PCIE slots not detecting riser cards on an ASRock mining board.

[marktwayne]

I built a mining rig about a week ago with 6 rx 580's, a 1200w power supply, and an ASRock H81 Pro BTC 2.0 motherboard. I tweaked the video cards and drivers, and the rx 508's started hashing Ethereum for almost 31 MH/s each. Success!

This afternoon, after a week of running constantly, the rig crashed around 4:00pm while I am still at work. I reboot the system. The monitor, connected to the rx 580 connected to the riser in the PCIE x16 slot, displays nothing. I switch around the HDMI cable and discover the Intel onboard video is displaying a login to Windows.

I log into Windows and discover it doesn't detect any of the rx 580's anymore. I shut the system down, and disconnect all the cards. I take one card sitting on a riser, connect up 8-pin power, stick the riser card in the PCIE x16 slot, making sure the molex/SATA connection and the usb cable is in securely. I start it up and the system doesn't detect the single card in the riser. I try this a few times with various combinations of cards and risers and still no luck.

I reset the CMOS and try all six cards, one at a time, and none of them are detected. I then think that all the riser cards couldn't be bad. Maybe the PCIE controller chip isn't working on the motherboard. I place a single video card directly into the PCIE x16 slot (rather than using a riser). The system detects the video card.

Any thoughts? What's wrong with this system? What are my next troubleshooting steps?

 

[TJ Hooker]

Look what is listed in under display adaptors in Device Manager. Be sure to check the option in device manager to show hidden items.

You could try reinstalling Windows + drivers. If possible, just create a new installation on a new partition so you don't have to blow away the previous installation before finding out whether that was the issue.

 

[marktwayne]

Yes. The rx 580's are there under "hidden devices." When I select properties on these hidden devices, I see "Currently, this hardware device is not connected to this computer" under "Device Status." But they are connected! All six cards are sitting in PCIE x16 risers. The power supply has 8-pin power to all cards and SATA power to the risers. The risers are also connected to riser cards in each PCIE slot. They are all connected no different now than when the system was working.

I reinstalled Windows. No devices that resemble rx 580's in their pre-driver state were present. AMD driver installation detected no rx 580's. I'm thinking the system is not detecting any of these cards at the lowest hardware level. If PCIE cards were detected, the integrated Intel video shouldn't display. Also, the video cards used to run their fans at boot. Now the video card's fans never start. It is as if these devices were disabled at the hardware, rather than OS, level.

 

[TJ Hooker]

The only other thing I can think to try would be to disconnect all the cards, remove all instances of RX 580 from device manager, and then reconnect one of the cards and see if anything is recognized. Although if you tried reinstalling windows already I don't know if this would help.

If you're reset CMOS already, I'm not sure what else you can do to get the cards to be recognized at a hardware level...

 

[4745454b]

http://www.tomshardware.com/reviews/amd-radeon-rx-580-review,5020-6.html

So a 580 needs around 230W. And you have/had SIX of them running on your system? 6*230 = 1380W. Plus the CPU and everything else, I'd expect this rig to need AT LEAST 1500W. I suspect you have somehow fried your PSU. Or perhaps blown something that outputs to the PCIe cables. You did say you came home and found it dead. Try the cards in a different system with a different PSU. If they check out you just need a properly sized PSU.

 

[marktwayne]

accidentally deleted post. 🙁

In this post I explained that I was really only running five GPUs rather than six at the time of the crash, and explained why I might have underestimated my system's power needs. I also thanked the previous poster for pointing out that this may be a power supply issue.

 

[4745454b]

I use a 200W figure when looking at the 580 because most of us only use our GPUs to game. If you look at the link I gave it shows the gaming load can be down as low as 209W depending on model. For mining you need to look at the torture power draw because that's what mining is. Card hitting 100% all the time. Even at 200W however you are looking at 1200W for just the cards. So a 1200W PSU just isn't enough. A 1500W with 5 cards should be ok. Or "fine tune" the cards to lower the power draw and run all six. My guess is however you stressed the PSU too much and something broke. I would try the cards in another system with a different PSU.

 

[TJ Hooker]

If you're mining eth only, power draw should be much less than 200-230 W. If you've undervolted/core underclocked properly, it should be low 100s per card, maybe less. If you're dual mining (or mining some other compute-oriented coin), power draw will be higher, but should still be less than max assuming you've optimized voltage/frequency properly.

 

[marktwayne]

I overclocked my memory generously and overclocked my core slightly until I got a 30.5 Ethereum hash rate with each card. Then I undervolted my cards slightly. I'm no expert at this, but I was happy with my results. When I was running on 6 GPUs, I was running into a few problems that could be explained by lack of power. However, with 5 GPUs, things were running smoothly. I had no crashes with 5 cards.

That means that 1200 watts was enough power to handle 5 video cards, but not 6. Sure, with 5 GPUs, it was probably running close to 1200 watts at max. If it went over 1200 watts while running 5 cards, I would have seen many system crashes before I ran my PSU into the ground. Power supplies are usually pretty sturdy. My gaming system has a 240-watt power supply and a GTX 1050 TI. I'm sure it is running close to maximum, but I've never had a problem.

I appreciate 4745454b's answer. It told me why I may have had trouble with my 6th card. He suggested that my power supply may have failed. If the power supply went bad after a week, I won't feel guilty about sending it back. Still, I've only seen power supplies work or completely fail. I've never seen one continue to operate, but at 10% power.

The 600-watt power supply is in my son's system. He has the only computer in the house with enough headroom to test my components. I'll have to ask him carefully.

 

[marktwayne]

So, what's the best way to find out whether my power supply is only producing 1/10th of the juice it should?

 

[TJ Hooker]

I think it's pretty unusual to have to overclock an RX 580 to achieve optimal eth mining results. For instance, with my own RX 580, 1200 MHz core clock was getting with ~0.1 MH/s of what I was getting with stock core clock (1340 MHz for my card I think). And with this core speed I was able to set a significant undervolt: 885 mV resulting in a measured Vcore of ~875 mV while mining. I think at stock settings Vcore was 1.1+ V.

Maybe your cards are different, but I'd definitely investigate underclocking/further undervolting if you haven't already.

 

[marktwayne]

I'm sure you know a lot more about tweaking cards for eth mining than I do. I think I was running each card at 1431 core/2180 memory and a very slight undervolt. I was getting 27-28 MH/s stock after BIOS flash with stock core/memory. I was getting about 30-31 MH/s with overclock. I have no idea if that is optimal. My electricity costs are about 9 cents per KW/hr. I'd love to tweak things further on my system, but I have to get it back up and running first.

I just booted the system with a 25-watt Nvidia gt 730 in a riser card. This would suggest that the power supply is producing enough energy to power a single gt 730, but not enough to power a single rx 580. Do power supplies really go from producing 1200 watts to less than 200 watts?

 

[4745454b]

I don't think the issue is that it's producing 1/10 the rated power, I think something might have fried in the secondary/daughter board/cables. Something that doesn't allow the GPUs to run, but allows the system to turn on and the IGP to work. You've never said what PSU you are using so I don't know if you even have a daughter board in it. (these are found on modular PSUs.) If the thing that broke was next to the PCIe cables, the PSU could work, but not be able to send power down the PCIe cables.

Edit: As for testing, one at a time try each card in sons machine. If they boot up and work, then it's a PSU or other issue. If they all failed then whatever happened killed the cards as well.

 

[marktwayne]

Thermaltake Toughpower 1200W Gold. The SATA cables provide power to the risers and I know those are working, but maybe the 8-pin power isn't. The PSU has 8-pin modular cables with a pair of 8-pin connectors each. I'm running three.

 

[4745454b]

You have to plug PCIe cables into the cards themselves. These are what I suspect are broken. Try the cards in the kids machine and see what happens.

 

[cpams322]

I had a similar problem as you are having, not with a mining rig, just a regular, run-of-the-mill PC. One PC was detecting a specific PCIe card while another was not.

It was the PSU. Checking the voltages on a PSU apparently is not a good test to see if it's good, even if you test under load. When I swapped the PSUs from each system, the problem transferred to the other PC and the problem PC was fixed. It was the PSU failing.

Put all your cards in and check to see if they are detected in you BIOS. If they are not detected in your BIOS, then they will not be available in Windows, not even as "other device" in device manager. Since Windows XP, Windows makes a device list based on what is found in the BIOS. If a certain device is not found in the BIOS, Windows thinks it doesn't exist. This is the problem I was having and it was the PSU.

The only way to find out if your PSU is the culprit is to replace it with a known good PSU. You can test and meter until your eyes turn black. But until you actually replace the PSU with a known good one (and keep all other variables the same), you will never know if the PSU is the culprit or not.

You can read about my dilemma at http://www.tomshardware.com/forum/id-3603615/dell-optiplex-760-strange-usb-pci-card-issue.html if interested. I would just read the first post and the last post.

 

[marktwayne]

I got myself a 1500 watt power supply. Everything booted up with no problem after I hooked it up. It looks like it was a subsystem of the power supply that was out. Thanks, everyone!

 

[4745454b]

How many cards are you running now? 5 or 6? Have you "tuned" the cards yet? From what I understand you should be able to bring the power usage down while not impacting the clock speeds too much. This will help your profitability a lot.

 

[marktwayne]

I'm running six on a 1500 watt PSU now. The optimization guides I have found so far suggest a substantial memory speed increase and a slight core speed increase. These rx 580's are now running around 1450 MHz core/2150 MHz memory/98% power limit each. Along with the BIOS flash, the brought my hash rates from stock 23 Mh/s to optimized 31 Mh/s.

I've heard of underclocking/undervolting as a strategy, and have tried to implement lower than 98% undervolt but I haven't been able to do it successfully and keep the cards stable at a 30+ Mh/s hash rate per card. I also haven't found any guides for underclocking/undervolting the rx 580. If you can point me in the direction of an article on the subject, or explain it yourself, I'd appreciate it.

 

[4745454b]

As long as it's working I'd say keep it. Your putting a heck of a load on that PSU. That's the only thing I worry about failing.

 

[marktwayne]

I've had two SATA power cables to the risers burn out. I switched to Molex now. That's what has me concerned now.

 

[TJ Hooker]

It's pretty unusual to require 1450 MHz core clock. For my RX 580/570s, there's less than 1 MH/s difference between running at 1150 MHz and 1300+ MHz. What utility did you use to try underclocking/undervolting? How much did you try reducing clock/voltage by?

And what miner software are you using?

 

[marktwayne]

Afterburner for clock and Claymore for mining. With the 1150 MHz core, what is your memory clock at? What mining speeds you get with that.

As far as underclocking, I tried going below 1300 MHz and my mining slowed to 27 MH/s.

Stock RX 580 - 23 MH/s
RX 580 with BIOS mod - 28 MH/s
RX 580 with BIOS mod and overclock - 31 MH/s

I have a gtx 1060 rig that is currently mining Ethereum as well. I slow down the core clock by 100 and speed up my memory by 600. It seems to mine pretty well.

 

[4745454b]

I see the hash rates and clock insinuations, but I don't see any power info. To be honest your middle figures of 28MH/s is probably "best". But without power info you are honestly guessing at the best settings to use.

 

[TJ Hooker]

I used to set my frequencies/voltages directly with Claymore using command line options. The only annoying thing is that the first time you run it you have to run it twice. The first time applies the settings, but you need to restart the miner to have those settings have an effect on your mining. Since then, I've just wrote all my settings directly into the vBIOS.

On my RX 580 Nitro+ 8GB, I'm currently running 1110 MHz core, 2190 MHz memory, with 850 mV core/memory voltage. I think I'm getting about 30.6 MH/s (using Phoenixminer). For comparison, IIRC the best hashrate I ever managed was 31.x MH/s, and I think that was with 1250 MHz core, 2220 MHz memory (and voltage around 900 mV I think), but that used more power and also resulted in some memory errors (as measured in HWiNFO64).

Hashrate also depends on what memory you have though. I have Hynix and Samsung (with Samsung apparently being the best), and used the Polaris Bios Editor one click memory strap option to change my timings.

Oh, and I'm running the AMD Beta Blockchain drivers.

If your electricity is very cheap and the heat/noise aren't an issue, then you don't necessarily have to worry about it though.