RTX 3060 Model 2504 (aka LHR v2) HiveOS Hashrate Limit's

Slxxpy · June 12, 2021, 2:58pm

Hey Guys I wanted to create this thread here so we could draw some more attention to this second model of 3060 that has the new model number (device id I presume as well). A lot of people on windows recommend using NVdriver 470 but that only works on the first-gen 3060, This newer version is not recognized by this driver on windows. Others will recommend Nvidia Driver 460.39 but due to my understanding and other member’s Due Diligence, This Modle 2504 is not supported in the driver that unlocks the Hashrate.

Im currently Mining Ergo @ 115 PL +0 on Mem and Core because the gpu will drop from 80 Mh/s to the low 50-60, Im curious if we will ever find a way to bypass this limit on the updated model Number.

It’s strange to me how a gpu limited on The Ethash is also limited on Autolykos as well.

Would love to hear the community feedback on this topic, Thank you!

eledush · June 12, 2021, 5:16pm

Try this

C 200
M 2300
PL 110

Slxxpy · June 12, 2021, 5:27pm

Putting anything over 200 on the mem drops it to 50-60 Mh/s, The card Thinks AutoLykos is Ethash.

eledush · June 12, 2021, 5:34pm

you are using nbminer?

Slxxpy · June 12, 2021, 5:48pm

Switching to it right now will update results

Edit: NB with +1000 on Mem bouncing between 76-86 peaks at 91 then drops back down, maybe it will stabilize over time

Edit 2: Still bouncing between the 70s to the high 90s

Edit 3: PL 115 Core 200 Mem 2300 causes me to crash, looks like there isn’t any fix yet for 3060 model 2504

eledush · June 12, 2021, 6:08pm

Slxxpy · June 12, 2021, 6:20pm

are you on driver 460.39? also have you owned the cards for more than 2 weeks?

eledush · June 12, 2021, 9:22pm

Yes, for 2 or 3 weeks I have been mining ergo. it is much more profitable than eth at least with 3060

dealbreak3r · June 13, 2021, 12:20am

Good place to gather all the details. Here’s what I got so far:

I had opened up GitHub issues for both NBMiner (Ergo mining, can’t downgrade driver for optimal hashrate) and lolminer 1.29 ( eth mining, 75% improvement hack):

Mining Eth it’s not viable with the LHR limiter. Lolminer 1.29 doesn’t give the 75% hashrate for it yet.

For right now all my 3000-series cards are on Windows rigs, NBMiner, mining ERGO. I can tune afterburner better than HiveOS.

I’ve been able to get 118 MH/s on the 3060 V2 (while everyone else gets about 125 MH/s on the 3060 v1). It’s close enough for me not to worry until this gets figured it.

hotness · June 14, 2021, 11:53am

Autolykos is very similar to Ethash, which is why you can essentially use the same overclocks profile, but this is also why the limiter is activated. The limiter doesn’t target algos by name, but certain behaviours linked to Ethash/Dagger Hashimoto.

aminer · June 18, 2021, 6:49am

@hotness
How certain are you of this statement? Because from what I know, the two algorithms are different enough so as not to be confused by any “hashrate limiting mechanism”. The memory allocation is different and also the core usage is different. I can elaborate more on this when I have more time.
The biggest example is Navi and Big Navi (AMD) cards who simply fail to produce good results in ERGO, whereas they excel in ETH mining. If the two algorithms were so similar as you say, these cards would have excellent results in ERGO as well.

hotness · June 18, 2021, 7:22am

Somewhat certain. On the latest Nvidia driver 3060 will drop down to 60 Mhs, so you need to use specifically 460.39 to get 120+ Mhs. I believe 460.39 is also the version required by Lolminer to run ETH at 75% hashrate.

I would say Navi cards are standard on ERG: 5700XT gets slightly less than double hashrate, 6800 gets about double hashrate. I would say Turing cards get about double hashrate also, so that seems to be the norm. So while Navi is not as good Ampere, it’s not necessarily bad.

Navi performs poorly due to an architectural difference. Here is a snippet from TeamRedMiner’s tuning guide:

This algo accesses mem in 32 byte chunks. This means that RDNA
generation gpus (Navi, Big Navi) will not perform well. Their 128 byte
cacheline size means that 128 bytes are read for every 32 byte
request, effectively halving the available memory bandwidth compared
to GCN (which uses 64 byte cachelines).

github.com

todxx/teamredminer/blob/master/doc/AUTOLYKOS_TUNING.txt

Team Red Miner Autolykos2 (ERGO) Mining
=======================================
This document provides some quick pointers on how to tune the
autolykos2 algo used by ERGO.


General background
==================
Autolykos2 is a memory-intensive low/medium power algo. However, with
the small memory accesses involved the algo behaves more like algos
like verthash rather than ethash. Performance is tied to the core clk,
and for max speed (especially for Vegas) core clk needs to be higher
than ethash to support driving the mem controller on the gpu(s).

This algo accesses mem in 32 byte chunks. This means that RDNA
generation gpus (Navi, Big Navi) will not perform well. Their 128 byte
cacheline size means that 128 bytes are read for every 32 byte
request, effectively halving the available memory bandwidth compared
to GCN (which uses 64 byte cachelines).

This file has been truncated. show original

aminer · June 18, 2021, 8:27am

A lot of confusion my friend, you are mixing a lot of things.

Lolminer and it’s “3060 booster” requires any driver from the range of 455.45.01 up to 460.39. Of course, naturally, we all use the latest working driver which is 460.39. Although, personally, I have used 455.45.01 on one of my rigs with a freshly installed HiveOS image (which comes with this driver version installed).

Even with 460.39, when mining ERGO, there are certain OC settings that makes your hashrate drop to 70-80 Mh/s. At the same time the wattage drops to the 80-90W territory. Does that mean that the limiter kicked in? No. It’s just that the card switches “performance mode” (similar to what AMD calls “Power state”). When you tweak the core again it goes back to full hashrate (120+). Similarly, if you mine ETH -even with 460.39- the limiter kicks in and afterwards, no matter what you do, ALL algos will be limited.

The above is proof that the limiter is only algo-specific and it will only kick in when mining ETH.

As for why the two algos are different. The fact that you can switch to ERGO with the same OC settings as ETH means absolutely nothing. As a matter of fact, in order to achieve the best ERGO hashrate and/or efficiency, you need to change the settings a lot. ERGO is a lot more core dependent than ETH, so you need to bump your Core Clock up instead of driving it all the way down as you do in ETH. Also, the fact that the algo accesses memory in 32 byte chunks, allows for a lot higher memory clocks than ETH. And that alone (the memory access allocation) is enough of a difference between the two algos so as not to be “confused” by nVidia’s limiter.

Very interesting discussion whatsoever to say the least!

hotness · June 18, 2021, 9:49am

It sounds like we are both confused.

I never had issues with driver 460.39 and low hashrate, it worked well as soon as I started using it. With the newer driver I was getting 60 Mhs, which is different from your 70-80 Mhs, so it must be a different issue. Also the behaviour was similar to the ETH limiter: it would start at 122 then drop to 60 Mhs.

The OC settings are much closer between ERG and ETH, as compared to say RVN or CTXC, where core clock and wattage go way up. For ERG I run my clocks a bit higher than stock, but somehow people still get good ERG hashrate with -502 core. I don’t know how to explain that. What I have found is that my ETH overclocks mining ERG were only about 3% less than hashrate than ERG optimised, so I would definitely classify that as “close enough”. Whereas if you used ETH overclocks to mine CTXC it would be something like 15-20% difference.

aminer · June 18, 2021, 11:07am

Just to get things on track again. You say that Autolykos and Ethash are very similar algorithms and that’s the reason why nVidia’s LHR kicks in for ERGO as well. You are basing this to the fact that both algos use somewhat similar OC settings.

To me, this is a wrong perspective. If you go on a statement about two algos and whether they are similar or different, you should be able to point facts about how each algorithm works.

I use entirely different settings for each algo. For example, with my 3060 v1, I am at -502 core 2600 mem with ETH, and -400 core 2800 mem with ERGO. To me, this is a big difference. Not to mention my Vega56 where I need to bump core from 1010 all the way to 1490 (this beast does 205MH/s on ERGO btw).

Also, a few posts above there is a guy (dealbreak3r) stating that he is able to get 118MH/s with the 3060 v2 (and apparently the latest drivers which is NOT 460.39). So there’s no LHR to ERGO and Autolykos.

Stop spreading misinformation please

saidikrimah · June 18, 2021, 1:08pm

I just bought 3060 v2 and combined with the 3060 v1 rig. The result is 3060 v2 cannot detected on the driver 460.39. So, I used the newest driver 465.31 using the same OC and it will drop to 50-60mh, but if using the default OC, the hashrate will stable on 102mh.

After a while, maximum got stable on 111.6mh using this OC settings:

my question is, my OC settings is danger or not? because the core value more than the memory clock value.

*hopefully my OC settings can help

hotness · June 18, 2021, 1:11pm

Your mem clock is very low - it should be 2400+.
You are right that the core clock is too high, but usually this would crash immediately.

hotness · June 18, 2021, 1:12pm

You are making some incorrect assumptions, which is forgiveable. It seems we are both spreading misinformation, since neither of us has a 3060 model 2504 to test. So we should please stop polluting the thread with useless banter.

Slxxpy · June 18, 2021, 1:25pm

I’m not sure if this will work in Hive OS but I managed to get the 2504 model to work with Windows.

I’m using the latest Nvidia driver as this is my gaming Pc but here are the settings

PL: 65% (NO MORE THAN 110W)
Core Clock: -525
Mem: + 1400
Fans: 75%

This gives me a stable 116-118 Mh/s on Windows

You guys could try to mimic these settings but double the value for memory in HiveOS ofc.

aminer · June 18, 2021, 1:25pm

My dear friend, never in my post did I make an assumption. I am only stating facts. The only person here making assumptions is you. You assumed that the two algos are very similar and the LHR kicks in for ERGO because the card thinks you are mining ETH. And you came to this assumption because you use similar OC settings for the two algos. That was were I came to correct you. If you cannot understand this then I can’t do more.

Having a conversation based on facts is not polluting. Making assumptions out of thin air is. Nevertheless, I will stop here since you offered to stop as well.

Have a nice day!