Jetway N2PAP-LITE motherboard recap

[momaka]

Is recapping motherboards a thing of the past anymore? The motherboard forums look kind of sad. Perhaps I shall try do to a bit of a revival, if no one minds.

I will start with a good old socket 462 motherboard – a Jetway N2PAP-LITE. If you ask why I post about such an old motherboard now… well, it is because I like fixing / refurbishing old reliable hardware. This board has leaded solder and is from the early 2000’s, when PC components were still designed with some reliability in mind (sans the bad cap brands). The hot red PCB color is also something that can’t be found nowadays – on that note, I really miss the days of colored PCB motherboards.

For those who have not heard the name before, Jetway was one of the many motherboard companies back in the early 2000’s. As with many motherboards of that time, they used obscure capacitor brands. In particular, they always seemed to stick with GSC capacitors (which are probably one of the worst cap brands out there - so bad that they changed names multiple times in hopes of fooling us.)

Nonetheless, I find that Jetway motherboards are usually well-designed, offer good amount of BIOS features, and work great after a recap. It’s unfortunate that they stopped making ATX desktop motherboards around the socket 775 and AM2 era. Jetway still produces motherboards today, but from what I’ve seen, it is mostly (if not all) ITX / mini-ITX form for industrial use and/or POS systems. On the positive side, it seems that they have finally ditched GSC/Evercon/Sacon and use solid polymer capacitors like everyone else.

Anyways, now the more interesting part of the recap: a motherboard diagram with a list of the original capacitors along with what goes where (and other details.)




There are important caps on 9 distinct voltage rails, more or less. Going by the cap diagram, these are:

#1: 5 Volt rail for the CPU. (a.k.a. CPU “high” side)
It comes directly out of the 5 V rail on the ATX connector.
- Original caps: 2x GSC ME, 16 V, 2200 uF, 12.5 mm ø, 26 mOhm ESR
- Replaced with: 2x Nichicon HM, 6.3 V, 3300 uF, 12.5 mm ø, 11 mOhm ESR

While it is not advisable to replace caps with ones with a lower voltage rating, I think the diagram clearly shows why I could do that: the Nichicon HM capacitors I used above were on the 5V rail, so there was no need to go with 16V caps at all. Jetway likely used 16V capacitors here to get lower ESR out of less capable parts. The Nichicon HM capacitors I used had lower ESR while also having higher ripple current rating, higher capacitance, and were smaller in size.

Perhaps that goes to show you how “good” GSC ME is . On that note, even Chemicon KZE would have worked here.

#2: CPU Vcore. (a.k.a. CPU “low” side)
This rail provides power to the CPU. It is generated by a 2-phase synchronous buck converter circuit from #1 above..
- Original caps: 5x GSC ME, 6.3 V, 1500 uF, 10 mm ø, 20 mOhm ESR, 2000 mA RC (RC = ripple current)
- Replaced with:
2x Nichicon FPcap R7 series, 6.3 V, 820 uF, 10 mm ø, 7 mOhm ESR, 6100 mA RC
2x Nichicon HZ, 6.3 V, 2200 uF, 10 mm ø, 7 mOhm ESR, 3770 mA RC

Needless to say, this combo of replacement caps was more than adequate for the task. Both the overall capacitance of the CPU Vcore was increased slightly (better for voltage regulation/stability) and the ESR was reduced (better transient response).
Worth noting is that I was actually going to use 6.3 V, 1800 uF Nichicon HM caps instead of the Nichicon HZ caps, but by the time I did the recap, I already had many used/pulled caps from dead Xbox 360 motherboards, so I used those instead. Using Nichicon HZ and polymer caps here was not the least bit necessary , as even the most power-hungry Athlon XP CPUs don’t use more than 80 Watts of power.

#3: V_dimm / V_RAM:
2.5-2.6 Volts for the DDR RAM slots. This rail is generated by MOSFET Q11 (CEB6030L) in a linear fashion.
- Original caps: 3x OST RLG,16 V, 470 uF, 8 mm ø
- Replaced with: 3x Nichicon HD, 10 V, 470 uF, 8 mm ø
Note: this is a linearly-regulated rail, so pretty much any 105°C cap can be used here. Hence why OST RLG worked. But I had the above Nichicon HD in stock, so I used that.

#4: Northbridge V_core/Vcc:
Power for the Northbridge chip, 1.60 Volts. This is another synchronous buck-regulated rail, so ultra-low ESR caps (including polymers) would be okay.
- Original caps: 3x GSC ME, 6.3 V, 1000 uF, 8 mm ø, 35 mOhm ESR, 1129 mA RC
- Replaced with: 3x Nichicon HM, 10 V, 1000 uF, 8 x 15 mm (ø x h)… ESR and RC: way better than GSC ME

#5: V_AGP (AGP bus voltage):
Typically 1.5 V or 0.8 V, but expect anything up to 3.3 V here, due to compatibility with old cards. It is another linearly-regulated rail.
- Original caps:
2x GSC ME, 6.3 V, 1500 uF, 10 mm ø, 35 mOhm ESR, 1129 mA RC
… and a *tiny* GSC T-series 16 V, 100 uF, 6.3 x 11 mm cap for whatever reason
- Replaced with:
2x Nichicon HM, 6.3 V, 1800 uF, 10 mm ø
1x Nichicon HE, 10 V, 220 uF, 6.3 x 11 mm for the tiny GSC cap

#6: Southbridge Vcc (always ON):
3.3 Volt standby rail generated linearly by an ALP1087 LDO from the power supply’s 5VSB rail, and it is always present as long as the PSU is plugged in.
- Original caps:
1x OST RLG, 16 V, 470 uF, 8 mm ø and 1x GSC T-series, 16 V, 220 uF, 6.3 mm ø
- Replaced only the GSC T-series cap with:
1x Nichicon HE, 10 V, 220 uF, 6.3 x 11 mm cap
(I ran out of the 470 uF Nichicon HD caps, but it should be okay, as this is a linearly-regulated rail and does not put caps under any stress. It is possible to get away even with GP caps here, which the OST RLG is.)

#7: RAM Vtt (termination) voltage:
This is typically half of the 2.5/2.6V DDR supply, so expect no more than 1.35 Volts. Another linear rail, generated by U2 (a Winbond W83310S-R) at the upper-right corner of the motherboard. Original cap (yes, ONE cap): GSC T-series 16 V, 220 uF, 6.3x11 mm. Replaced with a Nichicon HE 10 V, 220 uF cap, same 6.3x11 size. Honestly, I expected a bit more filtering from Jetway here, but I suppose that 220 uF cap is enough for two DDR slots.

#8: 3.3 V rail from the PSU:
This supplies various circuits of the board, but most importantly: RAM V_dimm, RAM Vtt, and V_AGP. Had two OST RLG 16 V, 470 uF caps and a single GSC 16 V, 220 uF, 6.3x11 mm cap. Replaced by two United Chemicon KY 10 V, 470 uF caps, 8x12 mm and Nichicon HE 10 V, 220 uF, 6.3x11 mm.

#9: 5 V rail from the PSU:
This supplies various circuits of the board, including the USB, PS/2 ports. But most importantly, it supplies the high side of the buck regulator for the Northbridge V_core/Vcc via caps C402 and C403. Because of that, low ESR caps are recommended for these two spots. This rail has the following caps:
Two GSC ME 6.3 V, 1000 uF, 8 mm diameter.
Two GSC T-series 16 V, 220 uF, 6.3x11 mm caps
One GSC T-series 16 V, 100 uF, 6.3x11 mm cap

I replaced the first two above (C402 and C403) with United Chemicon KZE 10 V, 1000 uF, 8x16 mm. The two 220 uF and one 100 uF GSC caps were all replaced with Nichicon HE 10 V, 220 uF.

#10: 5VSB rail from the PSU:
There is only one cap that filters this rail that should be considered important: It is marked with #10 on the cap map (though, it is not on the list below it). That cap is a 16 V, 470 uF OST RLG. I replaced it with a 10 V, 470 uF United Chemicon KY

Finally, I also replaced the audio output coupling capacitors (C454 and C450). These were standard 100 uF, 16V caps... GSC, of course. I used 2x 220 uF, 16V Panasonic blue GP caps in their place (85°C rated). The increased capacitance of those should help with the low-frequency response of the audio output.

I forgot to take good “before” pictures to show all of the GSC capacitors in their fully glory. Here they are nonetheless:
http://www.badcaps.net/forum/attachm...1&d=1498712430
http://www.badcaps.net/forum/attachm...1&d=1498712430

I suppose that is not a big loss, though, as none of the GSC caps on my mobo were bulged or leaking. That said, I did test a few on my TT / ESR meter and many of the caps on rail #2 (CPU V_core / low side) as well as #1 (CPU high side) were reading almost abnormally high on capacitance (right around the top of the 20% tolerance). I think only one or two were over 20%. However, they appeared to have good ESR. The high capacitance / good ESR combination often implies high internal leakage in the cap. Although I was able to reform them on my cap reformer, they took longer than a good cap and didn’t hold full charge for too long. So they have started to go high on the leakage current spec, indeed. Oh well, time to put them in my “spare junk caps” bin and in with the new ones. Pictures after the recap:

http://www.badcaps.net/forum/attachm...1&d=1498713285
http://www.badcaps.net/forum/attachm...1&d=1498712430
http://www.badcaps.net/forum/attachm...1&d=1498712430

Needless to say, my recap worked just fine. In fact, I took it out today to test some video cards and even just for the sake of powering that PC on. Played some older games too, while at it. It sure brought back some good memories. Oh wait, I forgot about the history of this motherboard / PC. Will be adding in the next post, in case anyone is interested. Stay tuned (or don't, if you have no time for a wall of text again )

[momaka]

Why the long post?
- This motherboard clearly has some sentimental value to me. It was in my first “own” computer as I as growing up… dare I even say I called it a “gaming rig” once upon a time.

I’ve had it from about late 2003 to mid-2006 or so. It sported a 1400 MHz AMD Duron Applebred CPU (Thoroughbred core with disabled L2 cache), 256 MB of DDR RAM, Asus Radeon 9200 SE 64 MB video card , and a 20 GB Maxtor HDD…yes, a far cry from a proper gaming PC for that era , but still quite an improvement over the Pentium II PC it replaced. Main reason I stopped using was because the CompUSA “500 Watt” PSU (a Powmax LP6100-D) in it developed bad caps and became unstable. This is what lead me to badcaps.net (thank you PowMax! ).

Around 2008, I put a new PSU in it (Thermaltake TR2-430W XP550NP.) That’s when I started reading about bad capacitors on the forums here and decided to stash the system away until I recapped the motherboard (despite it actually running perfectly stable). The motherboard probably never saw more than 4000 power-ON hours up to that point.

Then around late 2010, I finally made a capacitor order for the motherboard, along with a few other projects. However, it wasn’t until early 2014 that I sat down and recapped it. Unfortunately, I never really devoted enough time to play with it after the recap. As such, it still sits in my closet… working, with the original Windows XP first 2600-build that a friend of my family installed for me. The only upgrades I did to it after the recap was swap the 1.4 GHz Duron CPU for an Athlon XP 2500+ and upped the RAM to 1 GB (a dodgy Buffalo PC3200 stick that didn’t work in any other PC I tried it in, except this Jetway motherboard) from the upgraded 640 MB I had before. I also have an abundance of decent AGP video cards now: Radeon 9600, three Radeon 9700, Radeon 9800 XL (but now always working right), Radeon x800 GTO, and Radeon x1650, just to name a few.

With a new HDD and one of those cards, that should make a pretty cool early 2000’s retro game rig. I just need to get around and finish that PC. Today I tested the Radeon 9800 XL (after almost coming to tears trying to play anything with the Radeon 9200 SE that I originally had in there ), and I have to say, WHAT A FREAKIN’ IMPROVEMENT! (sans the 9800 XL artifacting in some games)
I could play all of my old games on 1280x1024 at nearly 60 FPS. Most notable was Need For Speed Hot Pursuit 2, which used to struggle at 20-30 FPS at 800x600 resolution with the Radeon 9200 in there. With the 9800 XL, I was getting over 180 FPS @ 1280x1024!

*Sigh*
I remember the days of playing Half-Life 2 Deathmatch multiplayer at 640x480 windowed mode and getting 15 FPS average, with dips as low as 10 FPS and “highs” as much as 30 FPS. If only I had this upgraded recapped PC back then…

Anyways, what is past is past. That PC had its glory days. Now it sits recapped just as a functional museum piece in my PC collection. I don’t think I will ever part with it, though.

This inspires me to make a thread about that PC in the "Post Your System" thread. I do have a few pictures of the PC "over-the-years" in use. An alternative place for that post could also be in the Ghetto Mod thread.

[Wester547]

Very nice repair!

What's funny is that GSC's / Evercon's ME series is supposed to be a 125ºC series on an equal footing with NCC KZH / Rubycon ZLH / Nichicon HV / Panasonic FM in terms of specs... yet these were still failing even if they weren't blown like all those other GSC capacitors. I highly doubt they would live up to the proclaimed 125ºC spec, nor the other numbers in the datasheet. Of the three brands that are "one in the same", Sacon seems to be the worst, GSC second, and Evercon third, but all are undoubted crap anyway... because they are all the same brand. Go figure. Of course nothing is as bad as Sacon FZ (except for Lelon's RXA series and YEC's LZ series, but both came before the "advent" of Sacon FZ... CapXon LZ could also be a runner-up).

Even though they're "ancient" now, I think the R300 and R350 cards from ATi / AMD are still nice legacy cards to have. The R300 (Radeon 9700 Pro) was one hell of a performer for its time, only truly upstaged by the NV40/NV45/NV48 (IE, the GeForce 6800 Ultra)... which had the bumpgate defect due to insufficient underfill material and the usage of leaded bumps with eutectic pads for the flip-chip. So it was short-lived even compared to the R300.

So you used Nichicon HZ and Fujitsu polymers for the VRM output... just because you can? I don't think those GP GSC series actually have any series designation. IIRC, "T" seems to be part of a factory / date code. Those linear regulated rails do not stress the capacitors after so long as the regulators themselves aren't dissipating too much heat, thus radiating too much heat toward the capacitors next to them.

Nichicon HM doesn't seem to be *too* much better than GSC ME's "claimed" HV grade spec... but being GSC, I wouldn't lend them any credence. Although it's interesting that those GSC capacitors are failing the same way as the failure-prone date range HM / HN / HZ do (of course any cap could have high leakage current, but abnormally high capacitance readings are another matter). Maybe someone from GSC was working for Nichicon?

...Kidding, of course. But it would explain that.

I have but one more question. You say this board was repaired in 2014. Were these GSCs measured in 2014 or did you put them in your junk caps bin more recently?

[pfrcom]

Originally Posted by momaka

Perhaps I shall try do to a bit of a revival, if no one minds.

Quite the contrary - go Momaka - I like Jetway too

[Dan81]

Originally Posted by momaka

Jetway N2PAP-LITE.

Just for fun, I tried flashing the Jetway N2PAP BIOS on my MSI K7N2 and it actually POSTed.

[mariushm]

re #5 : AGP has 12v, 5v , 3.3v and whatever voltage it uses (1.5v,3.3v) ..see http://www.interfacebus.com/Design_AGP_Pinout_Ver2.html

That 100uF 16v capacitor may have been on the 12v and you replaced it with a 10v rated capacitor.

Besides this.. good job. Surprises you went through all the trouble of using so many 10v rated capacitors, i personally would have used 16v rated ones as I "stock" only 6.3v and 16v.

[momaka]

Originally Posted by mariushm

re #5 : AGP has 12v, 5v , 3.3v and whatever voltage it uses (1.5v,3.3v) ..see http://www.interfacebus.com/Design_AGP_Pinout_Ver2.html

That 100uF 16v capacitor may have been on the 12v and you replaced it with a 10v rated capacitor.

Negative.
That small 100 uF, 16 V cap is connected in parallel with the other two caps labeled "5" on the cap diagram. And I put an arrow on that cap diagram to indicate that the Source of that CEB6030L MOSFET connects to one of the caps on rail #5. This is also where I measured the output of 1.5V. With the input of the CEB6030L MOSFET being only the 3.3V rail, I think it is safe to say those caps will see no more than 3.3V.

So no problem there going even with 6.3V caps.

Originally Posted by mariushm

Besides this.. good job. Surprises you went through all the trouble of using so many 10v rated capacitors, i personally would have used 16v rated ones as I "stock" only 6.3v and 16v.

Thanks.
Yes, I do the same now. Most of these 10V caps were bought way back when I was still a newbie... and a few of them due to being cheaper on Mouser than their 6.3V counterparts. Don't as me why.

Originally Posted by pfrcom

Quite the contrary - go Momaka - I like Jetway too

Me too!

They are also the only manufacturer I've seen use thru-hole CPU sockets instead of BGA for the AMD socket 939, AM2, and Intel 775 motherboards.

I have two more Jetway motherboards to post, but I haven't made a cap map for them yet. And as such, I haven't recapped them either. I think I will post one of them regardless of that. Perhaps in a few days, though. (Have to sort out the pictures from my camera).

Originally Posted by Dan81

Just for fun, I tried flashing the Jetway N2PAP BIOS on my MSI K7N2 and it actually POSTed.

But did it work normally?

FWIW, I've done that on a modern UEFI dual-BIOS motherboard. It worked... as in, the motherboard POSTed and seemed to function OK, until you tried booting into something or installing Windows.

Originally Posted by Wester547

Very nice repair!

Thanks!

Originally Posted by Wester547

What's funny is that GSC's / Evercon's ME series is supposed to be a 125ºC series on an equal footing with NCC KZH / Rubycon ZLH / Nichicon HV / Panasonic FM in terms of specs... yet these were still failing even if they weren't blown like all those other GSC capacitors.

Yeah, I find that very amusing as well. At 125ºC, they would probably puke their guts so hard that there would be nothing left to desolder from the motherboard.

Originally Posted by Wester547

So you used Nichicon HZ and Fujitsu polymers for the VRM output... just because you can?

Yup
Well, it really is a special motherboard to me, so I had to recap it honorably.

Originally Posted by Wester547

I don't think those GP GSC series actually have any series designation. IIRC, "T" seems to be part of a factory / date code.

I wouldn't be surprised if that is the case.

Originally Posted by Wester547

Those linear regulated rails do not stress the capacitors after so long as the regulators themselves aren't dissipating too much heat, thus radiating too much heat toward the capacitors next to them.

Exactly.

From what I have seen, only older ASUS, AsRock, and Dell/Foxconn motherboards (early socket 478 to socket 775 era) have that problem. This Jetway mobo runs quite cool, even with the ~30ºC ambient tempertures in my room right now. Only the CPU doesn't seem very happy (running around 50ºC with the CPU fan on full blast - but at that speed, it is LOUD. And at more normal speed, I easily get 55ºC.)

Originally Posted by Wester547

I have but one more question. You say this board was repaired in 2014. Were these GSCs measured in 2014 or did you put them in your junk caps bin more recently?

I measured them when I pulled them out with an ESR Micro v4.0 that I had borrowed. At that time, most of them showed up only about 10-15% high on capacitance, and ESR neither too low (like a shorted HM/HN/HZ cap) nor too high. Then I measured them again around mid-2015, or whenever I had that cap reformer built. That's when I noticed some of them had started to exceed 20% capacitance on the high side (or were just right on the border). ESR still showing up as "good". Then I spark-tested a few (charge to max voltage and short out against a metal surface), and they seemed a bit "lazy".

But I have them going in for another round right now - three or four of the the 1500 uF units and both 2200 uF 16V unit are installed in an Apevia PSU that I got for cheap. I'll post that another day (hopefully soon). But from what I read on BCN, those PSUs have the OVP set way too high and could kill PC hardware and caps... so that's why I went with the crappy GSC caps, as I didn't care about them. Miraculously, the PSU works fine. I just need to recap it properly now (and do a bit of re-work on its case too).

Originally Posted by Wester547

Even though they're "ancient" now, I think the R300 and R350 cards from ATi / AMD are still nice legacy cards to have. The R300 (Radeon 9700 Pro) was one hell of a performer for its time, only truly upstaged by the NV40/NV45/NV48 (IE, the GeForce 6800 Ultra)... which had the bumpgate defect due to insufficient underfill material and the usage of leaded bumps with eutectic pads for the flip-chip. So it was short-lived even compared to the R300.

The R300/350 cards are nice indeed. But IMO, they were even more short-lived than the GeForce 6800 variants - mostly due to running hot and developing artifacts. I don't think I've seen one still working and not dead in this way.

People say that all of nVidia's trouble started with the 6000 series, but I think the 6 series were nowhere near as bad as the 7000 and 8000 series (with the exception of the GeForce 6200 Turbo-Cache versions, which were manufactured on the same new 90 nm technology as the GeForce 7300/7600 and 7900 cores and are extremely prone to failures). Even the GeForce 7800 GT was a lot tougher than the 7900 GS/GT/GTO - partially due to better-designed cooler and also due to being made on 110 nm tech vs. 90 nm for the 7900, 7600, and 6200 TC. I'm not sure why the lager nm tech helps, but I have seen a lot more GeForce 7900 video cards with artifacts than 7800 and 6800 cards.

[Wester547]

Originally Posted by momaka

The R300/350 cards are nice indeed. But IMO, they were even more short-lived than the GeForce 6800 variants - mostly due to running hot and developing artifacts. I don't think I've seen one still working and not dead in this way.

I still have a working Radeon 9500 Pro and 9700 TX, but neither have seen much of any use besides 2D activities such as browsing and watching videos. So that's probably why they still work. The Radeon 9500 Pro I used to have, that I did droves of 3D gaming on, died 8 years ago (displayed omnipresent artifacts during 2D and 3D applications). Don't know if it was the GPU or RAM that died as the Infineon RAM ran hotter than the GPU under full load.

People say that all of nVidia's trouble started with the 6000 series, but I think the 6 series were nowhere near as bad as the 7000 and 8000 series (with the exception of the GeForce 6200 Turbo-Cache versions, which were manufactured on the same new 90 nm technology as the GeForce 7300/7600 and 7900 cores and are extremely prone to failures). Even the GeForce 7800 GT was a lot tougher than the 7900 GS/GT/GTO - partially due to better-designed cooler and also due to being made on 110 nm tech vs. 90 nm for the 7900, 7600, and 6200 TC. I'm not sure why the lager nm tech helps, but I have seen a lot more GeForce 7900 video cards with artifacts than 7800 and 6800 cards.

The issue with nVidia's cards started with the flip-chip variants of the FX series, and possibly some of the earlier notebook chips which used flip-chip technology rather than the older but more reliable wire-bond technology.

I think you said in another thread that the parts built on smaller micron technology are more failure-prone, especially when you increase the voltage and clock speeds.

Of course ample cooling would allow for those dreaded 6/7/8 series cards to last.

[Dan81]

Originally Posted by momaka

But did it work normally?

FWIW, I've done that on a modern UEFI dual-BIOS motherboard. It worked... as in, the motherboard POSTed and seemed to function OK, until you tried booting into something or installing Windows.

Yes. Actually, the only thing that wasn't supported was the diagnostic LEDs header (using the Jetway BIOS).

[RJARRRPCGP]

I thought that bumpgate was in the Core 2 era...

Especially right when even eVGA was using FZs! (7-series)

The Core 2 era apparently was the worst for Nvidia, except for the GeForce 9500 and GeForce 9400, which seem to be tough cookies.
At least my eVGA GeForce 9500 GT was a tough cookie, just slow...

The GeForce 9800 GTs seem to mysteriously drop like flies!

I knew someone who just got a GeForce 9800 GT in 2009, IIRC. One day in 2009, IIRC, it suddenly disappeared and the BIOS skipped it and booted with onboard video, the BIOS suddenly failed to detect the GeForce 9800 GT...

[momaka]

Originally Posted by Dan81

Yes. Actually, the only thing that wasn't supported was the diagnostic LEDs header (using the Jetway BIOS).

Lol, that's funny.
At least it works, though.

There was a popular mod a few years ago for the ECS MCP61xx-xx motherboards that involved flashing the BIOS on them from an aBit board and getting some overclocking features unlocked, as well more CPUs supported. Some people had a problem where their CMOS menus would "dance around" with the mod, but the board was still stable, lol.

I only came about finding that information when I got a 1st gen Phenom CPU recently and though about using it in the ECS MCP61PM-AM motherboard that came with my Gateway GT-5656 PC. Turns out, only the updated version of that motherboard could take that CPU, so I never tried the BIOS flash mod. And that PC already has an Athlon 64 X2 6000+ CPU, which can't really be overclocked much more, if at all.

Originally Posted by Wester547

The Radeon 9500 Pro I used to have, that I did droves of 3D gaming on, died 8 years ago (displayed omnipresent artifacts during 2D and 3D applications). Don't know if it was the GPU or RAM that died as the Infineon RAM ran hotter than the GPU under full load.

I've been wondering about that for a while now... but it seems it is the GPU, chip, unfortunately. And by that, I mean the BGA solder contacts between the core itself and the GPU substrate, not the BGA contacts between the substrate and the rest of the video card.

Main reason I am concluding this is because I bought a Medion Radeon 9800 XL (R360 chip) about half a year ago, which turned out to be faulty (artifacts in games, barely-visible yellow shadows behind text in Windows). It uses a much more different (and IMO improved) PCB layout/design than the standard Radeon 9800. As such, its RAM and MOSFETs run very cool. But the GPU chip, on the other hand, came stock with a tiny heatsink and ran extremely hot.

I put one of my modded heatsinks on it, and once in a while, that video card would work without artifacts. So I tested it in games, and the RAM still ran cool.

Thus, this is why I am concluding that it is NOT the RAM that is failing. Moreover, the R360 core uses the same low-K process that was developed for and used on the Radeon 9600. The 9600 cards run very cool - yet they fail all the time too. Just ask ChaosLegionaire - he has a few in his stock pile.

So IMO, all early flip-chip ATI video cards are prone to failure. I think their HD series was the first one where they finally got the issue taken care of. For nVidia, it was the 9000 series.

Originally Posted by Wester547

The issue with nVidia's cards started with the flip-chip variants of the FX series, and possibly some of the earlier notebook chips which used flip-chip technology rather than the older but more reliable wire-bond technology.

Well, the R300 was ATI's first flip-chip video card. So I don't find it too surprising that they didn't get everything right the first time.

Likewise, nVidia switched to flip-chip for the FX 5700 cards and above.

That said, GPUs made on wire-bond technology can fail too. I also recently acquired a GeForce 4 TI4600 and thought I had gotten another indestructible piece of history video card.

Well... I thought wrong! That TI 4600 was artifacting straight from the get-go. A reflow fixed it... and I added a much bigger heatsink. So let's see how long it lasts now.

Originally Posted by RJARRRPCGP

The Core 2 era apparently was the worst for Nvidia, except for the GeForce 9500 and GeForce 9400, which seem to be tough cookies.
At least my eVGA GeForce 9500 GT was a tough cookie, just slow...

Yes. nVidia finally came to their senses with the 9000 series and fixed / greatly improved on the bumpgate issue. Their cards seemed a lot more resilient after that, and probably up to the GTX 400 / Fermi series - then we started seeing issues again.

Originally Posted by RJARRRPCGP

The GeForce 9800 GTs seem to mysteriously drop like flies!

I knew someone who just got a GeForce 9800 GT in 2009, IIRC. One day in 2009, IIRC, it suddenly disappeared and the BIOS skipped it and booted with onboard video, the BIOS suddenly failed to detect the GeForce 9800 GT...

That's because many of those 9800 GT video cards come with a very under-sized single-slot cooler and they run super-hot - typically 60-65°C idle!

Under load, they hit high 80's °C. And sometimes even higher. I have a co-worker who built several older gaming PCs so we can do oldschool LAN parties in his house. After an hour of his PCs running, the room fills with that burning-hot PCB smell. At it is all coming from the 9800 GT video cards he used.

So at those temperatures, no BGA and flip-chip will last, no matter what ATI and nVidia claim to be "safe".

[Wester547]

Originally Posted by momaka

Thus, this is why I am concluding that it is NOT the RAM that is failing. Moreover, the R360 core uses the same low-K process that was developed for and used on the Radeon 9600. The 9600 cards run very cool - yet they fail all the time too. Just ask ChaosLegionaire - he has a few in his stock pile.

How cool is very cool? I would think they'd last long enough if you kept them below 50-60°C - even the bumpgate nVidia GPUs last a while at such temperatures.

Originally Posted by momaka

That said, GPUs made on wire-bond technology can fail too. I also recently acquired a GeForce 4 TI4600 and thought I had gotten another indestructible piece of history video card.

Well... I thought wrong! That TI 4600 was artifacting straight from the get-go. A reflow fixed it... and I added a much bigger heatsink. So let's see how long it lasts now.

Any BGA GPU can fail from too much heat. But the flip-chip GPUs seem significantly more failure prone than the wire-bond ones. I have GeForce 2 MX cards, without any fan or heatsink, that completely rely upon convection cooling, and all still work fine even though they've seen a ton of use (3D and otherwise) over the last 15+ years, and do get quite hot (over 60°C) under load. I also have a GeForce 2 GTS whose ADDA hyprobearing fan seized for ages (long enough for the GPU to overheat and cook the thermal compound), and it still works fine. I'm guessing that if they were flip-chip GPUs, they would have been toast by now.

Sorry momaka, I've derailed your thread yet again.

[momaka]

Originally Posted by Wester547

How cool is very cool?

35-40°C absolute max. The RAM is barely warmer than the PCB around it. And the PCB is cool too. Not like the reference 9700 and 9800 cards that have finger-burning MOSFETs and RAM.

Originally Posted by Wester547

Sorry momaka, I've derailed your thread yet again.

It was meant to be.

[Wester547]

Originally Posted by momaka

35-40°C absolute max. The RAM is barely warmer than the PCB around it. And the PCB is cool too. Not like the reference 9700 and 9800 cards that have finger-burning MOSFETs and RAM.

The GPU doesn't exceed 40°C under load? That seems awfully cool even for a fan-cooled heatsink. I could see the RAM and PCB staying that cool, but the GPU usually runs hotter, and even if the GPU's case temperature is 40°C, the core (junction) temperature is bound to be hotter.

[momaka]

Originally Posted by Wester547

The GPU doesn't exceed 40°C under load? That seems awfully cool even for a fan-cooled heatsink.

Oh sorry, I thought you were asking about the RAM on the Radeon 9800 XL.

The 9550/9600 non-pro heatsink runs about 35°C idle and maybe 40-45°C under load in open environment. In a closed case, probably 50-55°C. With my hot room, it might just hit 60°C. But they are some cool running GPUs, that's for sure. And they typically have a passive heatsink. The 9600 Pro has just slightly higher dissipation, but still a passive heatsink. I think only the Radeon 9600 XT has active cooling.

[ChaosLegionnaire]

Originally Posted by momaka

the R360 core uses the same low-K process that was developed for and used on the Radeon 9600. The 9600 cards run very cool - yet they fail all the time too.

just to clarify, only the 360 cores (r360 and rv360) used on the two xt cards, 9800 and 9600 xt respectively, use the low-k process. the low-k process improves the transistor clock gating thus allowing higher clock speeds to be achievable. the 9600 pros, non-pros and 9550 use the rv350 core which doesnt use this low-k process node. however, the rv cores use the lower 130nm process node which leads to the below point:

Originally Posted by momaka

I'm not sure why the larger nm tech helps

the high-end R cores use the larger 150nm node while the lower-end RV cores use the smaller 130nm node. the reason they do this is because the older process node is more mature thus less manufacturing problems and better asic quality. so the more complicated (and higher-end) chips are manufactured on this more mature process and the more simple, less complex (lower-end) chips use the newer funkier process node. just look up the video card list on wikipedia. u'll see this trend in many gpu family generations.

so the reason the 9600 series of chips keep kicking the bucket despite the lower tdp is probably because the semiconductor manufacturer's (it was tsmc at the time i think) newer 130nm process node wasnt mature at the time, so that would lead to reliability issues as time went on.

im now quite hesitant on purchasing used 9600 cards on ebay after having being bitten a number of times.

on ebay, first, i got a 9600 non-pro. i replaced the passive heatsink with a fanned small qq aluminium china mini-clone of the zalman vf700. even though it doesnt have a temperature sensor, i tested the heatsink on a 9600 xt which does have a temp sensor, and it wouldnt have run hotter than 40-50°C with the small qq cooler. and yet after running it 12x7 for a couple of months, it failed with artifacts in 2d mode when watching tv on my pc.

next, i got another 9600 xt with an arctic ati silencer 2. this one lasted less than a month before i started noticing tinted yellow lines when on a white background. the idle and load temps on the silencer were 38-45°C and yet it failed.

i also then got a 256mb 9550 and this one was doa. on power up at the bios screen, the picture is perfect. at windows desktop without the driver installed, the picture is fine too. but when the driver is installed, the screen becomes completely garbled, its impossible to see anything. i did some troubleshooting and disabled all agp and pci acceleration and that fixed the screen corruption so it looks like it has bad bga on one or more of the agp bus communication lines.

then i bought 3 nos fic 9600 non-pro cards for US$9.99 each on impulse. two cards had cosmetic nicks on the L shaped agp connector but were working fine. one looked fine cosmetically but when u run a 3d application, after several seconds or after about a minute or so, it either hangs or makes the driver restart the card because the hardware stopped responding but still with a corrupted display (this corrupted display after vpu recover successfully restarts the gpu seems to be a driver issue associated with the vpu recover feature, not the fault of the hardware itself).

so ugh what terrible rv350 lemons those are...

Originally Posted by momaka

The 9600 Pro has just slightly higher dissipation, but still a passive heatsink.

actually the 9600 pro (at least the oem ati ones) uses an active cooler but not a "beefy" one like on the 9600 xt. it uses a heatsink similar to the one on the 8500 or 9000 pro.

[ChaosLegionnaire]

Originally Posted by momaka

That's because many of those 9800 GT video cards come with a very under-sized single-slot cooler and they run super-hot - typically 60-65°C idle!

Under load, they hit high 80's °C. And sometimes even higher. I have a co-worker who built several older gaming PCs so we can do oldschool LAN parties in his house. After an hour of his PCs running, the room fills with that burning-hot PCB smell. At it is all coming from the 9800 GT video cards he used.

well one quick solution to prolong the life of these cards would be to mod the video bios to reduce the core voltage. if the video bios has it, it should come with 3 or 4 preset settings for the core voltage: 1.0v, 1.05v, 1.1v and 1.15v. for those single slot coolers, using the most conservative voltage of 1.0v is recommended.

however, the core clock and shader clock would have to be reduced as well. following the settings for the 9800 gt green edition, the core clock would have to be reduced to 550 mhz and the shader clock reduced to 1375 mhz. however, the benefit is that the tdp is drastically reduced to around 75w, making it run much cooler and last much longer. this also makes the cards smell much nicer without the burning pcb smell, dont u think?

[momaka]

Originally Posted by ChaosLegionnaire

well one quick solution to prolong the life of these cards would be to mod the video bios to reduce the core voltage.
...
however, the benefit is that the tdp is drastically reduced to around 75w, making it run much cooler and last much longer. this also makes the cards smell much nicer without the burning pcb smell, dont u think?

Yeah, on newer cards, it's always a good idea to try to reduce the stock core voltage.

However, I haven't talked to that guy in over a month. Not sure if he still has those PCs or what he is doing. Also, I don't think he cares enough to play with volt-modding them, even if I can get him convinced that the cards will last longer. People these days don't care much about old hardware like that like we do. :\

That aside, I need to get back in shape and post the rest of the motherboard recaps/diagrams I have for my other motherboards.

[Stefan Payne]

Originally Posted by momaka

Well, the R300 was ATI's first flip-chip video card.

Are you sure?
ATi had some obscure mobile packages with Memory on Package - much like HBM...

[momaka]

Originally Posted by Stefan Payne

Are you sure?
ATi had some obscure mobile packages with Memory on Package - much like HBM...

Hmm. Good point.
I guess the R300 was the first flip-chip desktop video card. Not sure if there was anything mobile before that. That said, I don't remember if the ATI M3 in my Dell Latitude C600 laptop had a flip-chip video IC or the oldschool wire bonds one.