VOGONS


Interposer for mobile Pentium MMX Tillamook

Topic actions

Reply 122 of 165, by Thermalwrong

User metadata
Rank Oldbie
Rank
Oldbie
Sphere478 wrote on 2023-11-27, 00:16:

Indeed, I concur with your conclusion based on measurements

What readings did you have by the way? How many ohms?

Just doing beep tests 😀 So I can't tell you how many ohms but it would've been less than 50 to make it beep.
Regarding the alternative multiplier settings, that's a maybe - the laptop's been put back together and is working fine for now. Cutting traces to re-route them on this board that can't be replaced isn't something I want to do.

edit: Hmm, realised that getting to the topside takes 6 screws and a minute to look at. Checked out both the Portege 3010CT and 3020CT - the topside looks like this:

The attachment 3020ct-topside.jpg is no longer available

So perhaps I can try the different multiplier setting without too much work.

Reply 123 of 165, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Check my pcb projects I made a small pcb for adding dip switch to laptops

Edit:

Multiplier Bodge Board For laptops 486, Pentium, K6 , Cyrix.

It’s very small. Tap the necessary pads, run small boge wires and glue the board somewhere easy to get to.

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 124 of 165, by Thermalwrong

User metadata
Rank Oldbie
Rank
Oldbie

There is genuinely no room in this laptop for a separate PCB, it's too thin 😀 I'll try moving around the 0-ohm resistors tomorrow, shouldn't be too hard

Reply 125 of 165, by Thermalwrong

User metadata
Rank Oldbie
Rank
Oldbie

Hmm, not a positive result sadly. I've verified that I've got BF0 now hooked to GND and BF1 hooked to VCC3, I think this should result in BF2:0 being 1-1-0?

컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴?
System Speed Test Ver 4.78 Report file - created on 01-01-1990 01:07:13
컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴?

CPU is in V86 mode: No
Processor : Intel Pentium(r) MMX(tm) 266 MHz
Frequency via TSC : 266.61 MHz
CPUID (TFMS) : 0582 Codename: Tilamook (0.25um)
Feature : 008001BFh MMX(tm):Yes, IA SSE:No, IA SSE2:No
CPU speed index : 198.33

The attachment 3020ct-110-maybe.JPG is no longer available

For the sake of completeness, putting it to 1-1-1 resulted in 300MHz the same as 1-0-1 did:

컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴?
System Speed Test Ver 4.78 Report file - created on 01-01-1990 01:05:01
컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴컴?

CPU is in V86 mode: No
Processor : Intel Pentium(r) MMX(tm) 300 MHz
Frequency via TSC : 299.94 MHz
CPUID (TFMS) : 0582 Codename: Tilamook (0.25um)
Feature : 008001BFh MMX(tm):Yes, IA SSE:No, IA SSE2:No
CPU speed index : 223.13

Total memory size : 96 MB
Memory Bandwidth : 153.36 MB/s

-----------------
Now it's put back to stock configuration 😀

Reply 126 of 165, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

>Thermalwrong
Thank you for your experiements on 300MHz version. It just prove what I expected. So your version has enabled one more mul step to 4,5x but it's still capped and cannot get higher. I still think that intel disabled it hard way - on a silicon, as intel likes to do. May be there exist some engineering sample that is fully unlocked but it would be even much rare than this version.

BTW I read that such thin carbide drills are suitable only for professioal drilling machines that are perfectly centered and automated movings. They tend to break very easily when used in hand-driller, it's better to use normal HSS steel drills. But they lost sharpness soon...
For probing there are manufactured thin needle probes for multimeters...

Last edited by RayeR on 2023-11-28, 05:41. Edited 1 time in total.

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 127 of 165, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Well, darn.

Good try 😀

You could try all possible combos but I bet result will be the same.

Btw, you gotta be careful when doing this not to connect bf to both. Usually they are all on a high value pull up resistor and a low value pull down resistor is used to flip flop

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 128 of 165, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

I did some more experiments on Gigabyte GA-5AA. I looked in my CPU box and get AMD K6-II-300 for testing.
It runs nice at 83, 95, and 100MHz FSB. I measured the PCI CLK to see how divider is set and I got 33,3 / 31,6 / 33, 3 MHz so it works fine and PCI is not overclocked. At 100MHz I need to push Vcore a bit higher to 2,7V to run it stable at least for Quake test. Despite a huge difference in CPU score and mem/cache speed shown in Speedsys utility (K6 was virtually better) the real performance in DOS Quake was very close to Tillamook at the same frequency (fractions of FPS).
So I proved that MB itself doesn't have a problem to run at 100MHz FSB. Unfortnuatelly my Tilly sample is so poor that it can't. I was encouraged to push Vcore at 2,9V (the VRM maximum by jumpers) for a short time, Tilly booted but restarted immeditatelly I try to run some other DOS program from command prompt.
It still makes me wonder how huge gap the Tilly has between 4x83MHz where it run stable at Vcore 2,2V, I even tried at 2,1V and also hadn't problem at least in Quake (overclocked K6 did crazy things in Quake when not volted up enuf) while it cannot run stable at 4x95MHz even with such extreme overvolting (I tried by smaller steps, on lower Vcore it even doesn't POST or hangs after a few POST codes). Maybe I could try to put some 0402 100n caps at Vcore on interposer if it would have some effect?

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 129 of 165, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

You ran the tillamook core (vcc2) at 2.9v?

No, dude never do that. Too much.

A good tillamook will do 400core/100bus at 2.0v all day long basically all my ppga tillamooks will do that. (Air cooling)

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 130 of 165, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

Sure I know, just for test, less than a minute. I tried with smaller steps and it shows that inceasing Vcore helps getting from no-POST to boot but still not enough. And MB is not the cause. What are "all your Tillys"? Do you have also 233MHz parts and all works at 400? How many samples? Maybe I jus had mo luck for good piece or it may be the interposer lowers the voltage stability or signal integrity. First I will check with scope that there are not some excessive vdrops at power planes...

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 131 of 165, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Umm, let’s see, I have a number of tillamooks, golden tigers, golden warriors. But the ones I’m talking about are the ppga ones (same package as 233mmx p55c)

I believe I have like three 266 sl2z4 units, and two sl2z3 units and every time I have tried 400mhz they have worked flawlessly.

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 132 of 165, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

Ok, 2 samples of 233 are not much big set to conclude that all 233 should work at 400 and those 266 are better bin...

I checked the ripple and it doesn't look nice, over 300mVpp at VCC3 and 200mVpp at VCC2. So I soldered on some mix of SMD caps (10u, 100n, 10n) about 16 total on top side of interposer between PGA pins and around IHS. It helped to lower ripple at 2/3-1/2 of previous level. But it didn't improve stability much. I can start POST at lower Vcore than before but it still restarts after boot very soon even at 2,8V. When I disable both caches it seems to be more stable but perf. Is useless. With disabled L2 once I was managed to run Speedsys at 95MHz FSB, getting CPU score about 282 up from 246 but it didn't finish cache/mem benchmarks...

Then I experimented back on PC-crap M590. I revealed that while this board boasts running at 100MHz the reality is such that when you select 100MHz in CPU setup it runs only at 90MHz (PCI is then at 1/3 = 30MHz). Other (lower) FSB values matches reality. But even at 83MHz I got a better score in Quake than on Gigabyte (57 vs 64FPS 320x200), maybe due to bigger L2? But in this MB I have populated only 2 EDO SIMMs 60ns as that PC133 256MB SDRAM DIMM doesn't work there, probably it supports only lower density modules. But important is that in this MB my Tilly ca run at 90 x 4 = 360MHz stable at Vcore 2,3V. It's only 5MHz from 95MHz FSB where I had no luck. Maybe on other MB with 100MHz support it would run....

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 133 of 165, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

I once encountered a bios bug where in my 3+ refused to run any faster than about the speed you achieved.

When I updated the Bios, the processor was able to easily do 600mhz

It may be bios related?

Btw, try the system bios cachable option I’ve seen weird issues with it on 3+. I really doubt it’s your issue, but maybe easy to try.. You can also fiddle with other settings.

Have you talked to Jan about these issues?

I’ll tag him in this post.

Btw, a thought about binning of cpus,

It may be that the ppga packages just overclock better, through ppga design or on purpose placing better cores there.

Have a look at the tillamook working L2 thread. Many people get 400mhz out of them all the time. It’s pretty standard for them. (Ppga)

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 134 of 165, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

On GA-5AA I have the latest beta BIOS F7 that I found, older don't work properly with Tillly.

I tried to desolder 2 elyts 1000uF/10V close to VRM Vcore output and measure it. The capacity was still in good condition ~ -5% and ESR also acceptable. For sure I replaced them by Nippon 1500uF but it didn have any effect on (un)stability.

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 135 of 165, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

I have found another bad/weird behavior of Gigabyte GA-5AA or used Ali Alladin crapset. I played with more CPUs and added some benchmarks to table http://rayer.g6.cz/hardware/tillamok.htm#CPU_TEST4 while I discovered that at 66MHz FSB I got much better memory performance than on higher FSB. Simply when stepped from 66->75MHz FSB there is big memory drop down like the chipset would add some wait-states or so. This led to paradox that I got slightly more FPS in Quake on my Tilly at 266 (4 x 66) MHz compared to 333 (4 x 83) MHz, that's insane, I have temption to throw the MB out of the window! Even that crappy PC-Chips M590 with heavy O'C SiS chipset has better performance than this. I would need to try on some better MB. I guess that on some intel chipset I would get better memory performance even at lower FSB, e.g. 75MHz that's probably intel can do max. What non-intel Super S7 chipset is the best that can handle 100MHz FSB and with bonus of support AGP and SDRAM?

Last edited by RayeR on 2023-12-25, 14:16. Edited 1 time in total.

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 136 of 165, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

The coolest socket 7 motherboard that you’ve never heard of

And

Asus p5a version 1.04 are my most favorite.

But the freeway has really good agp. It can run a radeon 9800 pro

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 137 of 165, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

Hm, but that freeway doesn't have any ISA - it's not cool. I though that ISA started disappearing somewhere later in Pentium III era with introduction of intel 8xx chipset...

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 138 of 165, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
RayeR wrote on 2023-12-25, 14:22:

Hm, but that freeway doesn't have any ISA - it's not cool. I though that ISA started disappearing somewhere later in Pentium III era with introduction of intel 8xx chipset...

Try p5a 1.04 then

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 139 of 165, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

This one?
https://theretroweb.com/motherboards/s/asus-p5a
I can see it has similar Ali chipset like mine GA-5AA (ALi M1541 vs Ali M1542 and the same M1543C) so I doubt it would perform better.
There's also important note:
Due to a hardware problem P5A Rev 1.05 and 1.06 will work extremely slow with a K6plus CPU. this has been noticed both with rev E and rev G of the chipset.

No mention about 1.04, does it mean that newer chipset revision has some HW problem? Maybe similar issue with M1542.

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA