Is GDDR SDRAM the same thing?

No. Different pinout, physical package and timings.

The Serpent Rider wrote on 2020-12-05, 19:16:

Different pinout, physical package

I compared datasheets and amount of balls and spacing is the same, pinouts also seem to match (except Hynix which has MCL pin unconnected).

Don't know how important are the timings, Hynix and Samsung seem to be rated for better speeds.

This is a picture of another Radeon 9000 which uses HY5DU283222AF-33 GDDR SDRAM. I have a card with different PCB but the same Radeon chip and it uses HYB25D128323C DDR SGRAM.

Wouldn't that mean that these chips could be interchangable?

I just replaced two (out of four) original Infineon HYB25D128323C-3.6 128 Mbit DDR SGRAM chips with two NOS Samsung K4D263238G GC33 128Mbit GDDR SDRAM chips.

I managed to get a clean image, but occasionally artifacts do appear. So it looks like DDR SGRAM may be compatible with GDDR SDRAM, I will replace remaining two chips and hopefully will have the card fixed.

//update: I set the preheater too high while replacing chips on the top layer and one of the bottom memory chips "desoldered" a bit, I gave it a quick reflow because I'm getting impatient with this board, and now I have good image in Windows... let's do some benchmarks

adalbert wrote on 2020-12-30, 22:13:

I just replaced two (out of four) original Infineon HYB25D128323C-3.6 128 Mbit DDR SGRAM chips with two NOS Samsung K4D263238G GC33 128Mbit GDDR SDRAM chips.

I managed to get a clean image, but occasionally artifacts do appear. So it looks like DDR SGRAM may be compatible with GDDR SDRAM, I will replace remaining two chips and hopefully will have the card fixed.

//update: I set the preheater too high while replacing chips on the top layer and one of the bottom memory chips "desoldered" a bit, I gave it a quick reflow because I'm getting impatient with this board, and now I have good image in Windows... let's do some benchmarks

I admire your determination, considering how inexpensive Radeon 9000s still are .

I also wish I had your skill and equipment to be able to reflow and/or solder BGA components.

Excellent work! and it's good to know that the two ram types are compatible. I do hope you are able to fully repair your card. Definitely looking forward to seeing the benchmark results.

darry wrote on 2020-12-31, 01:29:

I admire your determination, considering how inexpensive Radeon 9000s still are .

I also wish I had your skill and equipment to be able to reflow and/or solder BGA components.

Thanks. Well, yeah, they are, but this is a tiny mobile version for Inspiron 8100 laptop. And it seems to be a very rare variant of graphics for that laptop. I bought one ~5 years ago, fixed it and now bought this one as a spare. They both had defective memory, so this is some kind of factory defect. So it seems like it is nearly impossible to buy a working one, and BGA rework is necessary to have it working. I fixed first one just by reballing the chips, but with this one it seems that complete replacement of chips is the way to go.

I'm just using a 858D hot air station and a makeshift preheater. I learned most of the stuff by watching Louis Rossman's videos. Being able to fix own stuff is helpful when having a bigger collection of hardware.

chrismeyer6 wrote on 2020-12-31, 01:33:

Excellent work! and it's good to know that the two ram types are compatible. I do hope you are able to fully repair your card. Definitely looking forward to seeing the benchmark results.

Thanks, unfortunately looks like still some artifacts appear (although the image is OK most of the time; before the rework it was complete garbage), so I will replace the second set of chips. Should have replaced all of them in first place, doing half-arsed job is usually not a good idea. Probably after the New Year though.

This is how that board looks in current state, with different chips. Samsung ones are a bit bigger, but have identical pin spacing. Silkscreen on the PCB has two squares, smaller and bigger, so it may actually be designed to support different types of chips.

That's an interesting design being a two piece GPU. Hopefully after you get the last set of chips replaced you'll have a fully functional piece. How's the cooling for the GPU and ram chips in the laptop?

chrismeyer6 wrote on 2020-12-31, 02:04:

That's an interesting design being a two piece GPU. Hopefully after you get the last set of chips replaced you'll have a fully functional piece.

Oh, actually it's a one piece, I just combined the photos of top and bottom side in one image 😀 Large connector goes to the mainboard, smaller is for connecting the LCD.

BTW Here is how it looks with two memory chips removed. I pulled one solder pad (it was unconnected so that doesn't matter) as this board probably underwent too many heat cycles (I reballed original chips earlier).

chrismeyer6 wrote on 2020-12-31, 02:04:

How's the cooling for the GPU and ram chips in the laptop?

RAM chips have no additional cooling. GPU is cooled passively. Keyboard has bottom made of metal, it has thicker center piece and a heatpipe spreading the heat from the center to the right side of the keyboard. It touches the GPU through a thermal pad. GPU doesn't seem to be heating too much. I think that the memory failure may be caused by excessive mechanical stress caused to the board (you are putting weight on video board while typing), not necessarily because of heat. But this is just a guess.

I also have Inspiron 8200, which has identical construction, but has Pentium 4-M CPU. I8100 has Pentium3-M (Tualatin). If I get this card to work, then I will be able to run benchmarks on both of them with same CPU, comparing performance difference between mobile P4 and P3 Tualatin. I may also try overclocking the memory on the GPU with Samsung Chips, as they have better specs than the original ones.

The mechanical stress makes alot of sence causing the issues with the ram. You get someone who's heavy handed typed or just rough on their keyboards and I can really see that circuit board getting very flexed. You add in the new at the time and very problematic RoHs lead free solder and that will lead to damaged or cracked solder balls.

I set the preheater too high while replacing chips on the top layer and one of the bottom memory chips "desoldered" a bit, I gave it a quick reflow because I'm getting impatient with this board, and now I have good image in Windows... let's do some benchmarks

Interesting. That probably means they are both SGRAM chips and naming is just misleading.