Burnt my video card !!!

[rajaiitm]

an addition to the infamous xfx line of video cards .. .. i started ocing my new xfx geforcefx 5700le 128mb card.... it was one of the best overclockers i had ever seen untill an 8-pin DIP mosfet burnt out!!! is this expected from video cards to happen ??? someone please let me if they have experienced something similar..
as far as the oc ing goes .. the default speeds were 225 (2d core)/ 250 (3d core)/ 400(mem) .. ocing the 3d clock didnt give much returns for aquamark .. but 2d clock gave massive improvement .. i oc'ed to 490/490/490 .. (the mem maxed at 490 and i was trying to check for max core clocks) ........ this resulted in my aquamark graphics scores to increase from the default 1500 to 2500 !! at this point however the screen blacked out and i smelt the burnt chip.. i thought i'd burnt the core since it has a passive heatsink ..
but guess what the gpu and ram chips were safe .... an 8-pin (abt 3mm*3mm sized chip) mosfet got roasted !! the seller wouldnt take it back .. so i got that chip replaced for 350rs ~$10.. it worked fine until i started playing quake ...this one burnt up at default speed!! aaaaaaaaa ... it looks like the guy replaced the original with a mosfet of lower power rating (i somehow managed to get some info on these mosfets from net but im not sure as i cant trace the manufacturer... some really vague make) i'm still waiting to get the card back after giving it back a second time...

so .. someone plz let me know if mosfet burn up's are a known issue for video cards ??

 

[SomeDude22]

skipping lines is everyones friend

 

[Avalon]

Just goes to show that you need to learn up a lot of background knowledge on overclocking before you tackle it head on to the extreme. That's cool that you can get it fixed for $10, though, and I'm sure it was a learning experience.

 

[SomeDude22]

i think aquamark doesnt benifit because the card doesnt have full DX9 support

 

[blckgrffn]

Oh, the card has full DX9, just the somewhat shady/"better" implementation with 32 bit precision, etc...

 

[rajaiitm]

Originally posted by: Avalon
Just goes to show that you need to learn up a lot of background knowledge on overclocking before you tackle it head on to the extreme. That's cool that you can get it fixed for $10, though, and I'm sure it was a learning experience.

Sure, it was a learning experience .. the hard way though! what a way to greet my first graphics card !!

but one thing i need to know .. has this happened with oc'ers before?? or shall i say XFX has compromised with some cheap components ... and assume other makers would be better off and make sure that if something is going to blow up its going to be my gpu first !!

 

[cbehnken]

You are claiming that XFX is using bad components when you have problems after overclocking your graphics card nearly 100% ?!

Isn't that like running your street car at 12000 RPM and then bitching when the engine blows?

 

[rajaiitm]

Originally posted by: cbehnken
You are claiming that XFX is using bad components when you have problems after overclocking your graphics card nearly 100% ?!

Isn't that like running your street car at 12000 RPM and then bitching when the engine blows?

true ... but when i was expecting my engine (referring to my gpu) to blow up ... the fuel pipe gave way (the mosfet) .. this is what im bitching abt!.. frankly it wasnt something i expected .... i was waiting for artifacts to show up hoping the gpu would become unstable at some point .. but this thing blew up!!

 

[SomeDude22]

remember


if it aint broke dont mess with it

too late for you though

 

[impemonk]

4 words: knowing is half the battle.

 

[otispunkmeyer]

ho ho ho overclocking with a passive heatsink wasnt too smart lol, but least the gpu didnt go first

 

[mindless1]

this is what im bitching abt

Well if they all ran safely at very high o'c, then don't you think they'd come out of the box that way?

First and foremost, know what overclocking means to all parts. yes it will blow 'fets, and burst caps too in far shorter time. When i highly o'c cards (especially when volt-moddin'em) I replace the caps with higher spec parts and put 'sinks on the 'fets... not AFTER they seem to be having trouble but right away, right after card passes extended test runs at stock speed (so I know the card arrived in working order in case it needs RMA'd).

One cannot just switch one mostfet to a different make. It's not just their amperage but other properties too... usually they are being used in a 180' configuration into same load and when the two aren't matched, one of them is bearing a higher load than the other, so even if you'd replaced the faulty 'fet with a BETTER fet, you might then blow the other one because of the mismatch.

Heatsinks though are very important. FOr extreme o'c they may need epoxied on, frag tape may not be enough. Since I epoxy mine on anyway I even lap them a little because the 'fet surface is usually slightly dimpled.

 

[rajaiitm]

thanks for enlightening me on fets and caps ..... can u give me more info on mosfets though .. the 180' pair u were talking abt .. does that mean that the other mosfet is identical to it?? (i need to know this coz i could not note down the number of the original chip which went up in smoke.. the second 'fet has 4406 inscribed .. a check with vishay.com/mosfets tells me its max drain current is 20A .. after repair the guy had soldered a 4800 which is rated at 9A hence it promptly burnt up on starting quake at default core/mem speeds .... so if the two are twins i know which one to get) can i know what fets are normally used in video cards .. [this one was an so-8 package 8-pin 30Vds(drain-source),20Vgs(ground-source)] .. since they are used for DC-DC conversion i guess it shud be rated more or less the same for cards with similar power requirements.

abt my ocing ideology .. i followed these rules..

1) i did not increase agp voltage
2) kept increasing mem speed till i hit a limit with artifacts .. this came to 490 from 400 default
3) now with default mem i increased 3d core clock .. it went upto 500 before artifacts showed up
4) noticed that 2d clock increase resulted in gud fps increase in quake .. so started increasing that ... i didnt reach a limit .. meanwhile i took a break
5) tried to benchmark at 490/490/490 .... mosfet went up in smoke!!

assumption i made : if ram or gpu has to blow up it would show artifacts a long way before that .. so increase speeds till artifacts show up ...

mistake: the power consumption of all components together at 490/490/490 was too much for mosfet which cant show any artifacts .. so i had no clue it was going to blow up !!

big mistake: assuming that not increasing the agp voltage will leave me in safe territory for ocing and that i would not have major overheating problems!!!

out of interest ... how many of novice oc'ers look into 'fets and 'caps??

 

[mindless1]

Originally posted by: rajaiitm
thanks for enlightening me on fets and caps ..... can u give me more info on mosfets though .. the 180' pair u were talking abt .. does that mean that the other mosfet is identical to it?? (i need to know this coz i could not note down the number of the original chip which went up in smoke.. the second 'fet has 4406 inscribed .. a check with vishay.com/mosfets tells me its max drain current is 20A .. after repair the guy had soldered a 4800 which is rated at 9A hence it promptly burnt up on starting quake at default core/mem speeds .... so if the two are twins i know which one to get) can i know what fets are normally used in video cards .. [this one was an so-8 package 8-pin 30Vds(drain-source),20Vgs(ground-source)] .. since they are used for DC-DC conversion i guess it shud be rated more or less the same for cards with similar power requirements.

abt my ocing ideology .. i followed these rules..

1) i did not increase agp voltage
2) kept increasing mem speed till i hit a limit with artifacts .. this came to 490 from 400 default
3) now with default mem i increased 3d core clock .. it went upto 500 before artifacts showed up
4) noticed that 2d clock increase resulted in gud fps increase in quake .. so started increasing that ... i didnt reach a limit .. meanwhile i took a break
5) tried to benchmark at 490/490/490 .... mosfet went up in smoke!!

assumption i made : if ram or gpu has to blow up it would show artifacts a long way before that .. so increase speeds till artifacts show up ...

mistake: the power consumption of all components together at 490/490/490 was too much for mosfet which cant show any artifacts .. so i had no clue it was going to blow up !!

big mistake: assuming that not increasing the agp voltage will leave me in safe territory for ocing and that i would not have major overheating problems!!!

out of interest ... how many of novice oc'ers look into 'fets and 'caps??

Yes the 2nd mosfet should be identical. I'm not familiar with the exact circuit on that card but there should be a driver (controller) powering a couple of them, right before inductor(s). YOu should get exact same fet, NOT just one with same amperage spec, OR replace both of them. 20A is quite high for SO-8 though, or maybe I'm mistaken about the newer high-spec parts and should brush up some.

In the past one could just increase the speeds and note problems, but we're seeing modern cards with limitations- they're built to a budget with these surface-mount parts which have finite current limits but the core speeds and memory (amount and speed) go up. Basically parts are being ran closer to tolerance in some cases. As for your specific card, I can only speculate, don't have one.

I'd imagine too few OC'ers look into power regulation components but how many look into those on motherboards? You have to start somewhere I suppose, there would be no need to look into component problems until AFTER one is overclocking and hence facing these situations.

Frankly I think card manufacturers could spare the PCB real-estate and raise the cost by $1 or 2 to use more robust parts, but even on high-end cards somebody is making silly design decisions, using SO-8 parts then factory heatsinks... heatsinks on TOP of an epoxy-cased part are certainly not optimal, just what one is stuck with after the design mistake was made.

 

[Keysplayr]

Originally posted by: mindless1

this is what im bitching abt

Well if they all ran safely at very high o'c, then don't you think they'd come out of the box that way?

First and foremost, know what overclocking means to all parts. yes it will blow 'fets, and burst caps too in far shorter time. When i highly o'c cards (especially when volt-moddin'em) I replace the caps with higher spec parts and put 'sinks on the 'fets... not AFTER they seem to be having trouble but right away, right after card passes extended test runs at stock speed (so I know the card arrived in working order in case it needs RMA'd).

One cannot just switch one mostfet to a different make. It's not just their amperage but other properties too... usually they are being used in a 180' configuration into same load and when the two aren't matched, one of them is bearing a higher load than the other, so even if you'd replaced the faulty 'fet with a BETTER fet, you might then blow the other one because of the mismatch.

Heatsinks though are very important. FOr extreme o'c they may need epoxied on, frag tape may not be enough. Since I epoxy mine on anyway I even lap them a little because the 'fet surface is usually slightly dimpled.

So you feel ok with first extensively testing your card (before any mods) at stock speeds. If it does well, then you cut off capacitors and solder your own on. If It blow up you would still RMA it? Not very honorable advice your giving here. You mod it, it's yours forever, unless you sell it.

 

[mindless1]

So you feel ok with first extensively testing your card (before any mods) at stock speeds. If it does well, then you cut off capacitors and solder your own on. If It blow up you would still RMA it? Not very honorable advice your giving here. You mod it, it's yours forever, unless you sell it.

Of course I wouldn't RMA it and it is offensive that you stoop so low as to suggest I'd RMA it then after moddin' it. Perhaps YOU would do that, since you are the one who thought of it.

"You mod it, it's yours forever" is utter nonsense. It's yours until you sell it just like it was if unmodded, except a modded card (if done right) is more valuable. Maybe some feel it's worth less, but those would be the people who don't appreciate or mod cards, for each type of user there is a product they want, a market for it.

If we want to talk about dishonorable, that would be what often happens instead, that someone (doesn't mod their card), overclocks it, then trys to RMA it when it fails. Once a card is overclocked you are commmitting fraud by RMA'ing it, unless the product description very specifically specs it to overclock to that specific speed.

MOddin' it makes no difference in a moral or legal sense, it was the overclocking itself that mattered and towards that end a knowledgeable user will make their card fit for the target overclock if it wasn't that way out of the box. Note I wrote "knowledgeable", I don't advocate someone mod a valuable card if they've never done it before or dont' know how to make good (replacement) part selections, are comfortable reading component datasheets and understand them.