P55-UD6 socket burn

[cirthix]

cirthix, yes, but not every vodka and weapon from russia is good
what they really do well in russia and eastern europe are... women :p
ahum, enough ot

i thought we already ruled out that its a gb thing cause other boards are affected too?

???? (my first post in this thread)

[kiwi]

I think he confused you with me, I made that joke. Yes, women are nice here and cheap

[saaya]

This is still speculation really. No one has actually had a look at SF3D's or my board apart from the pictures posted here.

well isnt that obvious?
of course its speculation!
since quite some people read this and some even posted that they are worried about getting this or another 1156 board, i figured it might make sense to sum up what might have caused it and to list what people could do to avoid this from happening... the lowest vcore that you guys used and had it blow/melt a pad was 1.5v vcore though, right? so this def wont affect normal end users and even normal overclockers...

I don't think Intel can verify that particular cpu has been overclocked or not. Or they can?

they definately can, and amd too... whether they do it or just replace processors without checking is another question

cirthix, kiwi, my bad... no idea why i confused you with each other

[kiwi]

Are you sure? Do they store the maximum frequency ever in some registers? If so what is the purpose? I am just curious

[saaya]

Are you sure? Do they store the maximum frequency ever in some registers? If so what is the purpose? I am just curious

no im not sure, and if they store it its obviously for statistics or rma control or researching why a certain cpu failed.

[Raja@ASUS]

Wprime over 5180 @ -102 Celsuis on an 870 with 8 threads yesterday and checked the socket today and found this;



This is the ASUS Maximus III ROG this time using the same 'Foxconn socket' as the GB board. Same corner of the socket as the others too - coincidence? CPU pads were clean for those wondering.The CPU that was in the ASUS board this time has some discoloured and mishapen VCC pads, although none have gone missing completely. Power draw was in the region of 20.5 amps on the EPS12V line. Factor 15-20% switching loss, and also some power to the VTT rail from that figure - so what around 160-170w to the CPU @ -100 is enough to light up the sh1thouse in some cases. Ran the same CPU in the EVGA board at 5.2 the day before and no problems (it uses a CPU socket from LOTE I think)

later...

[eva2000]

Wprime over 5180 @ -102 Celsuis on an 870 with 8 threads yesterday and checked the socket today and found this;

http://img8.imageshack.us/img8/7877/pic23k.jpg

This is the ASUS Maximus III ROG this time using the same 'Foxconn socket' as the GB board. Same corner of the socket as the others too - coincidence? CPU pads were clean for those wondering.The CPU that was in the ASUS board this time has some discoloured and mishapen VCC pads, although none have gone missing completely. Power draw was in the region of 20.5 amps on the EPS12V line. Factor 15-20% switching loss, and also some power to the VTT rail from that figure - so what around 160-170w to the CPU @ -100 is enough to light up the sh1thouse in some cases. Ran the same CPU in the EVGA board at 5.2 the day before and no problems (it uses a CPU socket from LOTE I think)

later...

wow, foxconn cutting back of on quality during these financial crisis times ?

[Raja@ASUS]

Not sure if its just the socket as the first CPU I had burned out a pad in the EVGA board although the socket took zero damage. The thing with the LOTE socket seems to be that it can handle the current, but that does not protect the CPU if there's a problem at the CPU side of things.

[Raja@ASUS]

I just did a quick VCC pin check - looks like around ~ 175 VCC pins to i5 and around 250+ to i7. Power draw between both is very very close, i5 just is'nt designed for this kind of use under 8 thread loads (not that we did not already know...lol).


I should add at this point, if you run 8 thread Linx over 4.4GHz at ambient temps the power draw of some CPU's is very much the same as what I was pulling at near 5.2 @ -100. Only you don't have the luxury of the cooling preventing a serious burnout. So next time one of the stability boys harps on about needing to see 8 hours of Prime etc on 4.3GHz + OC's using the i5 platform , tell him to 'F*** O**!'.

[chew*]

Thx for the share Raja.

My 860 comes this week.....still not sure if i want to toss it in UD4P or UD6.

I think the phase will matter none at all as both use the same socket.

[borec]

raju: please what Vcore you used when socket "burn"? Thanks.

[Raja@ASUS]

raju: please what Vcore you used when socket "burn"? Thanks.

VID under load was 1.57V for 5.19GHz. It's the current that matters most, you can pull higher current even with lower VID. If the CPU consumes more than 160w to VCC then consider yourself to be in the red zone for some CPU's/sockets. Load up i5/i7 with Linx at 4.4GHz and it's likely you will pull this much on some CPU's even if VID is under 1.40V. Or if benching, use a clamp meter on EPS 12V, if you see 20 amps+ under load, you best back off unless you don't mind the risk. Other than that, there may be further precautions that I don't know about. It depends what the other guys were running when they burned out their boards/CPU's.

[borec]

yes it is clear for me...
this is big problem potentially for all mobos with foxconn sockets, not only GB...
I think, EVGA classified with a more gold in socket, will have not problem...

S*IT Intel with his 1156 layout!

[Raja@ASUS]

yes it is clear for me...
this is big problem potentially for all mobos with foxconn sockets, not only GB...
I think, EVGA classified with a more gold in socket, will have not problem...

S*IT Intel with his 1156 layout!

The extra pin thickness of some sockets may help a bit, but you still get burned CPU VCC pads even when the pins are fine. Bottom line is you have fewer VCC pads on i5 with current draw like i7 at the same frequency. On boards with thicker socket pins, the CPU pads stand a good chance of more damage than the pins (the thinner pins get hot and bend away from the CPU), so its kinda lose lose either way if you wanna push the platform real hard.

[AndreYang]

My R3F is still alive when I test it 5.4G+ 32M.

[Raja@ASUS]

My R3F is still alive when I test it 5.4G+ 32M.

32m is like 7 amps of draw max - no wonder you are ok Try 1024M @ 20+ amps

[SF3D]

It was obvious all the time, that amps can burn the socket that way. I was testing Wprime and Vantage as well.. low frequencys at that point, but still a lot more load for the board.

I think we should just test lighter applications

[Eeky NoX]

Seems to be clearer now whish for every body that should be fixed ...so the lesson is to be carefull now ^^
Thx for the share Raju (and I really liked your post up ahead : so true lol)

[Raja@ASUS]

32m, 3dmark, 01, 03 and 05 should be fine.

I only posted today because we had a few guys speculating in this thread that we don't know how to insulate or have too much lint in our sockets. I give the prize to the 1 VCC pin question myself

[thijs]

Just to be clear, is this risk present at the X58 i7's aswell? or does it have way more VCC pads?
Im not around for that long already, but no reports have been made regarding burned pads with the X58 i7's right?

[dinos22]

i dont know if i havent paid attention before or what but these P55 board pins do seem a bit more fragile or easier to bend

hopefully its just a socket replacement

[hipro5]

This socket and the shape of it's pins suck...I told you....
There are more to come out of it in time......

[Raja@ASUS]

The LOTE socket has better pins (the EVGA sockets are better), but you guys don't forget you have 75 less VCC pads on i5. SO just looking at the socket pins shape as the cause is not going to cure this issue. i7 has over 250 VCC pins while i5 seems to have about 175. ALmost same current draw with less pads. I lost one VCC pad even on the LOTE socket (which by the way has the same kind of pins we saw on the i7 sockets). So failing a complete redesign of the pin and contact area (pad size/density/count), there is not too much that can or will be done at this stage.

[saaya]

wow, foxconn cutting back of on quality during these financial crisis times ?

not really, they had problems with 1366 as well and that was before the crysis... :/
well it does what its supposed to do... see it from their point, why should they spend even a fraction of a cent more per socket if its good enough for the rated specs intel asks them for?

thats what foxconn is great at, good enough for as cheap as possible...
i dont want to defend them, for ocing their sockets seem to suck...

thx for all the testing raja! great stuff!
are you sure its the vcc pins though and not vss pins?
wasnt it the vss pins that had diagonal rows 1 pin thick and vcc were 2 rows thick? it looks like there is one burned pin, then right to it 2 pins are barely burned or ok, and then theres a clearly singed pin again...

160W only can cause this... damn...
thats def achievable on air without even trying too hard...
i checked the intel docs but couldnt find any rating as to current draw per pin... theres just a max resistance per pin value i found...

i wonder if intels specs are that tight or if foxconn actually fails the specs intel asks for...

ps, its lotes, not lote afaik?

[Raja@ASUS]

It is LOTES, just me being a bit hasty in typing and checking.

Whichever way you look at it, it's current draw related. 160+w is pushing the socket design quite hard it seems, and yes running Linx you can get to this quite easily over 4.3GHz, so people just need to be a little more cautious, no need to torture the design too much to prove its stable or otherwise.

Both times it has happened to me on two different CPU's (one was a real bad CPU for frequency only 4.8Ghz max), it has been when I have seen full load EPS 12v draw at 20.5 amps (both times different boards). PWM efficiency at these kinds of loads should be 80-85% at a guess, and then you have say 30-40w (guess) going to VTT as well, so 160-180w to VCC.

BTW even on the beefy pin sockets, the the contact point density is pretty much the same and limited by the pad design of the CPU. Real solution here would be just to have more VCC pads.