VOGONS


First post, by paradigital

User metadata
Rank Oldbie
Rank
Oldbie

Hello!

I've got a Pentium 4 (Northwood) Shuttle SB51G here that is based on the Intel 845GE and ICH4 chipset, it's running a Pentium 4 2.8GHz (133fsb/533bus) , 2x512MB of DDR 333, and a 40GB IDE (ATA100) Maxtor drive with Windows XP SP3 on it (no unofficial patches). I was attempting to use this system as a benchmarking platform for a few GPUs I have here (only for curiosities sake), planning to compare the GeForce 2 Ti, GeForce 2 Pro, GeForce 2 MX400, GeForce 4 MX460, Radeon 9200, and the PowerVR Kyro II (Hercules Prophet 4500).

After finishing the nVidia cards (unfortunately), I then plopped the Kyro II in, only to be greeted by a machine that wouldn't even power on, let alone POST or boot. I found that if I powered on the board whilst simultaneously pushing the card home into the AGP slot then the machine would POST/boot after a reset (using the reset switch) and from that point would happily reboot as many times as I wanted (but wouldn't power on again if the machine powered off). This behavior for an undamaged card and board is odd, something I'd usually attribute to a dead-short, but there isn't one (or at least, not one I could find with a multimeter). There were also some odd things happening in the BIOS, strings cut short (for example the CD-ROM drive was being named "H" during detection), and the error message that usually warns me of no A: drive was appearing at the top of the display, overlapping the memory count, when it's usually at the bottom of the screen!

Regardless, the card would happily boot to Windows and install drivers, this is then where things get worse 🙁 No accelerated 3D application would run at all (or at least wouldn't run for more than 10-15 seconds in the case of whatever application I ran first, as soon as the machine was fully booted to Windows, after a single 3D application had run for 15s, nothing else would run again until a reboot). Quake II, Quake III Arena, Unreal Tournament GoTY, 3DMark2000, and 3DMark2001SE were all tried, with both Quake games eventually going to a blank grey screen, Unreal GoTY turning transparent (as in I could see mostly the Windows desktop, but not interact with it until I CTRL-ALT-DEL into task manager). Oh and for what it's worth I tried all 3 driver versions that were greater than v2.0 (didn't really want to try the older ones as they don't have EnTnL).

I did some quick Googling but turned up relatively little (or old dead links that the wayback machine didn't have cached), though I did stumble across the AGP 4x mod. Having not a lot to lose I performed the hardware mod for the AGP 4x compatibility (0 Ohm resistor on R168) and now the Shuttle would power on with the card installed, and no BIOS glitches either! Booted into Windows but the 3D errors remain.

The only other AGP machine I had to hand was my Baby-AT Super Socket 7 system (AMD K6-III 450@550, MVP3 chipset, Windows 98SE) which is very picky on AGP cards usually, but the Kyro II was fine! Into Windows and whilst the performance was poor (which I'm attributing to the lack of CPU/platform grunt), it was flawless in terms of compatibility, stability and display quality.

Sorry for the wall of text to this point, but to save me from having to run 4 card's worth of benchmarks again, I was hoping one of you lovely ladies and gents could perhaps shed some light on potential BIOS tweaks to make, or if you have come across any similar issues or compatibility problems with the Kyro II. I did have a Kyro II back in the day and don't recall having any problems with platform compatibility, but I can't for the life of me remember what platform I had the card in.

Thanks in advance 😀

Reply 1 of 9, by Doornkaat

User metadata
Rank l33t
Rank
l33t

This sounds like typical Kyro II AGP2.0 weirdness.
The chip identifies as AGP 2.0 compatible but 4x is deactivated and since the AGP 2.0 indicator pin isn't wired to ground some boards assume a wrongly keyed card and refuse to work as a safety measure.

Now I assume 'hotplugging' the card forced 4x mode. The AGP 2.0 mod circumvents the safety, board and card negotiate AGP 2.0@2x and all goes well.
Avain: pure assumption but it would explain what you described.

Reply 2 of 9, by paradigital

User metadata
Rank Oldbie
Rank
Oldbie

Thanks Doornkaat, that does at least explain the lack of power-on when the card was inserted, however I'm still not sure on the stability issues in 3D modes when using the I845GE system. I'm going to try the Kyro II card on my nForce2 Ultra board when that arrives, and can try it on a Via Apollo Pro 133A (dual S370) board if I go drag one out of the loft.

Reply 3 of 9, by Doornkaat

User metadata
Rank l33t
Rank
l33t

You can still force 4x in BIOS. If the system proves unstable again it would support (not prove) my theory.
If it keeps running stable we can probably dismiss it. 😅

Reply 4 of 9, by paradigital

User metadata
Rank Oldbie
Rank
Oldbie

Unfortunately the Shuttle doesn't have AGP speed control (or indeed any control other than aperture size) in the BIOS.

Furthermore, the card isn't stable in the Pentium 4 Shuttle (I845GE chipset) at all, I can't run any 3d application (2d is fine, as is software rendered 3d).

The card is 100% stable in my K6-III PC (Via MVP3) in OpenGL, Direct3d, etc.

Reply 5 of 9, by Doornkaat

User metadata
Rank l33t
Rank
l33t

Sorry, misread the first post. 3D stability issues are probably linked to the card not correctly supporting 4x mode.

Reply 6 of 9, by paradigital

User metadata
Rank Oldbie
Rank
Oldbie

So, even though I had no issues with the Kyro II in my Super Socket 7 system, I figured I’d try the voltmod for core GPU voltage as the card would work for a few seconds before freezing.

It only worked. So the voltmod plus the AGP 4x mod has made the card compatible with this I845GE based Shuttle!

Here's the two hardware mods I had to do to the Kyro II to get it to play ball in the Shuttle SB51G Pentium 4 machine.

kyro2-mod1.jpg

As you can see, there were two areas of the card that needed modifying. R168 near the fan header is bridged to ground (bridges pin A2 on the AGP connector to ground, this tells the motherboard that the card is AGP 4x capable) using a 0 Ohm resistor, and R37 near the voltage regulator was given a 20KOhm resistor in order to raise core voltage from 2.08v to 2.5v.

kyro2-mod2.jpg

These things are bloody small! Sure I "could" have used a blob of solder or a bodge wire, but I wanted the card to look somewhat factory and not messed with, this way it's using proper SMD components that look "right".

kyro2-mod3.jpg

This one is even smaller, R37 is actually in parallel with R39 here (R38 and R39 form a potential divider which is used as reference voltage for the voltage regulator). Dropping in a 20KOhm resistor on the previously blank pads for R37 drops the resistance of R39, and raises the reference voltage up a tad. This results in 2.5v core rather than the standard 2.08v. Again, could have used a radial resistor and soldered directly between legs 3 (ground) and 5 of the voltage regulator, but this would look ugly as balls, and certainly not passable as "stock".

Hot-air rework stations are a godsend for stuff like this, both the 0Ohm and the 20KOhm were pulled from various bits of dead hardware using the hot-air station, I certainly don't keep such small SMD components lying around!

Reply 7 of 9, by 2Mourty

User metadata
Rank Member
Rank
Member

Does modifying the core voltage from 2.08V to 2.50 V have the potential of shortening the life expectancy of the gpu?

Reply 8 of 9, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

I think it's life expectancy went from an impending date with a hammer to running for a year or two, so assuming it had only hours more to live, this is probably in the neighborhood of a 60,000% life extension.

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 9 of 9, by paradigital

User metadata
Rank Oldbie
Rank
Oldbie
BitWrangler wrote on 2021-05-24, 04:51:

I think it's life expectancy went from an impending date with a hammer to running for a year or two, so assuming it had only hours more to live, this is probably in the neighborhood of a 60,000% life extension.

Indeed. However it still runs as cold as anything (the Kyro GPUs simply don't get hot, the HSF was for show more than anything as people were only buying gaming cards that *looked* like gaming cards by this point!), so I don't expect a massive issue. It's still happily running (when I use the P4 shuttle) and performing the same as ever.