if motherboard maker hooked it up to the socket you might get more multi control on a SiS board, but I think still it only goes down to 8x then.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.