Tale of the dodgy HD3870

[Th3_uN1Qu3]

Yesterday David called me to tell me there's a problem with my computer (which resides at his place for the time being, that's where we record our music). It would boot into windows but restart just after it finished loading everything.

For the record, the configuration is (and has been for the last 4 years): Gigabyte P35-DS3, E6550 @ 3.61GHz under Scythe Mugen, 8GB Corsair XMS2 1066, Sapphire HD3870 512MB DDR3 under Accelero S1, Seasonic SS-500HT power supply, Akasa Zen case.

Resetting everything to stock or running with less RAM didn't work, so that pretty much ruled out the mobo, CPU and RAM. Safe mode worked however... and it would sometimes fail IntelBurnTest when i would set it to test over 2GB. Tweaking the timings a little made it stable under IBT again, so we ruled out everything but the video card. We didn't have another card to test however. I disabled the HD3870 in Device Manager and still no go in normal mode... that threw me off, and i thought it couldn't be the video card either. Disabling all startup items had no effect either, it would reboot after exactly the same amount of time every time. I was stumped.

So i took the system back home. Tested the hard drives, no problem. Edit: No trouble in the PSU either. Just as Windows loaded in normal mode, i started IntelBurnTest. Strangely, the system didn't reboot this time... I was able to get into CCC and notice that the overdrive settings got reset. That was the first clue that it definitely had to do with the HD3870, as this particular card has a sweet spot and won't work at anything higher, or lower than that. I loaded up the settings off the back of my head and it worked at the desktop, but would still reboot after a minute or so of FurMark.

This card was dodgy since new. It had one of those god-awful single slot coolers, and wouldn't even run stock before i vmodded the RAM. Since i got the Accelero S1 it didn't have anymore problems.

I took the card out and noticed one of the ramsinks wasn't seated properly because the heatpipe of the cooler was pushing on it. The S1 has two sizes of ramsinks and a smaller one was supposed to be in that spot, but i had chosen a different arrangement based on heat distribution. I took the cooler off, grinded that ramsink a little and it now fit fine. I then focused my attention on the caps - two KZG 680u 16v on the input, polymers everywhere else. The KZG tested fine tho... I noticed that there was a unfitted spot for another poly cap for the Vcore, but i couldn't get the solder to melt and knocked off a really tiny SMD cap in the process. Don't know what size it is but it's smaller than 0603. Thankfully i knew they're all 100n bypass caps so i pulled one from a mobo and soldered it in. In fact, i pulled 3 and lost 2 of them. I then proceeded to drill thru the solder and installed the extra poly cap. Put the Accelero back on, and powered 'er up. I also decided to set the vmod to the minimum (1.9v, was set to 2.06) to see what would happen. Stock voltage for the DDR3 memory was 1.8v btw, and it wouldn't run any game more than 15 minutes with that...

Put the card back in, no reboot. Good. Started up FurMark, no crash but after a few minutes then artifacts started flickering every now and then. Mkay, time to touch up the memory vmod. 1.95v... still flickering. 1.97v... perfect.

The card is now 1 hour into Furmark with no trouble. Long story short, i will never buy Sapphire again. They are the #1 supplier of ATi cards, and since the yields are the same for everyone, you make more cards, you get more duds. I still don't understand what starting IntelBurnTest before everything loaded had to do with the system not rebooting anymore, but i'm sure the 4th Vcore poly cap (which should have been there in the first place) is what brought stability to the card.

[Scenic]

Never had probs with Sapphire cards before. My old dual slot HD3850 is still working fine, as is my not quite so ancient HD6870

I did have lots of trouble with XFX though. Overhyped crap. And to this day they use crapcaps on anything that isn't top of the line. meh..

One thought I had while reading that post though: if it was dodgy right out of the box, why didn't you RMA it?!

[Th3_uN1Qu3]

Originally Posted by Scenic

Never had probs with Sapphire cards before.

Well my old 9800 Pro is still alive and kicking and with some pretty sick vmods on it... but that's quite a few generations back

Originally Posted by Scenic

One thought I had while reading that post though: if it was dodgy right out of the box, why didn't you RMA it?!

Because there's no such thing as customer RMA here. You take it back to the store and they handle the RMA. I'd been thru 2 8600GTs before this one, and when the 2nd 8600GT blew (brand new out of the box, it worked for 1 hour and then bye bye), they took over 3 weeks to bring me my 3rd card. I sold that 3rd card and bought the HD3870. It didn't work properly, but i didn't want to be without a computer for 3 weeks again.

Also, i thought my PSU was crap (a Spire 420W) so i replaced the PSU hoping that the card would work then. The Spire PSU wasn't stellar indeed, same old halfbridge platform and CS caps, but it was built well (no missing parts, big heatsinks, 1000u primaries, TO-247 13009s, 120mm fan and passive PFC) and lasted for 3 more years in another computer till i did some radical mods and turned it into a +/-30v 600W amplifier power supply last year.

[mariushm]

Sapphire is the official partner for AMD - they get preferential treatment, they get the designs ahead of the others by a few weeks, they do the OEM designs for AMD and guarantee quantities and so on.

The Radeon 4850 I have has the same issue with overclocking - the memory and cpu is VERY sensible to what frequency you set, I think there must be a certain ratio between them to work. I've set once the cpu by about 100 Mhz higher and the memory about 50 mhz lower just for test and basically i got the card in a sort of out of sync making it blue screen the system.

Everytime I went after that in Control Center , the moment it loaded it applied those preset values without giving me a chance to change them and a couple of seconds later the system would blue screen. So I just gave up and uninstalled the Control Center - not like I use it much anyway.

Here's the card with me mounting the Accelero s1 rev 2 on it - the same issue with the heatsink on a memory chip... but I bent the fins from the start :

http://www.youtube.com/watch?v=y8pcK...ure=plpp_video
http://www.youtube.com/watch?v=sMKUH...ure=plpp_video

ps. about the RMA thing ... things have changed a bit lately here... there's a law now saying that if you bought something online you can return it within 10 days from the moment you received the package and get your money back no questions asked.
Also, the local stores here where I live usually replace a broken card if it dies within the first 2-3 weeks and they also offer you some sort of "Trade up" where you pay the difference for a better card if you're not happy with it in the first 30 days or so.

When I got my first computer, I got it with an S3 Trio 3D, upgraded to S3 Savage 3D that had some issues with the bios and couldn't do any 3d stuff, got it exchanged at the store with a Savage 4.. developed the same issue a week later and finally the store owner gave up and gave me a nVidia Vanta for no extra charge (it was more expensive back then).

[Th3_uN1Qu3]

Originally Posted by mariushm

The Radeon 4850 I have has the same issue with overclocking - the memory and cpu is VERY sensible to what frequency you set, I think there must be a certain ratio between them to work.

I don't think there's such a thing as a "golden ratio", i think the board layout is dodgy.

It's unfortunate that i no longer have a scope, but things are looking good on the horizon and i'll be able to afford some proper test gear (and a new video card ) not too far from now, and with the mods this card has on it i won't be able to sell it anyway, so i promise to bench the hell out of it one day and show you all the signals i can capture, what goes wrong and where, and how it can be fixed.

Adding extra bypass capacitors to improve overclocks (not only as in big filter caps, but also as in those SMD ones) isn't unheard of, so clearly there must be something wrong that those caps are fixing. Sometimes, adding a cap is more helpful than increasing a voltage, AND without the extra power consumption and heat!

Edit: Heh, S3... brings back memories. Remember the Virge, aka the "3D Decelerator"? I was just a kid back then continually breaking Win95 and crying for my dad to fix it. And one day dad left the CD and boot floppy at home and i learned to install it myself. The rest, is history.

[lti]

Originally Posted by Th3_uN1Qu3

Remember the Virge, aka the "3D Decelerator"?

I have one of those. I think mine is a Virge GX with 2MB of video RAM. It is a good 2D decelerator too. My Compaq's SiS 5598 onboard graphics have better 2D performance.

[RJARRRPCGP]

Originally Posted by Th3_uN1Qu3

Long story short, i will never buy Sapphire again. I still don't understand what starting IntelBurnTest before everything loaded had to do with the system not rebooting anymore, but i'm sure the 4th Vcore poly cap (which should have been there in the first place) is what brought stability to the card.

Sapphire has been caught using a POS heatsink before.

And, Linpack failures= Usually unstable processor.

And now I'm wondering if that's why the second hand Radeon 9800 Pro I got in 2007 had video corruption. Even the BIOS screen was corrupted!

But, it seemed to go away when I moved the card a little, only for it to start again! Seemed to be bad contact somewhere. Looks like maybe crappy soldering. Shame on you.

[Th3_uN1Qu3]

Originally Posted by RJARRRPCGP

Sapphire has been caught using a POS heatsink before.

You can say that again.

Originally Posted by RJARRRPCGP

And now I'm wondering if that's why the second hand Radeon 9800 Pro I got in 2007 had video corruption. Even the BIOS screen was corrupted!

I'll be getting a 9800 XT for free, with the only issue being a seized fan. Not sure who made it tho, we'll see when i go pick it up.

Originally Posted by RJARRRPCGP

And, Linpack failures= Usually unstable processor.

Moar vcore will fix (almost) anything.

[RJARRRPCGP]

It looks like the solder on my Radeon 9800 Pro hairline-fractured when being used with cold air blowing into the case. And it wasn't this sub-zero-cooling stuff.

I came back to the PC to see the screen looking like a bad game cartridge!

[momaka]

I think many of those older Radeon 9700 (R300) and Radeon 9800 (R350) video cards just had inadequate coolers. I've seen quite a few people complain about getting artifacts after some years of use. A reflow of the GPU/RAM should fix it easily. I have 3x 9700's come back form the dead so far. 3 more are awaiting reflow. They use leaded solder, too.

[ratdude747]

Originally Posted by momaka

I think many of those older Radeon 9700 (R300) and Radeon 9800 (R350) video cards just had inadequate coolers. I've seen quite a few people complain about getting artifacts after some years of use. A reflow of the GPU/RAM should fix it easily. I have 3x 9700's come back form the dead so far. 3 more are awaiting reflow. They use leaded solder, too.

OT: Should I replace the cooler on my 9800 pro? I overheated the graphics a while back... no issues so far after fixing the half-seized fan (drop of ATF).

[Th3_uN1Qu3]

I've been using a Pentagram XC-70 for years on my 9800. Then again, my card is vmodded and overclocked like mad.

[Toasty]

I have to agree with Scenic. I never had any problems with my Sapphire's either.

I have 2 of these cards, both used when I got them. They have been poly-modded now and had all the open cap spaces filled in. Never had any problems with them before I did that. Both were used for folding for about 6 months before the poly mod with no issues. Sorry I can't speak to Vmoding them.

[momaka]

Originally Posted by ratdude747

OT: Should I replace the cooler on my 9800 pro? I overheated the graphics a while back... no issues so far after fixing the half-seized fan (drop of ATF).

IIRC you had the Saphire-made one, correct? If so, I'm not sure. All I know is the stock ATI-made coolers used on the Rad. 9700 and 9800 are somewhat inadequate and they often used that yellow gummy gunk that was supposed to be thermal compound. It would eventually harden and not transfer much heat to the heat sink, cracking the joints on the GPU/RAM over time. Yours used silver compound IIRC, so that's much better already. Definitely consider fully taking apart and cleaning the fan, though. Drop of ATF is only a temporary fix. Need to clean the shaft and sleeve bearing.
Also consider RAM heat sinks if space allows it.
Don't worry that it overheated once, though. Those cards use leaded solder, so they can take a lot more abuse from heat cycles than lead-free solder (hence why they often lasted over 3 years, even with partially seized fans).

Originally Posted by Th3_uN1Qu3

I've been using a Pentagram XC-70 for years on my 9800.

They have this on ebay for <$3. I wonder how well it performs. Years ago I bought a cheap replica Radeon 9800 Pro cooler on ebay for $7. That's actually what I'm using right now to test the Radeon 9700's. The surface that goes over the die of the GPU is a little rough and the fan is sleeve bearing and somewhat slow, but it does a decent job keeping the video card cool.

[ratdude747]

The cooler is a clone of the ATI one. I swapped the dried up silverish stuff for some fresh silver compound.

I ran the ATF though the bearing, drained it, and refilled as a way to clean it... unlike the other fans I have worked with, this one had no obvious clip on the fan shaft...

I kept the heatsinks stock due to space constraints... every card slot on the board are in use; I put the smallest card (a 1394 card) beneath the GPU...

heat wise, the card didn't get too hot once the fan was fixed... no issues even under full load.

---

that cooler you linked is junk... that fan is trash.

[Th3_uN1Qu3]

Originally Posted by momaka

They have this on ebay for <$3. I wonder how well it performs.

That one looks exactly like the XC-70 i have, minus the sticker and blue LEDs... Also mine came with a fan controller. And yes the fan did crap out after 3 years or so IIRC, but a relube got it going once again. How much longer are you planning to keep a Radeon 9800 anyway?

[Wester547]

Originally Posted by momaka

Need to clean the shaft and sleeve bearing.

All the Radeon 9500/9700/9800 cards (the stock ones from ATI) use sleeve bearing fans? D: That only worries me because I've read that sleeve bearing fans don't cope well with being horizontally mounted and that they may seize early in such a position. Though, I suppose you could always replace the fan on the card with a better one (ball bearing would be good). I'm guessing that there's a 45mm fan on the card (the one from ATI) or something like that. And I wouldn't mind keeping R300/R350 boards myself. For their time they were amazing performers and well usurping of anything in competition with them. I feel that those series of cards aged very well, many generations beyond their prime, even though they obviously wouldn't be suited to recent games/apps or systems.

[ratdude747]

Originally Posted by Th3_uN1Qu3

That one looks exactly like the XC-70 i have, minus the sticker and blue LEDs... Also mine came with a fan controller. And yes the fan did crap out after 3 years or so IIRC, but a relube got it going once again. How much longer are you planning to keep a Radeon 9800 anyway?

considering that it is in a supermicro sytem, quite a long time...

[momaka]

Originally Posted by ratdude747

that cooler you linked is junk... that fan is trash.

Have you actually bought one like that?
Like I mentioned, I got a cheap Radeon 9800 clone heat sink from ebay/China (the seller is actually long gone now) but it works fine and the quality isn't that bad. I'm actually curious to buy one of those that I linked to above, just to try it out. Might grab some $1 10g jar of thermal compound to go with it as well (it's got free shipping) .

Originally Posted by Wester547

All the Radeon 9500/9700/9800 cards (the stock ones from ATI) use sleeve bearing fans?

Either 1 ball bearing with sleeve bearing for the rear or full sleeve bearing... I fixed a partially seized fan on my friend's 9800 Pro about a year ago so I don't remember exactly. Might have been a single sleeve bearing, though. Those fans are very skinny.

[ratdude747]

Originally Posted by momaka

Have you actually bought one like that?
Like I mentioned, I got a cheap Radeon 9800 clone heat sink from ebay/China (the seller is actually long gone now) but it works fine and the quality isn't that bad. I'm actually curious to buy one of those that I linked to above, just to try it out. Might grab some $1 10g jar of thermal compound to go with it as well (it's got free shipping) .

Ive seen many more of that type fan dead or useless than good... even new ones have been bad.

the fans are slow and poorly made... go ahead and try it, but don't expect quality.