VOGONS


Reply 980 of 1184, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
feipoa wrote on 2023-06-18, 09:36:
If you have already connect FLT# to GND and it caused an over current condition, then try lifting up the pin as I've shown in my […]
Show full quote

If you have already connect FLT# to GND and it caused an over current condition, then try lifting up the pin as I've shown in my photo and bend it to the side (GND). Thinking about this further, I think you can only wire FLT# to GND if the motherboard isn't connecting it to Vcc. FLT# has an internal pull-up resistor and, according to the datasheet, if FLT# is not used, it should be left disconnected. Thus I am surprised that your motherboard was sending FLT# to Vcc. However, if you look at the photo I provided, it looks like FLT# on my motherboard was also being sent to Vcc, in contravention to the datasheet. It has been many years since I did this mod, but it looks like I did the pin left up because I noticed FLT# was going to Vcc. Sorry, I didn't remember this.

You really need to get any PGA132 CPU to continue testing, preferably a Cx486DLC or Tx486DLC. These should be like $15 max. EDIT: I just looked on eBay, and am in disbelief, wow! There's tonnes of these chips, but how did the price of jump to $50 or more? Cheapest I could find was $40 https://www.ebay.com/itm/134349014775 I suggest hit up CPUShack to see if he has any extra he doesn't need. He probably does because I cannot see these selling on CPU-World since collectors already have them.

The datasheet for the A80386DX can be found on digikey. They are still selling these!
https://www.digikey.com/en/products/detail/ro … DXL-25/12095484

By the way, you should be running your SXL2 at 3.6 V or higher, not 3.3 V.

Okay, let me adjust to 3.6v and re test. Perhaps this is the problem.

Thanks for the links 😀 you rock!

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 981 of 1184, by feipoa

User metadata
Rank l33t++
Rank
l33t++
Sphere478 wrote on 2023-06-18, 16:37:

Okay, let me adjust to 3.6v and re test. Perhaps this is the problem.

Probably won't help. You need to disable or remove the onboard Am386.

Plan your life wisely, you'll be dead before you know it.

Reply 982 of 1184, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
feipoa wrote on 2023-06-18, 20:20:
Sphere478 wrote on 2023-06-18, 16:37:

Okay, let me adjust to 3.6v and re test. Perhaps this is the problem.

Probably won't help. You need to disable or remove the onboard Am386.

I also share this opinion. 😀

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 983 of 1184, by rafpereira

User metadata
Rank Newbie
Rank
Newbie

I was able to set clock double to 66mhz and 8k cache using a TI486SXL-40, and the TI486SXL2-50, even the slower variant seem to be stable at standard 3.6v

This board has a Kyocera 66mhz oscillator soldered, so I won't be testing at higher clock.

Compared to the DRx2 33/66 in this board with 128k 20ns cache, 1ws and isa clk/3 the difference so far has been minimal to TI486SXL2-66 advantage in the current configuration.

Attachments

Reply 984 of 1184, by feipoa

User metadata
Rank l33t++
Rank
l33t++
rafpereira wrote on 2023-06-29, 17:40:

I was able to set clock double to 66mhz and 8k cache using a TI486SXL-40, and the TI486SXL2-50, even the slower variant seem to be stable at standard 3.6v

This board has a Kyocera 66mhz oscillator soldered, so I won't be testing at higher clock.

Compared to the DRx2 33/66 in this board with 128k 20ns cache, 1ws and isa clk/3 the difference so far has been minimal to TI486SXL2-66 advantage in the current configuration.

The TI486SXL-40 is stable at 3.6V and 66 MHz? That is really curious. What software did you run and did you stress the system?

Plan your life wisely, you'll be dead before you know it.

Reply 986 of 1184, by rafpereira

User metadata
Rank Newbie
Rank
Newbie

I changed the ISA clock from clk/3 to clk/2 and with regular fan cooling on the TI486SXL-40, boot in MSDOS 6.22, usual startup with himem, and cyrix -e -b -cd

then this sequence ran without a reboot or reset at 33mhz*2 (66mhz)

Phil's dosbench option:
b) Doom timedemo - Timed 2134 gametics in 9156 realtics
c) Quake timedemo - 969 frames in 452.4 seconds - 2.1 fps
m) Topbench 3.8 - 111
t) Cache test - screenshot

CheckIt 2.1 - Memory test - full continuous test (one full iteration done)

No lock-ups, no beeps, no bruises.
This setup/run does not look at all stressed at this clock, is there something specific you were looking for?

Attachments

Reply 987 of 1184, by feipoa

User metadata
Rank l33t++
Rank
l33t++

From my experience, usually Windows apps tend to catch more instabilities compared to DOS, however I do not have a set recipe. It normally it involves opening IE5 and letting google load a search criteria a few times. Then I throw all my Windows software titles at it, e.g. Word Perfect, Photoshop, Acrobat, various benchmarks, mp3 player (very down sampled), etc. Sometimes I run CAD rendering in DOS, but I found my blast of two dozen Windows software titles to be sufficient. If this passes, I may run NT 3.51 just to be sure it is stable. I also like to let Quake run in loop for an hour or so, check DOOM and sometimes Duke3d if it happens to be on my test HDD.

I think pshipkov has a standard routine he runs, however I've in the past run his routine which passed, but then get an IE 5 error.

Next time I have an SXL2 setup going, I will have to test my 5V SXL-40 and SXL2-50.

Plan your life wisely, you'll be dead before you know it.

Reply 989 of 1184, by feipoa

User metadata
Rank l33t++
Rank
l33t++
pshipkov wrote on 2023-06-30, 16:15:

I still cannot believe that there are 40MHz rated parts able to do 66MHz.
I have 2 50MHz rated SXL2s and none of them can do it.

Did you try the PGA-168 variants?

Plan your life wisely, you'll be dead before you know it.

Reply 991 of 1184, by feipoa

User metadata
Rank l33t++
Rank
l33t++

He is saying that his 5V PGA-168 40/50 MHz CPUs run at 66 MHz, not the PGA-132 variants. I find this very surprising and will have to confirm on my end.

SXL_variants.JPG
Filename
SXL_variants.JPG
File size
563.28 KiB
Views
1089 views
File license
CC-BY-4.0

Plan your life wisely, you'll be dead before you know it.

Reply 992 of 1184, by pshipkov

User metadata
Rank Oldbie
Rank
Oldbie

Yes, same here - regardless if 168 or 132 it is unexpected that to see that.
PGA-132 SXL-50 just do not scale well on fast boards. But on some slow ones they can get in the 60-65 zone ok with some effort.

retro bits and bytes

Reply 993 of 1184, by rafpereira

User metadata
Rank Newbie
Rank
Newbie

To be clear, the PGA168 TI486SXL-40 mentioned (which compared to your picture does not have the "FC2-" marking), can enable, and I do enable clock double at 33Mhz FSB at 3.6v (not 66mhz FSB), the SXL was rated 40mhz FSB, so enabling clock double in the SXL for me was the only surprise. The specs call for 3.3v and 5v versions, how do we differentiate? (Never seen a V or G variant of the 40 and 50mhz)
https://ia903003.us.archive.org/35/items/bits … rence_Guide.pdf

at CLK/3 and 1ws setting, so far, I was unable to crash it no matter how many times I opened and closed IE5 on Windows 3.11 while on Solitaire. My current CF card is 1Gb and it is not setup for a MS Word, Excel installation.
I did install a Crystal CX4236B-X03 sound card and with Unisound I had a few games in SBpro setting: Jazz Jackrabitt (486DX2-50 setting, High detail, Turbo mode), Sam & Max, Tyrian, Darkforces... all with sound, and working as expected.

If I need to go and install NT 3.51 in this board, and spend time figuring out how to call cyrix.exe before the bootloader, then maybe I should ask if you have this installed in a disk image you can share. Going the long way will take me a while, and I am not sure it is worth for me.

Update: Windows 95 installed, Office 95 installed, Tom 2D benchmark ran for 6+ hours. Ethernet card added, IE searching on Google while Kayaker video playing. RAM upgraded from 8Mb to 16Mb.

Do you know what FC2 means? My SXL2-50 has it as well.

Attachments

Reply 994 of 1184, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Had some time (and motivation) today to play with this some more,

Removed the varta,
Reinstalled the over clocking device (anyclk) using pins and reinstalled the factory socket
Put a housing on the removable battery pins

But none of that is interesting to this thread really,

Next I did a little investigating on the flt pin, wasn’t really able to see anything interesting or useful by examining the motherboard images without a processor available online next to the actual back of my motherboard, maybe I missed something but it seems that they may have actually not made this socket usable without modifying or removing the cpu.

So I did the pin pull up (physically) pull down (electrically) mod, but with a small wire and some kapton tape.

The system boots, but shows up as a frx 386dx in post. Is there a patch that can be applied?

Windows is even weirder, mentions cyrix instead (not a typo) in system properties.

Working on getting cpu z on the system.

Attachments

  • image.jpg
    Filename
    image.jpg
    File size
    851.8 KiB
    Views
    1000 views
    File license
    Public domain
  • image.jpg
    Filename
    image.jpg
    File size
    1.99 MiB
    Views
    1000 views
    File license
    Public domain

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 995 of 1184, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Well, I gave up for now trying to get phill’s or cpu-z on there. Cd drive isn’t reading RWs.. oh well. Shal resume at another time.

The system is a fair deal faster, so pretty sure it worked. I did notice that the boot summary screen was showing this now. So it was a success!

IMG_3644.jpeg
Filename
IMG_3644.jpeg
File size
1.09 MiB
Views
993 views
File license
Public domain

Now to learn about the cyrix/TI registers..

As far as the pcb, and going to release version, there doesn’t really seem like there is any significant changes to be made, this works about as well as one could hope for.

Could probably put silk indicating locations that can be used on the pins for 5v capacitors, but that isn’t really needed.

I’ll have to report over clocking results later on, for note, I used the capacitor values my friend and I came up with for the vcc3 from the capacitor response testing, curious how it stacks up against what feiopa came up with. I suspect feiopa’s efforts probably are a cleaner result given the number of tests he did. would have to scroll back and see what those values I used were, I forget. As for the 5v caps, all are 10uf and all are populated. (Pads were there, and I had them…🤷‍♂️ why not. Lol)

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 996 of 1184, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Let's see some noise measurements. For an equal comparison, run the chip in 1x mode at 40 MHz (no clock doubling), 4.05 V, and x10 low-inductance probe right on the VRM's pins. My results for both VRM options were 80-90 mV Vp-p.

What cap values did you end up with?

I'm using Cin1 ceramic 10uF, Cout1 tantalum 22 uF, and the others as shown below. I don't recall the exact values I used for Vin on the central region of the PCB, but they aren't important for Vout measurements. I tried these two configurations for the bottom PGA caps:

PGA_Vout_caps_bottom_options.JPG
Filename
PGA_Vout_caps_bottom_options.JPG
File size
216.43 KiB
Views
976 views
File license
CC-BY-4.0
PGA_Vout_caps_bottom_alpha2.JPG
Filename
PGA_Vout_caps_bottom_alpha2.JPG
File size
164.74 KiB
Views
976 views
File license
CC-BY-4.0

I suppose you could silkscreen these 0805 caps for the top pins. You can also silk the bottom Vcc5 if you wish.

PGA_Vin_caps_top.JPG
Filename
PGA_Vin_caps_top.JPG
File size
189.16 KiB
Views
976 views
File license
CC-BY-4.0

I think you should remove the three Vout capacitor pads on the bottom of the PCB entirely. In general, the Vin caps have almost zero effect to clean up Vout. IIRC, they helped reduce noise by ~5 mV only. As for the capacitor pads on the top of the PCB, if you are wanting to make the planes more robust, the below configuration is probably more than sufficient. Keep the purple circles, remove the other pads, and add silkscreen to where the red circles are. For consistency, maybe make the two remaining (purple) cap pads 0805.

PGA_Vin_caps_top_options_these_locations_sufficient.JPG
Filename
PGA_Vin_caps_top_options_these_locations_sufficient.JPG
File size
349.38 KiB
Views
976 views
File license
CC-BY-4.0

I'm not sure what else to alter. We are already nit picking.

Plan your life wisely, you'll be dead before you know it.

Reply 997 of 1184, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
Sphere478 wrote on 2023-03-15, 01:41:

Conclusion, what we can try to target 80mhz and 5khz is to populate the 3.3v pga pin pads with alternating 470pf and 10uf alternating on between half, to all the pga pin cap points.

This is what I did on all the vcc3 pga pins

Is there a bios patch I can apply for this cpu or no choice but to run as is?

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 998 of 1184, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

I heard back from chkcpu,

Hi Sphere, In interesting project, getting a PGA168 TI486SXL2-66 running on a PGA132 socket board! Although I never did work on […]
Show full quote

Hi Sphere,
In interesting project, getting a PGA168 TI486SXL2-66 running on a PGA132 socket board!
Although I never did work on these 386/486 1991/1992 BIOSes, I know that the AMI 1991/1992 core BIOS uses a quite generic CPU model detection. When it detects a 486, it looks for an FPU and displays a 486SX or 486DX accordingly.
So on the 12/12/1991 AMI core, even a DX2, or DX4 with interposer, are indicated as ‘486DX or 487SX’ CPU. But experience shows that it runs! 😉
So if the TI486SXL2 boots and runs, that’s all you need and Cyrix.exe will take care of the rest.
No need for a BIOS update or patch.
Cheers, Jan

I’ve been having trouble getting files on this system. It isn’t reading burned CDs for some reason, and the files are too big for floppies. I may have to get creative and hook up a ide drive. The system is scsi.

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 999 of 1184, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

I’m voting for round pads, that whole D pad thing was a fiasco that I don’t think helped anything. The board is better with round pads anyway. I can make a note to elevate capacitors above pcb to deal with bridging conserns?

If you are really set on having the D pads I will add them back. 😀 Reminder though, jlc sometimes sends them clearanced for D but pads themselves are round. Another reason to just make them round..

Feiopa, could you whip up a recommended BOM with alternates and explainations for components as well as assembly notes I will include it in the final file. Kinda like a pdf instruction sheet for the end user ya know?

Attachments

  • image.jpg
    Filename
    image.jpg
    File size
    1.52 MiB
    Views
    875 views
    File license
    Public domain
  • image.jpg
    Filename
    image.jpg
    File size
    1.5 MiB
    Views
    876 views
    File license
    Public domain

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)