Team Red Miner R-Mode 10% Efficiency Improvement

Very interesting. I have 16 5700XTs, so this would be a significant power savings. However, I would have to convert the rigs over to Ubuntu again (currently using Awesome Miner and Windows).

Also, all the GPUs are running modded BIOS for memory straps. I did this back when I got them all in December 2020. I think nowadays you don't really need to do the straps in the BIOS as it can be done on the fly.

ClockerXP Would you mind providing some more details about your rig configuration and if you're running stock BIOS on your cards?

This is what my most efficient rig is currently doing: (running PhoenixMiner)

miner002_5_14_22.jpg
 
Last edited:
Thanks for the info. I haven't really been spending much time messing with my 5700XT rigs for probably over a year now. So these are my 5700XT rigs:

AMD GPUs.jpg


miner004 is running on Ubuntu 20.04 and using TRM. The other 4 are Windows rigs using Phoenix. Interesting point about the Phoenix hash rates being inflated. As you can see above, miner004 (TRM) is reporting 110.70 MH/s and miner008 (Phoenix) is reporting 111.56 MH/s.

Looking at the stats poolside, I see:

miner004.jpg


miner008.jpg


So yeah, it looks like I should definitely switch to TRM on the other rigs if nothing else, even if I stay on Windows.

Here is what I currently see on the TRM rig:

trm_before_R_mode.jpg


Even though I do have a grub file, adding the "--kernel_vm_mode=RR" argument to the miner on startup, just causes it to error out. So I would need to do some digging to figure out why.

I'm pretty much married to Awesome Miner at this point, and have all my clocking profiles defined there:

clocking profiles.jpg


Yes, I know the clock profiles are the same for almost all the GPUs currently, but I used to have them tuned individually, but after fighting stability issues, I went back to known stable clocks/voltages.

I used HiveOS back in 2018 but switched to Awesome Miner. For a while I was running the AM client on Ubuntu rigs, but dealing with driver updates was a huge hassle, as was troubleshooting issues and initially AM had limited GPU clock/fan control, which caused me to switch my rigs to Windows as that seems to be better supported by AM.

Been meaning to consolidate my 5 AMD rigs into just 2 rigs with 8 GPUs each, but haven't had the time to mess with it and they seem semi stable right now, so I don't know if I want to mess with them. But potentially shaving off 250 watts of power is pretty tempting.
 
Yeah I know, but if I reduce core voltage, stability is out the window. I'll need to set aside some time to trouble shoot one of these days.

I tried switching from Phoenix to TRM miner on my other rigs, but they don't start mining at all and just sit there with no errors on the console. I'll need to set aside some time to play with them, or just throw HiveOS on there, or perhaps try out mmpOS. I spent many a night doing base Ubuntu 20.04 installs and while I had decent lock on my Nvidia rigs, the AMD ones were a bear so I went back to Windows on those.
 
Hi, you have a nice tutorial how to do this on YT, and also Vega cards can be made more efficient.

THIS IS HUGE!!! TeamRedMiner v0.9.4.7 R Mode is Awesome! Is 5700xt now the most efficient card?
 
Yeah,
Hi, you have a nice tutorial how to do this on YT, and also Vega cards can be made more efficient.

THIS IS HUGE!!! TeamRedMiner v0.9.4.7 R Mode is Awesome! Is 5700xt now the most efficient card?


So far this is what I've seen:

Vega 56: 30-31W decrease, same hashrate
(core voltage 800mv now instead of 900mv, 925mhz core clock now instead of 1075mhz).

6700XT:
GPU1: 1-2W decrease, 200Mhz core clock decrease (1200mhz to 1000mhz) (102ish Watts to 100Watts in software)
GPU2: 2W decrease (89W to 87W)

5700XT:
GPU1: 10W decrease (95W to 85W)
GPU2: 13W decrease (107W to 94W)
0.1MH/s loss
Decreased core clock from 1400MHz to 1075MHz, decreased core voltage from 785mV to 705mV.


Only stability issues so far I've seen is that I got a "gpu detected dead" error message on one of my 5700XTs so I've raised the core voltage to 725mV to see if that'll fix it.


I don't like tweaking my GPUs too much so I'm sure you could get more out of this, but throwing rough values in there based on TRM's recommended values seems to work pretty well for the most part. I'm going to lower my core clocks on my 6800s slightly and see how much power I save but it doesn't look like much is to be gained there. Generally my experience matches up with TRM's notes, major efficiency gains on older AMD GPUs but not much on newer models. I expected better results on the 6700XTs since one of mine is a dud, but not seeing that so far.
 
That gain on Vega is because 900mV was too much to begin with. I was running my vegas stable with 800-812mV. For 5700 xt, you can lower you core clock to 990-1030 Mhz depending on the memeory clock. This will allow you to go below 700mV on the core voltage.
 
That gain on Vega is because 900mV was too much to begin with. I was running my vegas stable with 800-812mV. For 5700 xt, you can lower you core clock to 990-1030 Mhz depending on the memeory clock. This will allow you to go below 700mV on the core voltage.
hmmm, yeah weird. I have my other vega based cards running at between 800-820mV. Some are unstable below 820mV. Not sure why I had this one running higher. Now that I've had a few rigs running r-mode for awhile my 5700XTs seem stable now with 725mV core voltage. The vega56 is crashing every couple of hours but recovering so I'll try tweaking it a little more.

I'll try messing with the 5700XTs now that I have a stable baseline. Memory clocks are currently 900MHz.
 
hmmm, yeah weird. I have my other vega based cards running at between 800-820mV. Some are unstable below 820mV. Not sure why I had this one running higher. Now that I've had a few rigs running r-mode for awhile my 5700XTs seem stable now with 725mV core voltage. The vega56 is crashing every couple of hours but recovering so I'll try tweaking it a little more.

I'll try messing with the 5700XTs now that I have a stable baseline. Memory clocks are currently 900MHz.
For now, Vegas seem hard to be stable if mem clock is above 960. I hope this will be better in the final release of the miner.
 
Very interesting. I have 16 5700XTs, so this would be a significant power savings. However, I would have to convert the rigs over to Ubuntu again (currently using Awesome Miner and Windows).

Also, all the GPUs are running modded BIOS for memory straps. I did this back when I got them all in December 2020. I think nowadays you don't really need to do the straps in the BIOS as it can be done on the fly.

ClockerXP Would you mind providing some more details about your rig configuration and if you're running stock BIOS on your cards?
Hi pc- As stated by others, I believe the BIOS mod is still required and I have done it. Note that one of my cards is a 5700 flashed to 5700XT. I have tweaked up voltages slightly on a few of the cards to get better stability. I'm using Minerstat as I have been from basically the start, I highly prefer the UI over Hive or Rave. Plus Hive has been having a lot of issues lately.

For details? My rig is an old Asrock H81 BTC Pro R2.0. MSOS works great on it and it is easy to configure. First rig is free. 1.)Install the latest 1.8.0. 2.) Type amdmode at the commandline to set the variables/settings that TRM needs. 3.) Set clocks in Clocktune 4.) Set Worker Config 5.) Start mining
 
Last edited:
Thanks ClockerXP I was able to download and install the beta version of TRM and issue the command to set R mode from the terminal. I'm running a base install of Ubuntu 20.04. I use Awesome Miner to set the GPU and Memory clock. For some reason AM is not sending the Core power I set to the GPUs, but the results looks pretty good anyway:

core_volt_4.jpg


I'm still running AMD 20.30 drivers which are quite dated compared to the latest, which appears to be 22.10 for Ubuntu. I got a note in to Patrik (the AM developer) about not being able to set the core clock.
 
Used powerupp to lover the core voltage on a single GPU to compare:

powerupp_1.jpg


Made absolutely no difference in power or hashrate:

powerupp_2.jpg


Btw, the power numbers shown by AM are after AM adds 35% to account for the true power each GPU consumes since AMD cards only report Core power usage, not the memory and other components like Nvidia cards.
 
I have a sonoff S31 smart plugs on each mining rig that monitors power in real-time and feeds it to my Home Assistant server:

power_usage.jpg


I haven't gotten around to setting up history for power yet on the individual rigs. I'll do some more testing to see if I can see the power drop at the rig.
 
Most of my 5700 XTs won't take a core clock lower than 750 mV. I went back and looked at my notes from when I flashed them all originally back in fall of 2020 and found this:

AMD_flashing.jpg


So I guess that unless I want to go through the hassle of re-flashing those cards, I'm SOL as far as lowering the core voltages further. I did it in a Windows box originally and it was a pretty lengthy process doing one card at a time. I understand that HiveOS has a built in function to flash AMD cards. Maybe that would be the quickest path if I wanted to go through the trouble?
 
Just saw this nice, dropped my handful of 5X00 cards a nice drop in wattage. I almost put them up for sale to grab some RTX 3000 cards but not after this. I have a binned 5700XT Liquid Devil that's a total beast lol

5700.png
 
I moved a couple of 5700 XT cards to a Hive rig and it ran great for 24 hrs:

miner010_02.jpg


I then moved it to another location and fired it back up, and I lost 12 MH/s per GPU:

miner010_03.jpg


No amount of rebooting beings it back to the correct hash rates, nor does reseating the GPUs (they are both plugged directly into PCIe slots on the mobo.

Any idea what happened?
 
Have you tried increasing core clocks? The lowest I've set mine is to 1075 but there's one 5700XT struggling to go below 1275.

Also, while it works pretty well on my last vega56, I haven't really had great results on my AMD MI25s. If I don't mess with my stable overclocks it runs about as well as without r-mode, but with no change to the power consumption once I apply memory clocks. Without memory clock changes r-mode seems a little more efficient. I have one group of cards that looks like it saves about 5-6W per GPU with r-mode but the other two rigs are unstable so far.
 
Last edited:
I tried bumping core to 1100 and vcore to 700, and that got me to 48 MH/s. Still way less than before. These cards ran 56.5 MH/s since December 2000 until a few days ago on Phoenix Miner. And on TRM for the last several days until I moved the rig today. Guess I'll tear it down and put it back together to see if that helps. I could understand if one GPU flacked out, but both of them, and the exact same reduction in hashrate? Really weird.
 
Last edited:
R-Mode also works for the 6700XT and 6800 non XT. Each 6700XT was using around +20 watts more compared to when I was using NBminer with around the same hash rates, but the 6800 was only about 12 watts less. I haven't found anyway to improve the efficiency for 6750XT.
RX6000_20220520.jpg
 
There's issues with trm/hive conflicting and not applying grub edits properly (that are required to run R mode). Sounds like that's what's going on and they're running in B mode. One of the (many) reasons I really prefer mmp over hive.

Yeah, I ended up doing a fresh HiveOS image, but hardcoded a bogus rig ID and then installed the Awesome Miner Remote Agent on it and control it from AM. I then manually switched to R mode from the local console. Works like a champ now:

trm_01.jpg


I'm still having some issues adjust core voltage for better efficiency and not everything is reporting correctly back to the AM console:

trm_02.jpg


But I'm working with Patrik (author of AM) on those issues. Hope to have them licked soon.

HiveOS is nice and all, but being cloud based is a non-starter for me. Awesome Miner is all local (unless you want/need cloud access as well), so no worries about their APIs being overloaded, Internet issues, etc.
 
Last edited:
Yeah, I ended up doing a fresh HiveOS image, but hardcoded a bogus rig ID and then installed the Awesome Miner Remote Agent on it and control it from AM. I then manually switched to R mode from the local console. Works like a champ now:

just in the hive config file? was there any tricks to using a custom pool? ive been meanin to look and see how hard it would be to customize hiveos to NOT use any of the hiveos cloud crap.
 
Hi all, i ended up using trm v 0.9.4.2. any newer version make my three rx5600xts reboot. Error comes up saying insufficient buffer size. I have mixed rig of 2x 5700xt, 3x 5600xt, 4x 5500xt, cpu 3600 mining monero.
5700xts and 5600xts are bios moded

Could someone help me with this issue please?
 
Ok, so I've done some more tweaking with my AMD MI25 GPUs (similar to a vega56/64). I was also moving everything around at the same time so I'm not sure which it was, but after using r-mode I ran into a lot of stability issues even on previously stable configs. I replaced all my thumb drives with SSDs just to see if that would help, reinstalled all updates, and now I think it's working out ok.

Previous settings (42MH/s @ 120W ish)
1010-1050MHz core
820-830mV core voltage
830-850MHz memory

New settings (46MH/s @ 120W ish) with r-mode enabled
900MHz core
800-825mV core voltage
945MHz memory (unset, stock)

The 46MH/s I'm getting now isn't 100% stable but I'm not really getting restarts anymore, it's just that the hashrate of the cards fluctuates sometimes. Memory temps get close to 95C so I think that's the issue, because when I run the same exact settings without applying timings (which raise the memory temps) then I have pretty solid stability and hashrate (at a lower hashrate though). I'm going to experiment with lowering the core clock even further but so far I haven't had much luck. Lowering.

[Edit: The reason why I don't run higher memory temps is that even in a server case running max fan speed with low ambient temperatures it's difficult to keep the memory cooled enough. That's why previously I had to lower the memory clocks below stock to keep them stable. I think with a modded BIOS and better cooling on the memory that these could do much better.
 
Last edited:
Hi all, i ended up using trm v 0.9.4.2. any newer version make my three rx5600xts reboot. Error comes up saying insufficient buffer size. I have mixed rig of 2x 5700xt, 3x 5600xt, 4x 5500xt, cpu 3600 mining monero.
5700xts and 5600xts are bios moded

Could someone help me with this issue please?
Are you running windows?
 
just in the hive config file? was there any tricks to using a custom pool? ive been meanin to look and see how hard it would be to customize hiveos to NOT use any of the hiveos cloud crap.

In the hive config file, I just set the rig id to like 99999999 and then ssh into the rig (username "user" and password "1") once it comes up and then do the following:

#Host Rename
sudo vi /etc/hostname
sudo vi /etc/hosts
hostnamectl set-hostname miner001.home

#Allow root to login remotely
# Become root
sudo su
# Set up root password (Enter and confirm new password)
passwd root
# Modify ssh config
echo "PermitRootLogin yes" >> /etc/ssh/sshd_config
# Restart ssh daemon
sudo systemctl restart sshd

I then ssh back in as root and install Awesome Miner Agent:

wget http://www.awesomeminer.com/download/setup/awesomeminer-remoteagent.tar.xz
tar xvJf awesomeminer-remoteagent.tar.xz
cd awesomeminer-remoteagent
sudo ./service-install.sh

I then configure the host in Awesome Miner and set the rig to use TRM and start mining for like 10 seconds just to let Awesome Miner install TRM.

#Team Red Miner Mode R
# Config Kernel
cd /root/.config/AwesomeMinerService/teamredminer-v0.1.10.0-linux/
chmod +x ubuntu_mode_script.sh
./ubuntu_mode_script.sh R
reboot

You can of course also just install TRM manually and start it manually without install Awesome Miner first. This is what Awesome Miner sends to the rig to start mining:

Code:
/root/.config/AwesomeMinerService/teamredminer-v0.10.0-linux_1/teamredminer-v0.10.0-linux/teamredminer  -a ethash -o stratum+tcp://us1.ethermine.org:4444 -u 0x304AcBc3bfa713AC08bD2F6d5a59AaE2df047F72.miner001 -p x --watchdog_disabled --api_listen=0.0.0.0:4028
 
Last edited:
Back
Top