Hi guys!
I’m trying to update drivers and after choosing the option with right driver version I receive this:
Cannot write to ‘amdgpu-pro-19.50-967956-ubuntu-18.04.tar.xz.1’ (Read-only file system).
How I can solve this?
Moved from anther thread for relevance
I have a slightly different issue I’m dealing with, and I was hoping it could be cured in the next upgrade, to avoid me really, really screwing things up in my system (again) just to try to install a card that should be mining with no problems.
As you know, the market for new cards has gone insane overnight due to renewed interest in mining. I REFUSE to pay $800+ for an RX 5700 XT I paid $440 just one month prior, or $400+ for an RX 5500 XT EVER.
I’ve been researching older server cards, and have found some with specs suitable for mining. Preferable, actually. I have three AMD Firepro S9150’s that I recently purchased for around $150 each. They spec out very nicely for an older card. Simliar transistor count to a 5500, but with twice as many cores, and twice as much (16gb) GDDR5 RAM. The main problem is, It seems to want certain lines of code that aren’t installed in the standard AMD “openCL” driver, unless you either do a full “pro-install” which causes my setup to black out and reboot midway through due to attempting to setup a different “dkms” file. When it gets to that point, it goes “hard reboot.”
If I try to “uninstall” then reinstall, I says I have to run;
sudo dpkg --configure -a but goes into hard reboot while doing so. Then I have to wipe the hard drive and start over.
It wants to use a “legacy-open source” version.
While I’m starting to pick up on basic programming and setup with LINUX, this seems to be a couple of Notches above my pay-grade, so to speak.
this is what I supposedly need help with, except it would be 20.40 for Ubuntu 18.04 (same procedure, just for the server version of the driver) ;
https://math.dartmouth.edu/~sarunas/amdgpu.html
Update, 2021.01.20.
To install OpenCL part from AMDGPU-PRO, run:
./amdgpu-install --opencl=legacy --headless --no-dkms
Last tested with:
- Ubuntu 20.04.1 LTS, Linux 5.4.0-62-generic, AMDGPU-PRO 20.45-1188099-ubuntu-20.04, Radeon RX 480 (Ellesmere).
N.B.:
- As of AMDGPU-PRO 20.45-1188099-ubuntu-20.04 (and perhaps earlier),
amdgpu-installdoes not honor--no-dkmsoption when run with--opencl=rocr, i.e.amdgpu-dkmsis intalled and kernel moduleamdgpu.kois compiled, replacing the one from Linux kernel tree. - amdgpu-install may complain that it can’t install, because this is not Ubuntu version expected. Edit /etc/os-release and replace Ubuntu version with what installer wants. After amdgpu-install completes /etc/os-release may be changed back.
- In Ubuntu 20.04 /dev/kfd is owned by root.render, not by root.video. Add user intended to use OpenCL to render group.
‘amdgpu’ is an open source Linux kernel module for AMD graphics. It is available in stock kernels starting with Linux 4.2. OpenCL support can be added by installing libraries from binary AMDGPU-PRO driver download, provided by AMD. AMDGPU-PRO itself only works with certain distributions/releases/kernel versions.
Check lsmod and /var/log/Xorg.log that you have amdgpu loaded and working.
Download AMDGPU-PRO package for your Linux distribution from amd.com.
``**tar -xvf … ** to upack the downloaded file and run:
amdgpu-pro-install --compute ( see Update above )
which is for “compute only”, i.e. it will install only the following:
clinfo-amdgpu-pro
opencl-amdgpu-pro-icd
amdgpu-pro-dkms
libdrm2-amdgpu-pro
libdrm-amdgpu-pro-amdgpu1
Apparently, I can’t do this unless I either run;
$ sudo apt install build-essential dkms
or it tries to install it.
Any thoughts? A $5000 (new) server card with minor mods for effective cooling and all for less than $250, that should be able to hash in the 25-30 MH/s range with enough memory to say “DAG file? What DAG file? Pfft. DAG file.” is worth trying to get running, IMO.
Thank you for any help you can provide.
*
Whoever is trying to help hiveos recognize my cards, thank you I’m very grateful, but it isn’t quite there. The os recognizes my card (sort of) as a W9100, slightly later model with a fan instead of the S9150. But teamredminer is still having issues with the processor itself. At least it’s recognizing the processor it isn’t recognizing.
I’m betting it’s just a couple more lines of code.
Again, thanks for your help.
Interesting what you have going on here. I also was looking for alternatives to popular and now expensive GPU’s and discovered a FirePro w7100 8GB card.
I plugged it in to see what would happen. It ran at 42 MH/s out the gate but HiveOS was running it as a unknown card. The only settings I could change on the card therefore was the Fan speed. Even at 100% and running external fans to compensate the card would get up to 75C in like 5 minutes and I would have shutdown the miner.
The card really shows promise and I have 5 of these. Any help you might have on at least allow me to under volt the card would be really helpful.
Thank you,
zero
1 Like
I would just like to be able to mine in teamredminer. I think I may be able to use phoenixminer, but I’m also curious how my fan solution will perform. If it isn’t up to snuff, I already have upgrade ideas.
I meant to ask you what miner you were using when you had your FirePro running. teamredminer won’t recognize the chip in my card. I was thinking of switching over to Phoenixminer when I change my flight sheet, after my next payout.
Any input is positive input, so thank you in advance.
I was using Phoenixminer latest version I believe. The issue I can’t control any voltages so it runs at max speed and voltage and quickly over heats. I am looking for aftermarket cooling options for the cards as well.
Cooling is one big issue with server cards. They were designed to be in an enclosure that forced a lot of air through them, and most of them weren’t using the processor itself at full speed.
If you pull up “amd-info” on your rig, you’ll see that while most “commercial” cards have only one speed for the processor, that (at least with my S9150) they have several. If you run the processor in one of these slower states, it will probably run cooler.
Also, with the latest updates, it looks like you might have a tad more control with these cards.
I’ll let you know more when I switch flight sheets.
any updates with the latest version of HiveOS?
The current cooling solution I’m using is the one I had planned for my NVidia Tesla K80, that I had determined was way underpowered for a dual GPU card. It just happened to fit my FirePro S9150 very well with slight modifications. It even worked out to have a slot for the mini 4 pin adapter I used for my fan.
I did finally get my cards to run. I had to wait until another flight sheet had completed, so I could switch miners to phoenixminer.
When I switched, I looked up the basic arguments I would need to add for the cards to work properly, including cooling. By adding { -ttli } you tell the gpu that when it hits a specified temperature, to throttle back until it reaches the specified temperature. It reduces the hash rate momentarily, but it keeps it from rebooting.
I have my temperature set at 53*c. Right now the 5700’s get nowhere near that. The S9150’s can get close, even at 100% fan speed for right now. I’m also researching a solution, but this works for right now. The server cards have been solving shares that are over One Terahash occasionally.
The other arguments are turning on the drive that server cards use, so that will work more efficiently. I can’t BIOS mod them but, They can be worked with. And for a tad less than $200 (American) They were roughly what i paid for the RX 5500 XT they replaced back in Nov 2020, that are now selling for $500 (FFS! WTFBBQSauce?!?!)
But the string of arguments makes them run a bit smoother, until you can mechanically modify them for better cooling. There are several inexpensive modifications if you’re bold enough to pop the case.
1 Like
fantastic work! looking forward to seeing more results. they do draw a good bit of power for their hashrate, though. have you been able to check at the wall?
any updates to share here?
Updates.
Yes, they are power hungry.
Yes, they are a beast to keep cool. The current modification is only effective with ambient air temps under 20* C. I’ve found a company that sells a universal GPU heat sink built more like a traditional PC style heat sink. I’m waiting for one to come for number 3 that I tried to outthink the engineers. It can only function properly around 5* C and under.
Currently, a window fan, and a ton of ducting is the “until I get my parts” solution.
Those temperatures make the rest of the computer unstable. When I get the temperature workaround figured out I will update you.
As for control, they do what they want to. I can control fan speed, and I can control wattage with temperature arguments, but that also lowers hash rate. Most of this boils down to miners not using the same Kernel as Tech companies designing CAD-CAM software and say, mapping the universe.
People writing mining software weren’t expecting people to use cards that were the cost of a decent used car from a New car dealer, that wasn’t used for PCs in the first place.
I’m doing the hot rod equivalent of stuffing a B-6 Cummins into a '32 Ford rat rod, and using a Banks CPU to get it rolling down the road. No big deal. Just another day.
1 Like
I bought this fan from amazon and jury rigged it to mine. I am getting 26.5MH/s using the hawaii bios editor and this seems to be fairly stable. I get around 1 reboot a day but it doesn’t bug me as the card is running its its own ‘quarantine’ box.
https://www.amazon.com/gp/product/B08P1S5DBN/ref=ppx_yo_dt_b_asin_title_o03_s00?ie=UTF8&psc=1
@FreemanMaudib I noticed you have OC settings there. I wasn’t able to get any OCing to stick to mine using the hive UI, so the hawaii bios editor was the route I took.
I haven’t done any overclocking, I’ve just told it to use the temp sensor as a throttle pedal, and from boot either card only use around 135-137w when they’re running at peak.
My next step in cooling solutions is to basically convert it into a “naked” R9 “Deluxe.” with a real GPU fan and heatsink with individual heat sinks for the memory chips and MOSFET transistors.
Where did you get the BIOS for the card? I have a 3rd card that only maxes out at 23.5 MH/s, and If I were to edit the bios, I’d like to do all 3.
I started using S9150s due to their low price and high amount of VRAM.
I have 4 of them, with 6 more coming from eBay ($88 EACH WITH SHIPPING!! 
)
My cooling solution was just to use a holesaw to drill a hole in side of the plastic shroud, bolt a 90mm fan to the hole and connect the fan to an external power splitter. Then just use tape to block off the rear opening of the shroud. The whole process takes like 5-10 mins per card, max.
Temps are usually 60-75C for the four S9150s I’m using now as long as I keep airflow on the backplate to cool the rear of the PCB as well.
Power consumption is about 180W per card from the wall with a Kill-A-Watt meter.
Modded Dell Firepro S9150 16GB BIOS ROM with 1000Mhz core clock and 1000Mhz memory clock, and 1100mv core/mem voltages. Constant 28.6Mh/s in Phoenixminer in HiveOS. Works with Dell and HP Firepro S9150 16GB cards.
My custom BIOS is here:
https://gofile.io/d/8Tfjg7
Flashing the ROMs took a lot of tries… kept failing in my Windows desktop over and over again. I had to go into Safe Mode AND uninstall the card from the Device Manager before it would flash with any reliability.
I tried that with one of my cards, except I bought an adapter to use the onboard fan control/power so i could regulate it from the control panel, even though I leave them at 100% almost all of the time anyways. I have concerns about running the cards at those temperatures, as they just don’t seem to run that stable for me, and keep rebooting. They love operating in the 45-50*c range, and can go an entire week without rebooting, if there aren’t any other reasons for a restart.
I will give the custom BIOS a try on the card I’m customizing right now. I’ve removed the cover and the huge passive heat sink, and I’m basically turning it into a “R9 Deluxe” by using individual heat sinks for the memory and CMOS transistors, and an “Active” fan powered heat sink for the GPU.
It’s going to be the ‘rat rod’ of server cards. Hopefully, this will make them a viable alternative to the crazy prices and non-availability of new cards.
Thanks for the tips.
I considered watercooling the chip and putting those heatsinks on all the RAM and voltage regulators, but the cost per card would have been prohibitive.
It’s just way cheaper to run the original heatsink and add a fan. I also happen to have a ton of old fans around my house.
Temps in the 60’s are nothing for the Hawaii chips… they run HOT … if you look at people mining and gaming on the R9 290X and 390X, they usually run in the mid 80’s the entire time!
Tell me how the BIOS works… my cards have been rock solid.
1 Like
Hi there. Could it be a driver error? : Autofan: GPU temperature 511 is unreal, driver error
I got random 0 Mh/s with 100% in my rig. There are 6 cards in it. I have tried to lower the tuning, but as the miner starts, randomly 2 cards failing…
Is there any solution for this issue?
In 95% cases it’s power issue and rest 5% some cards return similar data when it totally hangs due OC
1 Like