GPU 7 dropps to a very low hashrate while the rest mine at a normal MHS. Can anyone help fix my problem?
@catfish Drop mem clock on GPU7 to 1140 or even 1130. Clearly 1150 is too much for it. Once you clock memory too fast, hashrate drops sharply.
Firstly, Your comments on this thread on overclocking are much appreciated…have been extremely helpful getting my 6600xt’s tuned in. Thx
Second, whats your procedure for HiveOS when resetting/recovering from a “bad” overclock/underclock after a gpu crash?
@m2internet If we’re talking specifically 6600 XT and ETH, my procedure (if I start the OC from scratch) is always the same:
- First and foremost - establish max VRAM clock. This will define your hashrate. Keep VRAM voltage set to its default 1350 mv (or simply don’t put any value in there). Start with a conservative clock value. For me a conservative value would be 1050 for Micron and 1100 for Samsung RAM, but no harm starting from 1000. Then keep pushing it up in small (I use 5 MHz steps) increments, testing each step for at least a few minutes. Always write down the exact hashrate after those 5 min, at each step. Ideally, also reboot the miner 5-10 times with each setting, to make sure it is stable. It should increase with each step at the same exact rate. The moment it doesn’t increase by the exact same amount any more, that’s when I stop overclocking VRAM. Then I would restart the rig 15-20 times in a row, waiting 5 minutes after each reboot, making sure the GPU always spins up to the exact same hashrate after the miner starts. Any deviation from the exact expected hashrate, I would lower the mem clock by 1 MHz and then do another 15-20 reboots in a row. And again. And again, if needed. Until fully stable. Yes, it takes a while to do all this, but you end up with a max value, which will be a constant for this particular GPU and will guarantee that VRAM clock isn’t the source of any GPU crashes going forward.
- Find the minimum core clock, which supports this newly found max hashrate. In case of 6600XT, this is easy, it’s a constant - set the core clock to exactly 901.
- Find the minimum core voltage, that is stable at the selected core clock. This will be in the 605-660 range, depending on the card. I usually start with 700 mv and test in 5mv decrements (5 min each step). The moment GPU crashes, I back off 5 mv. Again, do 15-20 restarts in a row and raise the voltage if any crash occurs. Notice at this point the only source of a crash can be the core voltage. Once you found a stable setting (after minimum 24 hours test), raise it another 5-10mv, for good measure. Or if you’re like me, you’ll keep this at the EXACT minimum value where the card still remains stable, down to a millivolt. But that’s my OCD hehe.
- Find the miminum VRAM controller voltage. Start with 750mv, same drill, decrease in small increments, testing each step for a few minutes. Once the GPU crashes, back off 5 mv. Memory controller doesn’t experience power spikes like the GPU does, so this setting will be much more stable, even if it’s right at the edge.
- The other settings, like SoC frequency and voltage. SoC frequency, in case of 6600 XT, is a constant - 418 is the minimum stable value you can set in HiveOS. The minimum SoC voltage will be in the 740-800 range (usually). Start with 850 mv and - the usual - test in small decrements, 5 minutes each step. The moment it crashes, back off 5-10mv. Test, adjust and retest if needed. Once you have your minimum value, I would just leave the card running, for probably 1 week at least.
At this point you end up with a 6600 XT which will be pulling around 67W and do about 32MH. Any GPU crash will almost certainly be due to core voltage being set too low to accommodate occasional power spikes. This is where the R-mode in TRM comes handy, it really minimizes these spikes.
1 Like
Wow, again … I can’t thank you enough for your time and detail in these posts. Truly helpful.
See my biggest problem I have is getting that old hashrate consistency back once I go back to a “working” overclock set/version (I write them all down, and version them in “Overclocking templates”). See you reboot your rig 15-20 times, guess my procedure isn’t so nuts.
Once a GPU crashes from a “bad” overclock my concern is getting the GPU’s back to a stable setting for their speed & voltages. On crash I would:
- Put the rig into “Maintenance mode”
- Then “Shutdown and Boot in 30s”.
- After booting into “Maintenance mode”, using the rigs “Overclocking” section, I would then run a “Clean” to reset the GPU’s to their default state.
- Reboot again using “Shutdown and Boot in 30s”
- Use the rigs “Overclocking” section to load the saved/stable overclocking template for the GPU’s
- Turn OFF “Maintenance mode” for the rig
- Reboot again using “Shutdown and Boot in 30s”
- Make sure the miner is running stable as it should.
I figured this was overkill but I definitely noticed it took the GPU’s sometimes even pulling the power cords to get them to reset properly. I thought I was doing too much, but youre restarting 15-20 times in a row to do reset the GPU’s off a “bad” overclock as well.
Wish there was a more sidelined procedure tbh. So helpful to see how others are doing it. Thx again 
@m2internet Interesting how different our approaches are. I never had to do any of the things you did. I’ve never used Maintenance mode, etc. My OC template for a 6600 XT literally just has 901 core clock and 418 SoC frequancy, that’s it. I set all other settings manually for each card. Whenever I had a mystery crash (as in, after days/weeks of stable work, GPU suddenly crashes out of nowhere), it was always a case of changing one of two values in the OC section. It was either:
- mem clock too high (those 15-20 reboots in a row really help catch that), or
- core voltage too low (locking ETH profiles in the miner will help here + TRM R-mode).
1 Like
I updated my AMD drivers and manually changed my tuning configs to “R”, but r-mode is not working.
What am I missing?
You’re missing the equal sign “=” after the word “eth_config”.
Do you make changes to your OC’s with the miner running? I have been trying both, but am accustomed to shutting down the miner first before my OC’s since using LOLminer. I do think you can perform the OC’s with the miner running, then it run stable for over 24hrs, but then on restart/reboot it would crash. So I’ve been performing the OC’s after i shutdown the miner…its more procedure, but def more accdurate/efficient.
You don’t actually need the"="
But I tried that too, no change.
It works fine with my “A” settings, but when I change "A"s to "R"s it auto-tunes. No errors, it just ignores all tuning configs.
Try setting “–kernel_vm_mode=RR” in the miner config, instead of “–eth_config” and see what happens
I got that working. So it’s now working in R mode with auto tuning. I think the rig had to reboot or something. Thank you!
Actually what was missing was almost certainly the entry “amdgpu.vm_block_size=11 amdgpu.vm_size=2048” in grub.custom file. This is described in the R-Mode tuning guide, line 70. What you needed to do, is issue this command in the shell:
echo -e “amdgpu.vm_block_size=11\namdgpu.vm_size=2048” > /hive/etc/grub.custom && selfupgrade --force -g && sreboot
But when you run TRM with “–kernel_vm_mode=RR” once, it does that change for you.
Either way, you can now use your previous “–eth_config=R480…” line.
1 Like
Yes, you are correct. As always, thank you for the help.
1 Like
@m2internet I always restart the rig after changing any setting. GPU might be stable during the usual ethash calculations, but you would want to know if the new setting withstands a cold start - power spikes during mining app start, DAG file generation, etc. At least that’s my thinking.
What will everyone be mining with their 6600XTs after ETH merge? I’m probably going to try ERGO.
What’s good fam.
Anyone have an overclock config to share for FLUX mining on 6600 / 6600 XT please ?
1 Like
I have been mining ERGO. I’m trying to find the most profitable coin for 6600 XT. WhatToMine is now showing the following in order of most profitable:
Kaspa(KAS)
Ryo(RYO)
Conceal(CCX)
Nicehash-ZelHash
Neoxa(NEOX)
Flux(FLUX)
Vertcoin(VTC)
Nicehash-KawPow
Cortex(CTXC)
Ravencoin(RVN)
Nicehash-Autolykos
Firo(FIRO)
QuarkChain(QKC)
Equilibria(XEQ)
Ergo(ERG)
Is this correct? We should now mine Kaspa (what ever the heck that is)?
What are you now mining and why?
depends if you are mining for cash or coins. If cash maybe, if you want to hodl the coins mine whatever project you believe will make it.
If you like ergo and believe in it, then it depends on your electricity cost. Otherwise it probably won’t make sense to mine at a loss, buying the coin (with what you would pay for energy) might be better
