VOGONS


First post, by Kahenraz

User metadata
Rank l33t
Rank
l33t

A GeForce 4 MX 440 I ordered exhibited characteristic display corruption when rendering 3D scenes. The card in question has BGA memory which is both expensive and difficult to repair, so I tried a cheaper alternative. I downclocked the memory from 333 Mhz to 290 and the problem went away.

I'll be attaching a note to the card for the future, so that I can remind myself how to deal with this fault. There's nothing wrong with it otherwise.
We
Here are some neat photos of the display corruption.

20220322_173753_resize_65.jpg
Filename
20220322_173753_resize_65.jpg
File size
145.44 KiB
Views
1282 views
File license
CC-BY-4.0
20220322_173557(0)_resize_31.jpg
Filename
20220322_173557(0)_resize_31.jpg
File size
260.75 KiB
Views
1282 views
File license
CC-BY-4.0
20220322_174627_resize_58.jpg
Filename
20220322_174627_resize_58.jpg
File size
203.99 KiB
Views
1282 views
File license
CC-BY-4.0
20220322_174302_resize_87.jpg
Filename
20220322_174302_resize_87.jpg
File size
144.22 KiB
Views
1282 views
File license
CC-BY-4.0

Reply 1 of 34, by Horun

User metadata
Rank l33t++
Rank
l33t++

Nice ! Maybe age + heat has caused the ram to not function at default speed. Could also be a capacitor issue....

Hate posting a reply and then have to edit it because it made no sense 😁 First computer was an IBM 3270 workstation with CGA monitor. https://archive.org/details/@horun

Reply 3 of 34, by Kahenraz

User metadata
Rank l33t
Rank
l33t

Everything was fixed by downclocking the memory.

Out of curiosity, I looked up the datasheet for these chips, Infineon HYB25D128323C-4.5. According to the specifications, these 4.5ns memory chips are only rated for 222 Mhz. It looks like some manufacturer tried to be cheap. I'm even more surprised that it's PNY.

Reply 4 of 34, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

Is that the base clock or the DDR speed? I was casting around to try to find some vmod info in case a minor bump would get you stable, for which all convenient links seem to be dead and unarchived so far. However, came across talk that the mem on "great" MX440s crapped out not far past 550 DDR, which is 275 base. So wondering if someone modded the BIOS or flashed some MX460 golden sample type BIOS to it.

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 5 of 34, by Kahenraz

User metadata
Rank l33t
Rank
l33t

This is a bog standard MX 440SE and is clocked at 250/333 by default. Golden sample it is not.

At 250/290, it is very close in performance to my MX 440 8x. I overclocked the 440SE core to 400 Mhz, but there was only a very slight improvement. This card seems to respond more to memory bandwidth than core clock. I tested this on a fast Pentium 4 3 Ghz with an AGP 4x slot.

Reply 6 of 34, by Doornkaat

User metadata
Rank l33t
Rank
l33t

Did some previous owner maybe mess with the VGA BIOS to overclock the card and forgot to differentiate between real clocks and effective DDR clocks? As far as I understand the MX 440 SE has its VRAM clocked at 166MHz (resulting in 333MHz effective with DDR memory).
The chips are rated 4.5ns which is faster than that (222MHz real/444MHz effective).

Reply 7 of 34, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
Kahenraz wrote on 2022-03-22, 22:42:
A GeForce 4 MX 440 I ordered exhibited characteristic display corruption when rendering 3D scenes. The card in question has BGA […]
Show full quote

A GeForce 4 MX 440 I ordered exhibited characteristic display corruption when rendering 3D scenes. The card in question has BGA memory which is both expensive and difficult to repair, so I tried a cheaper alternative. I downclocked the memory from 333 Mhz to 290 and the problem went away.

I'll be attaching a note to the card for the future, so that I can remind myself how to deal with this fault. There's nothing wrong with it otherwise.
We
Here are some neat photos of the display corruption.

20220322_173753_resize_65.jpg

20220322_173557(0)_resize_31.jpg

20220322_174627_resize_58.jpg

20220322_174302_resize_87.jpg

see if it maybe needs new capacitors sometimes line noise can cause errors

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 8 of 34, by Kahenraz

User metadata
Rank l33t
Rank
l33t
Doornkaat wrote on 2022-03-23, 04:52:

The chips are rated 4.5ns which is faster than that (222MHz real/444MHz effective).

The NVIDIA overclock tool expresses the clock as double, so it was actually 165 (333) Mhz stock. When I said 290 Mhz, I meant 145 at a single clock. It's DDR, so I can understand the confusion.

Maybe I'm reading the datasheet wrong. I'm pretty sure that the clocks it mentions there are also the clock doubled DDR speeds.

Reply 9 of 34, by Doornkaat

User metadata
Rank l33t
Rank
l33t
Kahenraz wrote on 2022-03-23, 10:12:
Doornkaat wrote on 2022-03-23, 04:52:

The chips are rated 4.5ns which is faster than that (222MHz real/444MHz effective).

The NVIDIA overclock tool expresses the clock as double, so it was actually 165 (333) Mhz stock. When I said 290 Mhz, I meant 145 at a single clock. It's DDR, so I can understand the confusion.

Ah, ok, then that's certainly not it.

Maybe I'm reading the datasheet wrong. I thought the clocks it mentions there are also the clock doubled DDR speeds.

In my experience it would at least be unusual.🤷‍♂️
Maybe they actually used marginal chips on this card and one was more marginal than they hoped for?😅

Reply 11 of 34, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Sometimes they are rated at x at xyz but if you drop latencies to uvw you can now run at clock speed y

to an limited extent mhz and latencies/timings are interchangable.

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 12 of 34, by Kahenraz

User metadata
Rank l33t
Rank
l33t

I see. Maybe PNY modified the latencies then? Here is a picture from the datasheet for this chip.

Screenshot_20220323-103551_Drive.jpg
Filename
Screenshot_20220323-103551_Drive.jpg
File size
56.53 KiB
Views
1035 views
File license
CC-BY-4.0

Reply 13 of 34, by Kahenraz

User metadata
Rank l33t
Rank
l33t

This card was particularly dusty, so I decided to wash it in the sink before putting it away. When I did so, I also inspected the capacitors closely and noticed that the top of one appeared to be bulging ever so slightly. I was sure that this wasn't a capacitor issue, but decided to try replacing it anyways. After removal, I found that its capacitance measured fine at 980uF compared to the rated 1000uF. I also replaced the one next to it, which measured lower at 860uF. My capacitance meter doesn't measure leakage, so they could still be much worse than this.

I was very surprised. Now at stock 165 (333) Mhz it's rock solid. I didn't even use anything special; just cheap low ESR capacitors that I order from AliExpress. I was certain that the failure was one of the memory chips, not a capacitor.

Attachments

Reply 16 of 34, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

I almost said check the caps, but I was thinking that was deeper than you wanted to get into it.

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 18 of 34, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
Doornkaat wrote on 2022-03-23, 20:53:
Sphere478 wrote on 2022-03-23, 07:48:

see if it maybe needs new capacitors sometimes line noise can cause errors

You called it! 😄👍

w00t! what did I win?

I love it when we figure out a hardware mystery! good job!!!

Kahenraz wrote on 2022-03-23, 14:38:

I see. Maybe PNY modified the latencies then? Here is a picture from the datasheet for this chip.

Screenshot_20220323-103551_Drive.jpg

yep. you may be able to dump the bios and use nbitor to read the timings? idk though, may be too old to be supported

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)