rtx 3050 very weird behaviour, only works slot 2

[davetheshrew]

Hi I would like some help please because for the life of me I cannot pinpoint this issue. The card gets all its voltages, resistances are fine but in the pcie top slot the card wont detect but it works perfectly in the second slot so Im guessing it works fine in pcie 8x but wont work at all in pcie 16x. Has anyone ever had this issue?

On stand by with my multi ready to answer anything. I want this gpu for my son for xmas so Im really hoping It can work fully in the top slot. Thanks

[davetheshrew]

Re: rtx 3050 very weird behaviour, only works slot 2

Update: Ive taped up all the pins that go beyond 8x and the card boots perfectly on the top slot so there must be a short somewhere along the line, so far I cant figure it out but Im glad at least it runs 8x. More than willing to pull it out for exploration if anybody wants to chime in to figure this issue out.

[Elektrolite]

Re: rtx 3050 very weird behaviour, only works slot 2

Update: Ive taped up all the pins that go beyond 8x and the card boots perfectly on the top slot so there must be a short somewhere along the line, so far I cant figure it out but Im glad at least it runs 8x. More than willing to pull it out for exploration if anybody wants to chime in to figure this issue out.

Please post the PCI-E versions (3 or 4) and number of data lanes (8x or 16x) for the motherboard, CPU, GPU, slot 1 and slot 2.

[davetheshrew]

Re: rtx 3050 very weird behaviour, only works slot 2

Please post the PCI-E versions (3 or 4) and number of data lanes (8x or 16x) for the motherboard, CPU, GPU, slot 1 and slot 2.

Massively appreciate the reply thank you. Ive replicated this issue on three test systems so its the card. The mobos I used were P87, x570 and h110 so a healthy mix of various gens, all produce the same behaviour.

Am I correct in thinking the next best points of interest for me should be in diode mode and check every transmitter and receiver lane after pin 48 and maybe PRSNT and RSVD?

[ktmmotocross]

Re: rtx 3050 very weird behaviour, only works slot 2

have some bad pcie lines beyond x8

[davetheshrew]

Re: rtx 3050 very weird behaviour, only works slot 2

yeah so ive tested all the transmitter lane pairs and they read 1.8074v , perfectly normal, the last PRSNT and pin 82 both lead to unpopulated pads, no missing components noticable on microscope and checked against board pictures on techpowerup (here https://www.techpowerup.com/review/p...ormx-oc/3.html ). So I think now its a short receiver lane, I get no reading on any of these because they are direct to gpu core as far as i understand?

As far as Im aware this card has never worked on pcie 16x slot so is it possible this card might have a shorted solder ball under gpu or something? no idea how this got passed quality control test if thats right.

I will see if any of the traces lead to any caps or resistors on route to gpu, might be something bad on the way.

[dellxps15]

Re: rtx 3050 very weird behaviour, only works slot 2

check cpu pins... bend cpu pins can give that kind of error

[DynaxSC]

Re: rtx 3050 very weird behaviour, only works slot 2

If you have a short on the pci lane, it's almost sure it's inside the gpu (RIP). If the pci lane is open, it might be a contact issue pcb pad - gpu, then reflow or better reballing/reseating of gpu can help. You can also try bending the pcb (dismount the factory cooling, attach some small cooler to GPU, try to boot the card applying some constant bending pressure, do it with care, limit the power on time to keep GPU temperature safe), with luck it can establish the contact again for a moment, but usually not permanently. Short between balls under gpu almost impossible (card would'nt pass QA), unless somebody did reflow/reballing/reseating gpu before or you work in a metal machining workshop and metal particles are flying in the air

BTW I do not know how you can read 1,8V on the lanes, the normal value is around 0.25V (address, "A" side) or 0,4V (data, "B" side), so maybe you do it not correctly. You should measure in diode mode, positive probe always connected to ground on pcb, negative probe to pci data or address line - each one separately (omitt the capacitors on the address lines).

For address lines you can also check the 32 pcs 220nF capacitors for shorts or bad contact - "A" side of the pcb (inspect with magnifier glass for almost ripped off pcb ?), but your problem is more in the data lines I guess, as bad address lines would effect in not working card at all.

"A" side is the one towards the CPU on the mobo PCB.

There are also GPU pci interface test cards available on "the" chinese portal, with such a tester you can check all 64 lanes at once, LED's show bad lines. Saves a lot of time, if you repair lot of cards.

[davetheshrew]

Re: rtx 3050 very weird behaviour, only works slot 2

If you have a short on the pci lane, it's almost sure it's inside the gpu (RIP). If the pci lane is open, it might be a contact issue pcb pad - gpu, then reflow or better reballing/replacing of gpu can help. You can also try bending the pcb, if this changes something, with luck it can establish the contact again for a moment, but usually not permanently. Short between balls under gpu almost impossible, unless somebody did reflow/reballing/replacing gpu before or you work in a metal cutting workshop and metal particles are flying in the air

BTW I do not know how you can read 1,8V on the lanes, the normal value is around 0.25V (address, "A" side) or 0,4V (data, "B" side), so maybe you do it not correctly. You should measure in diode mode, positive probe always connected to ground on pcb, negative probe to pci data or address line - each one separately (omitt the capacitors on the address lines).

For address lines you can also check the 32 pcs 220nF capacitors for shorts or bad contact - "A" side of the pcb (inspect with magnifier glass for almost ripped off pcb ?), but your problem is more in the data lines I guess, as bad address lines would effect in not working card at all.

"A" side is the one towards the CPU on the mobo PCB.

Thanks very much I will have a look and get back to you. For diode mode I was in diode mode with negative on data and all of them read 1.8, but will check again. I will also try to flex the pcb slightly as you say and see if I get a post, if so I will reflow it and retest, I have a pretty good preheat table and a decent hot air so dont worry no 650c hot air guns or ovens going near it!