VOGONS


Reply 60 of 75, by PC@LIVE

User metadata
Rank Oldbie
Rank
Oldbie

I saw Phil's latest YouTube video, and it was just talking about the latest version of this program, well I have a number of PCs where I can try it, some have K6-2 CXT with multi 6X (i430TX/VX), others instead use FSB 95-100, and they are K6-III and K6-2+, but surely the performance for some is not the best, because for example, I forced a K6-2 CXT at 400 on a MB that sees it as K6-300, probably all the improvements of the CXT are ignored.
Then I read something 😨 in the previous messages, which I didn't quite understand, we're talking about a multi 6.5X, here I know the maximum is 6X, but having a 433 CPU in my collection, I have doubts that it was planned with FSB 66 and multi 6.5X, instead of the strange 96.2 with which it was put on the market.
I'll see if I can do some tests with the latest program, I currently have an M550 with a K6-233 on the bench, I don't know if I would have any improvement with that CPU, but I can easily upgrade with other K6-2, up to 400 - 450 MHz, beyond that I could get to 500 but I would have the PCI over 41 MHz, and I wouldn't want to have problems in the future for this.

AMD 286-16 287-10 4MB HD 45MB VGA 256KB
AMD 386DX-40 Intel 387 8MB HD 81MB VGA 256KB
Cyrix 486DLC-40 IIT387-40 8MB VGA 512KB
AMD 5X86-133 16MB VGA VLB CL5428 2MB and many others
AMD K62+ 550 SOYO 5EMA+ and many others
AST Pentium Pro 200 MHz L2 256KB

Reply 61 of 75, by analog_programmer

User metadata
Rank Oldbie
Rank
Oldbie

PC@LIVE, there's no such thing as x 6.5 multiplier on (s)s.7 platforms. I'm not sure if K6 family CPUs officially support x 6 multiplier (as they use x 2 multiplier motherboard's settings for it). K6-2 at 433 MHz (427.5 MHz ??) seems like weird thing, but can be achieved on some ss.7 chipsets through some weird bus frequencies using x 5 or x 5.5 multipliers. K6INIT actually provides x 2 multiplier for CXT core K6-2 (or newer K6-III, K6-2+/III+) CPUs, but I can't recall now if I successfully tried x 1.5 and x 1 multi through this software tool. If you have an working (s)s.7 system with K6-II CTX or newer K6 family CPU, just test these multipliers and give some feedback 😉

from СМ630 to Ryzen gen. 3
engineer's five pennies: this world goes south since everything's run by financiers and economists
this isn't voice chat, yet some people, overusing online communications, "talk" and "hear voices"

Reply 62 of 75, by PC@LIVE

User metadata
Rank Oldbie
Rank
Oldbie
analog_programmer wrote on 2025-01-13, 15:46:

PC@LIVE, there's no such thing as x 6.5 multiplier on (s)s.7 platforms. I'm not sure if K6 family CPUs officially support x 6 multiplier (as they use x 2 multiplier motherboard's settings for it). K6-2 at 433 MHz (427.5 MHz ??) seems like weird thing, but can be achieved on some ss.7 chipsets through some weird bus frequencies using x 5 or x 5.5 multipliers. K6INIT actually provides x 2 multiplier for CXT core K6-2 (or newer K6-III, K6-2+/III+) CPUs, but I can't recall now if I successfully tried x 1.5 and x 1 multi through this software tool. If you have an working (s)s.7 system with K6-II CTX or newer K6 family CPU, just test these multipliers and give some feedback 😉

Yes ok thanks ☺️
As for the multi 6.5X, I thought that AMD had actually designed them, and maybe 🤔 they gave up stopping at 6X, maybe they could be useful on MB with FSB 66 (or 75 83 Max), a bit like the K6-2 333, which can be used with both FSB 95 and 66, so for example the 433, it could ideally work at both 96.2 and 66 (having the multi 6.5X).
Maybe 🤔 they thought it was easier to go up with the FSB, rather than with the multi, but strangely they stopped at 100 MHz, even if the latest motherboards, it seems that there are two or three models (or a little more), did not have L2 cache on the MB, since it had now been inserted in the CPU with the K6+ versions.
I'll try the multi you suggested (1X or 1.5X), I hope 🤞 they work, I could run the CXT from 75 MHz up 🔝, it depends on whether the MB allows FSB from 50 or what would be the minimum, let's say that I converted most of the S.7 cards to AMD, while the older ones with minimum voltages of 3.3V (VCORE), I have them with Intel and Cyrix and AMD K5.
I hope 🤞 to have news by the weekend, and bring you good news 📰

AMD 286-16 287-10 4MB HD 45MB VGA 256KB
AMD 386DX-40 Intel 387 8MB HD 81MB VGA 256KB
Cyrix 486DLC-40 IIT387-40 8MB VGA 512KB
AMD 5X86-133 16MB VGA VLB CL5428 2MB and many others
AMD K62+ 550 SOYO 5EMA+ and many others
AST Pentium Pro 200 MHz L2 256KB

Reply 63 of 75, by mrdjey

User metadata
Rank Newbie
Rank
Newbie
Geoman wrote on 2025-01-13, 10:12:
IBM Aptiva 2168 / AMD K6 III+ ATZ / IBM PC DOS 7.0 gives me, while using DEVICE=K6INIT.EXE /auto in config.sys […]
Show full quote

IBM Aptiva 2168 / AMD K6 III+ ATZ / IBM PC DOS 7.0 gives me, while using DEVICE=K6INIT.EXE /auto in config.sys

Keep C fatal error: Unable to fixup segment references

running as an executable works, though I get a partial error:


ERROR | You are trying to configure too many MTRRs, maximum is 2!
ERROR | PCI/AGP FB detection failed! Skipping...
ERROR | Preparing automatic configuration

SUMMARY | some actions failed

after that, the computer locks up.

i tried K6INIT for the first time yesterday on a K6-2+ 570 running PC-DOS 7.1 but it just freezes my system without giving any output 🙁

AMD K6-2+ 570ACZ | Aopen AX59 Pro | Diamond Speedstar A90 | AWE32

Reply 64 of 75, by mrdjey

User metadata
Rank Newbie
Rank
Newbie

i was able to run K6INIT if i don't load EMM386, so i put it in my CONFIG.SYS before the EMM386 line, and that seems to work and it got me +20 FPS in Quake benchmark.
but still not able to run if from the command line to make cache of multiplier adjustments.

AMD K6-2+ 570ACZ | Aopen AX59 Pro | Diamond Speedstar A90 | AWE32

Reply 65 of 75, by Oerg866

User metadata
Rank Member
Rank
Member

Hello,

with EMM running it is not possible to run a lot of these priviledged CPU instructions, that's why I wrote it in a way that you can load it from CONFIG.SYS... If anyone with low level DOS systems programming knowledge can chip in - it may be possible to thunk those instructions out to some real mode code, much like how BIOS calls are wrapped; otherwise it is not possble and the crash is expected (I could maybe add detection for it to make it more graceful)

It's the same reason the program doesn't run from within Windows...

Best
Eric

Reply 66 of 75, by Oerg866

User metadata
Rank Member
Rank
Member
SteveC wrote on 2025-01-03, 14:05:

Just tried this on my Aptiva K6-2 450 and it is great - doubles the Quake and PCPlayer 640x480 performance! It gives some errors though so will need to experiment a bit. /auto only worked on a clean DOS mode, with sound drivers etc it crashes for me.
If this is of any interest, this is what I see when I run it...

Ah this is a very tricky case... It detects the voodoo card's BARs as VRAM that should be accelerated, and that in conjunction with the memory on your TNT1 makes for 3 potential MTRRs needed, so it doesn't know what to do and gives up.

I don't know how to handle this case better, because the tool can't know what you actually want to do. I could convert this error into a warning maybe, or stick to true VGA adapter LFBs only (i think it can be discerned via the PCI subclass code)...

Any input how the logic can be structured better is welcome.

EDIT: wait - you're running a voodoo 3 in parallel with it? Yeah, then that can't be done automatically, you need to set the MTRRs you want by hand, I don't see any way to handle that automatically 🤔

Chain-Q wrote on 2025-01-04, 12:38:
First of all thanks for doing this tool, I'm blown away by the performance increase I'm seeing, from most graphics stuff using […]
Show full quote

First of all thanks for doing this tool, I'm blown away by the performance increase I'm seeing, from most graphics stuff using the LFB. (Demos!)

However, I also have some problems. At least with version 1.1, on my system, I can't seem to be able to load it from CONFIG.SYS. I get a message: "Keep C fatal error: Unable to fixup segment references." (see the attachment; direct capture from the VGA out) and then the system is locked up. Same before/after HIMEM.SYS is loaded. Also when EMM386 is loaded, K6INIT dies immediately after pressing ENTER at the prompt. No messages. Even if I just do K6INIT /?, the executable just stops. Note that the two things are independent from each other. K6INIT dies from CONFIG.SYS, even if EMM386 is not loaded.

I'm using the 1.1 release binary from GitHub. I use PC-DOS 7.01 (a.k.a. PC-DOS 2000). But as far as I know, this should be just MS-DOS with some tweaking/tuning from IBM, so the compatibility should be very high. My system is a K6-III+/400@550, AGP Riva TNT 1, Acorp 5ALI61 mainboard, 512MB RAM.

Do anyone else see similar problems, or it's just my setup, as usual?

It is probably a bug within Geoff Chappell's KEEPC library, a implemention detail discrepancy in PC-DOS that trips it up. I can see if I can fix it manually but there is some wizardry in there that's way over my head....

Also since I rarely check this place for obvious reasons - please feel free to open bug reports on github (https://github.com/oerg866/k6init/issues), so I can respond and fix problems a bit faster, and it also helps me keep track of them 😀

Last edited by Oerg866 on 2025-01-18, 15:51. Edited 1 time in total.

Reply 67 of 75, by Chain-Q

User metadata
Rank Newbie
Rank
Newbie
Oerg866 wrote on 2025-01-17, 23:33:

Also since I rarely check this place for obvious reasons - please feel free to open bug reports on github (https://github.com/oerg866/k6init/issues), so I can respond and fix problems a bit faster, and it also helps me keep track of them 😀

As you wish. Opened two bugreports on GitHub for the EMM386 lockup issue (it should still exit gracefully, IMO) and for the CONFIG.SYS load issue have.

AMD K6-III+/400@550, 512MB, Diamond Stealth V550, Gravis Ultrasound MAX 1MB, Sound Blaster AWE64
C= Amiga 2000, Motorola 68060@50, 128MB, Cirrus Logic 5446A 4MB, ESS 1688 + TI 486sxlc2@50 8MB
Mastodon

Reply 69 of 75, by SteveC

User metadata
Rank Member
Rank
Member
Oerg866 wrote on 2025-01-17, 23:33:
Ah this is a very tricky case... It detects the voodoo card's BARs as VRAM that should be accelerated, and that in conjunction w […]
Show full quote
SteveC wrote on 2025-01-03, 14:05:

Just tried this on my Aptiva K6-2 450 and it is great - doubles the Quake and PCPlayer 640x480 performance! It gives some errors though so will need to experiment a bit. /auto only worked on a clean DOS mode, with sound drivers etc it crashes for me.
If this is of any interest, this is what I see when I run it...

Ah this is a very tricky case... It detects the voodoo card's BARs as VRAM that should be accelerated, and that in conjunction with the memory on your TNT1 makes for 3 potential MTRRs needed, so it doesn't know what to do and gives up.

I don't know how to handle this case better, because the tool can't know what you actually want to do. I could convert this error into a warning maybe, or stick to true VGA adapter LFBs only (i think it can be discerned via the PCI subclass code)...

Any input how the logic can be structured better is welcome.

EDIT: wait - you're running a voodoo 3 in parallel with it? Yeah, then that can't be done automatically, you need to set the MTRRs you want by hand, I don't see any way to handle that automatically 🤔

Chain-Q wrote on 2025-01-04, 12:38:
First of all thanks for doing this tool, I'm blown away by the performance increase I'm seeing, from most graphics stuff using […]
Show full quote

First of all thanks for doing this tool, I'm blown away by the performance increase I'm seeing, from most graphics stuff using the LFB. (Demos!)

However, I also have some problems. At least with version 1.1, on my system, I can't seem to be able to load it from CONFIG.SYS. I get a message: "Keep C fatal error: Unable to fixup segment references." (see the attachment; direct capture from the VGA out) and then the system is locked up. Same before/after HIMEM.SYS is loaded. Also when EMM386 is loaded, K6INIT dies immediately after pressing ENTER at the prompt. No messages. Even if I just do K6INIT /?, the executable just stops. Note that the two things are independent from each other. K6INIT dies from CONFIG.SYS, even if EMM386 is not loaded.

I'm using the 1.1 release binary from GitHub. I use PC-DOS 7.01 (a.k.a. PC-DOS 2000). But as far as I know, this should be just MS-DOS with some tweaking/tuning from IBM, so the compatibility should be very high. My system is a K6-III+/400@550, AGP Riva TNT 1, Acorp 5ALI61 mainboard, 512MB RAM.

Do anyone else see similar problems, or it's just my setup, as usual?

It is probably a bug within Geoff Chappell's KEEPC library, a implemention detail discrepancy in PC-DOS that trips it up. I can see if I can fix it manually but there is some wizardry in there that's way over my head....

Also since I rarely check this place for obvious reasons - please feel free to open bug reports on github (https://github.com/oerg866/k6init/issues), so I can respond and fix problems a bit faster, and it also helps me keep track of them 😀

Yeah it's a Voodoo 3 card. It has onboard ATI AGP but I'm not using that, there's no TNT1 in there!

YouTube: https://www.youtube.com/c/StevesTechShed
Twitter: https://twitter.com/SteveTechShed

Reply 70 of 75, by tauro

User metadata
Rank Member
Rank
Member

I'm testing it with a K6-2 450@533 on a GA-5SMM (SiS 530 + SiS 630).

With /auto it doesn't improve performance, but /auto /vga does work.
Is this the best way to use the software?

Previously I used to run:
K6WCX E4000000 64 A0000 1

QUAKE 640x480 goes from 27.2 FPS (K6WCX) to 31.2 FPS (K6INIT /auto /vga).

Reply 71 of 75, by RoberMC

User metadata
Rank Newbie
Rank
Newbie

I tried this program and... I am experiencing a performance boost that exceeded my expectations by a lot, an abnormal performance boost, and my system was already optimized with write combining enabled and all the stuff, so I thought I would do some investigation...

My system:
AMD K6-III+ 575Mhz (115*5)
Asus P5A-B (Ali Aladdin V)
Voodoo 3 3000 AGP
256MB RAM

This motherboard has options in bios to enable write combining and frame buffer stuff, so i wasn't using any of the other WC enabler tools, like K6WCX, as they made no measurable difference when I tested them.

I was getting 29.4 fps in Quake 640x480 dosbench test, which is a bit more than what Philscomputerlab was getting in his review video of this tool, so I thought this tool would not do anything for me, another K6WCX tool but fancier... oh, I was so wrong...

With the exact same configuration I executed K6init /Auto and now I am getting 40.1 fps in Quake 640x480, that's a 37% increase, and this made Quake2 (Q2Dos port) playable over 30fps at 640x480!.

Here's some test results from dosbench with Bios tweaks and K6INIT:
Chris's 3D bench 640x480 (4) : 212.7 Score
PC Player 640x480 (6): 81.0 fps
Quake 640x480 (e): 40.1 fps

And here the same tests with K6INIT only:
Chris's 3D bench 640x480 (4) : 136.8 Score
PC Player 640x480 (6): 49.0 fps
Quake 640x480 (e): 28.6 fps

This lead me to think there's more room to improve what this tool is doing to squeeze even more the k6, at least on ALi chipset.

I remember years ago I managed to trace what the bios in this MB is doing with chipset registers and used pciset tool to replicate the behavior in order to transfer this tweaks to motherboards lacking this bios options, it is flipping some bits in 3 southbridge registers 84 85 and 86. South Bridge is Ali M1541, Vendor 10B9, Device 5243

Bios configuration by default (wc fb off in bios):
PCISET 5243 10B9 84 1111 0000
PCISET 5243 10B9 85 1111.1111 0000.0000
PCISET 5243 10B9 86 1111 000

Bios tweaked configuration (wc fb enabled):
PCISET 5243 10B9 84 1111 1111
PCISET 5243 10B9 85 1110.0000 1110.0000
PCISET 5243 10B9 85 0000.1111 110
PCISET 5243 10B9 86 1111 111

I tried this in other Aladdin V motherboards which do not have this options in bios, like the GA-5AX and Diamond Mironics C200, and got about the same performance as with the Asus and now with K6INIT the performance went over the roof. This bit masking and flipping thing makes my head hurt and probably the same results can be obtained with less commands, i'm not an expert. The bits to set can be deducted from the PCISET syntax.

The idea would be to integrate this chipset tweaks into the k6init tool if that's possible and... boom, more free performance for everybody!

Also, as a feature request, I would like the tool to have an option to output less text, a line or two would be great to know it's working as I like to have my boot screen less bloated, I use it in config.sys, so >NUL doesn't work.

I hope this post is helpful.

Reply 72 of 75, by 1541

User metadata
Rank Member
Rank
Member
tauro wrote on 2025-01-23, 11:52:

With /auto it doesn't improve performance, but /auto /vga does work.
Is this the best way to use the software?

"/auto" sets

"- Finds linear frame buffer memory and sets up write combining for it\n"
"- Enables Write Allocate for the entire system memory range\n"
"- Enables Write Ordering except for uncacheable / write-combined regions\n"

but excludes unsafe features that might freeze the machine, so you need to set "/vga" explicitly.

"/vga" sets

"Enables Write Combine for the VGA memory region"
"(A0000-BFFFF). WARNING: Potentially unsafe."
"You MUST NOT use this memory region for UMBs."
"This parameter is equivalent to /wc:0xA0000,128,1,0"

💾 Windows 9x resources (drivers, tools, NUSB,...) 💾

Reply 74 of 75, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++
marxveix wrote on 2025-02-10, 19:39:

Its not directly related to k6, but is it possible to enable similar boost with slot a k7 or socket a k7?

Along with being more or less a clean sheet design and getting a lot of DEC Alpha features, the Slot and Socket A CPUs were on motherboards specifically designed to use them, all the magic was implemented in chipset and BIOS. This is in contrast to the more "multipurpose" recycled socket 7 nature of the platforms for K6-x.

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 75 of 75, by Falcosoft

User metadata
Rank l33t
Rank
l33t
marxveix wrote on 2025-02-10, 19:39:

Its not directly related to k6, but is it possible to enable similar boost with slot a k7 or socket a k7?

K7 uses the same memory type range registers as Intel P6+ so you can use the same tools to set write combining for linear and VGA frame buffer:
https://www.philscomputerlab.com/dos-graphics-boost.html

But in case of setting WC for the VGA frame buffer there is a quirk:
MTRRLFBE and AGP/PCIe cards in DOS

Website, Facebook, Youtube
Falcosoft Soundfont Midi Player + Munt VSTi + BassMidi VSTi
VST Midi Driver Midi Mapper