Firepro w5000 DOA?

Hardware Hub
1

Hi Guys,

I bought an AMD Firepro W5000 off of eBay and have been having a hell of a time trying to get it to work on any of my systems. The actual posting is here: https://www.ebay.com/itm/2GB-AMD-FirePro-W5000-GDDR5-DVI-2x-DisplayPort-PCI-Express-3-0-x16-GPU-TESTED/224308381126?ssPageName=STRK%3AMEBIDX%3AIT&_trksid=p2060353.m2749.l2649

I bought it because it seemed small enough to mount to the side of a Dell/Wyse 5020 as an eGPU (which I realize is in itself stupid in pragmatic terms) and seemed reasonably powerful for the size/price. I have previously tried other cards as eGPUs for this system without troubles, both a Firepro V4900 that I also got around the same time from eBay (a much older/wimpier card BUT it works out of the box accelerated in NetBSD which was my original goal when the search began) as well as an HD 7970 I had on hand, which works fine but is much larger, more power hungry, and absolutely overkill for a 1.5GHz SoC. The 7970 is the same generation as the W5000, GCN 1st.

Anyway, I have had no luck getting the w5000 to work at all on any of the machines I’ve tried, which are:

  • An Ivy Bridge i7 3770S running on a Gigabyte H77N-WIFI – my main system which runs an RX 580 day to day
  • An Acer 4820TG – An old laptop I have with a busted screen but one which I use hooked up to the TV with the HD 7970 as a pretty decent Kodi box/Steam console. This one has no UEFI support at all and runs a legacy BIOS.
  • The Dell/Wyse 5020 – This is an AMD Kabini SoC based Thin Client which I’ve been playing with lately. I’ve used a PCIe adapter with it without troubles with both a FirePro v4900 and also the HD 7970

Basically, with the w5000 plugged in, I have no BIOS option to select the discrete card which I believe means the BIOS can’t see the card. This is with a DVI->HDMI adapter plugged in to the card’s DVI port and an HDMI dummy monitor dongle plugged into that adapter, so it should think it’s hooked up to a display.

If I boot to the UEFI shell and list cards, it is not shown. lspci on linux doesn’t show it (even if I leave it completely unplugged while the system boots, then plug it in and echo 1 >/sys/bus/pci/rescan it).

This problem exists on all three machines, none can seem to detect it at all and behave like there’s nothing there.

I de-soldered the pm25LD010 SPI flash chip from the card and used my BusPirate device to dump its contents with flashrom. The dumped ROM is identical to this one: VGA Bios Collection: Dell FirePro W5000 2048 MB | TechPowerUp

I thought, OK, maybe Dell are some buttholes and maybe the vBIOS locks it to Dell hardware. So I used the BusPirate to flash this other ROM, VGA Bios Collection: AMD FirePro W5000 2048 MB | TechPowerUp , to the chip and resoldered it. No change. This second ROM doesn’t support UEFI at all according to TechPowerUp so I made sure everything was using Both or Legacy (and the Acer 4820TG doesn’t support UEFI at all).

I’ve also been using a Dell DA-2 220W power supply to power the 12V rail on the mini PCIe adapter since this card has now auxiliary power input connectors as it’s meant to get power from the PCIe slot.

I’m at a loss of where to go from here. I’m thinking the card just doesn’t work at all although it was sold as tested. The only thing that gives me pause is this guy’s story where the card wouldn’t work on his board but worked just fine on his work PC: Solved: HPE-510t does not boot with AMD FirePro W5000 (Windows 8.1) - HP Support Community - 4027072

I’m happy to perform more diagnostics, I have a good multimeter and also an Agilent 54622D 100MHz oscilloscope if there’s a way to probe the PCIe at relatively slow speeds. I also have an OpenBench logic sniffer in a box somewhere, but it’s 20MHz and I doubt it can do anything the oscilloscope can’t.

Thanks in advance for any ideas y’all have!

1 Like
2

After messing with this over the weekend and giving way too much time to an admittedly pretty crap card (it’s the mystery that’s interesting!), I’m pretty sure it’s DOA. I had the idea to hook it up to the scope and watch the CS (chip select) line on the SPI flash containing the ROM. If I do this on a known good FirePro V4900, I can see the CS line go high when the PC’s power is turned on and then after about a second, the CS goes low indicating that something is reading it on its SPI bus.

If I do the same on the w5000, the CS line goes high with the PC’s power but then stays high indefinitely. So it seems like nothing is even attempting to read the vBIOS.

Using the multimeter on the 3.3V line to ground on the V4900 shows ~3.2kOhm when it settles although this is probably specific to however my meter measures resistance as it looks capacitive. Capacitance 3.3V->Ground measures ~500uF on the V49000.

On the w5000 3.3V → GND measures 220ohm resistance and 11mF capacitance. So way more capacitance and way less resistance, although again this is in situ with the multimeter so who knows what’s really going on there. And also 10mF is pretty large but seems reasonable for combined power decoupling for the 3.3V rail. Reading 220ohm resistance on the rail seems suspicious but this is about [email protected] so not like something failed shorted or anything.

Furthermore, quick back of the hand check on the PCB on the w5000 seems like everything is ice cold. So dunno what the issue is but I think I should try to get my money back :smiley:

1 Like
3

Oh yeah, I also gave it the eyeball test yesterday and one of the PCIe lanes had an AC coupling cap that clearly had been whanged pretty good at some point.

w5000-whanged-cap-small
w5000-whanged-cap-small1518×2024 109 KB

The SPI Flash containing the vBIOS is in the background and this is after I desoldered, flashed, then resoldered it. I’m Pretty sure I didn’t do this by flailing around wildly with the iron as I was pretty careful and that looks mechanically dislodged instead of melted off. It is odd that nothing else I could see showed mechanical damage though? Also I masured this with the scope and it wasn’t shorted to the one next to it and seemed to be reading electrically fine. It’s also a lane that isn’t used at all with my mini PCIe adapter (which is PCIe 1x). I did fix this though and nothing changed.

Sorry for the meh picture, this is my phone looking through a 10x loupe.

1 Like
4

Just because I linked the original auction I figured I should post the resolution. Seller refunded the money no problems, don’t even have to send the card back. So A+ on their part in terms of followup!

3 Likes
5

are you going to try and repair it and try to get it working anyway?

6

Maybe, though now I’m thinking oven :smiley:

I would love to solve it from a forensic fascination standpoint but nothing seems really obvious. I honestly have no idea what the boot flow chart looks like but I think my personal best guess is that a GDDR5 chip went bad since none of them have any sort of cooling solution applied. Not sure if that would fault it before even being detected by the system though.

There’s only two electrolytic caps and they’re not that old/from the plague years and they visually look ok as well. I’m really confused how that tiny ceramic got knocked off without any other obvious marks on anything else though.

7

Maybe the cap was never on properly on from the start. If it is just a filtering cap that could probably get on okay with it.

8

It’s an AC coupling cap so its job is essentially to remove any sort of DC bias (e.g. if the card’s ground and the PC’s ground are slightly different). So it definitely has to be there* since it’s in series.

You are correct for a power supply decoupling cap, which has the job of removing AC components from the DC power rails. I have an older 'scope that I repaired where some tantalums failed shorted. I removed them and ordered new ones but ran it without problems for a week while waiting for the new parts.

Quick 'n dirty explanation is that capacitors pass changes along. In the case of the pictured capacitor that would be the signal less any slow noise (e.g. 60Hz from the mains).

For the power decoupling caps you’re talking about they do the same, but they’re oriented to bypass that change to the ground. In other words, variations get removed and your chip sees a dead solid 3.3V or whatever.

Essentially, inline (aka serial) they pass the higher component, if thety’re in parallel (to ground) they pass the lower component.

*For that lane to work

1 Like
9

Ah okay, thanks for that, as you can see I am clearly not an elechicken/chip-wizard. I only have the basest of understandings.

1 Like
10

My training is computer science and I have no idea why I know this :joy:

My job should be telling EEs that it works on paper and they need to fix it lol

1 Like
11

@byteporter I was doing a search and came across your post. Was the card bad - what was the outcome. I ask since I own 2 of these and one has a bad fan and runs 15+ degrees celsius hotter than the other. I cannot find a replacement fan for it and if your card is bad and you find it not worthwhile to fix it I would like to buy the fan - hopefully it will work. Thank you for the time.
Regards

12

@ProfileTrader my bad, just seeing this now. Yeah, AFAIK my card is bad. Seller refunded everything so I essentially didn’t pay for it, you can just have the fan for whatever it costs to ship.

To save shipping though, you might be able to just refresh the fan you have now. I had a crappy fan that barely spun on another card (Firepro V4900) and I was able to pull the fan blades off of the center shaft without any trouble. Cleaned it up and put a drop of sewing machine oil on the shaft and then just pushed the blades back on. It works great now!

So you’re welcome to the cooler from my dead card if you like, but if you want to save some shipping you can try the above to repair the fan you have now possibly :slight_smile:

13

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.