search

trexminer / T-Rex

T-Rex NVIDIA GPU miner with web control monitoring page
2.64k stars 438 forks source link

Same GPU - different Alph hash rates on T-Rex 25.8? #1129

Open LADGRES opened 2 years ago

LADGRES commented 2 years ago

I am dual mining ETH-Alph on a 6x 3060ti rig in HiveOS. For some reason, the first 3 cards have really low Alph hash rates (100-160Mh) and the last three are normal (400-500Mh)/ I have tried to set each to 74, same result. Tried to just remove the intensity setting, same result. I set up another rig with just 2x 3060ti gpus and get the same thing - the first gpu only gets around 140 Mh and the second gets over 300 Mh? Eth hash rates seem to be doing well ... anyone else see this? Alph issue

tjayz commented 2 years ago

https://github.com/trexminer/T-Rex/issues/1115

LADGRES commented 2 years ago

tjayz - I read that post (#1115) as well and already verified that ALL of the gpu's are starting with the same intensity of 21.3 - LHR 74.0. After it has been running for awhile, I have seen the LHR going as high as 77 which answers another post I read (forget which one) ... the odd blake3 hash rates remain about the same.

I also realize that some of the cards have slight differences in the overclocks, but GPU 1, 2, and 4 have same overclocks. GPU 3.4.5 have higher allowable W though - maybe that has something to do with it?

LADGRES commented 2 years ago

Well that theory has nothing to do with it (max wattage) - look at this test I did .. only 2 cards and GPU 0 "should" get much higher hash rates ... Test .

LADGRES commented 2 years ago

Ohhhh ... there is another line that pops up during startup showing "dual ratio" numbers. Why are they different?

Test2

LADGRES commented 2 years ago

Sorry for the multiple posts, but just sharing my tests ... I reduced the Memory on each GPU to 0 and now get much closer ratio numbers ... but now my ETH hash sucks :-( This blake3 algo sure is finicky ...

Test3

tjayz commented 2 years ago

Per README.md

    --dual-algo-mode           Controls GPU behaviour in dual mining mode (default: a12).
                               Format: <algo>:<tuning coefficient>
                               "<algo>" can be one of: "a1" (first algorithm), "a2" (second algorithm), "a12" (both - dual mining)
                               "<tuning coefficient>" is optional, and can be either
                                     "rXX" ("dual ratio" coefficient set to XX), or
                                     "hXX" (primary algo hashrate percentage XX).
                               Examples:
                               a12:r15 - GPU will mine both algorithms with dual ratio 15
                                         (second algo hashrate is 15x primary algo hashrate)
                               a12:h95 - GPU will mine both algorithms maintaining primary algorithm hashrate at
                                         about 95% of the maximum
                               a1      - GPU will only mine the primary algorithm (single mode)
                               a2      - GPU will only mine the secondary algorithm (single mode)

Dual ratio can be set, lhr dual ratio is set automatically based on whatever checks trex is making at start.

LADGRES commented 2 years ago

I have been messing with easy settings first (ie. Memory) and noticing that multiple GPUs need to have the same memory set when the mining starts - otherwise, it does some really weird initial ratio settings. Like any other method, it helps to 'tweak' the settings to get your best return. For my 3060tis, that is currently around 1400 Core and 1500 Memory which gives me the following:

Test4

I can get higher ETH hashrates, but then the Alph hashrates really start droping - these settings are the lowest memory while still maintaining the same return using both algos. MUCH better than when I started :-)

I will start messing with the --dual-algo-mode, but I need to figure out how the different ratios work. Doing the testing above, I have seen these ratios vary from 2/2-8/8 on the first ratio and 11/11-17/17 on the LHR ratio. Lower numbers = High ETH/low Alph

zulu1905 commented 2 years ago

your cards are not the same. asus gigabyte evga ....... and naturally different bioses. you need to increase powerlimit for that low hasrate cards.

LADGRES commented 2 years ago

zulu1905 - Agreed. That is why I set up the test with the 2 Asus cards ... they are still VERY different cards (one is a ROG STRIX and the other is a Mini Dual), but it shouldn't matter when using the same settings. You can really tell the difference by looking at the temps - this is just what I had available to test with on short notice. Figured it would give me some results that I can use on the bigger rig ...

LADGRES commented 2 years ago

Just for reference, here is the startup for the 6x 3060ti rig - notice the wide variance in the default ratios which I don't understand at all... they should all be the same or very close. I will see if I can get these closer using manual settings per tjayz

Test5

tjayz commented 2 years ago

I settled on a12:r0 because :h99 still is lottery what it pumps out for ratios. r0 will give max hashrate to eth and every increase in r by factor of 1 decreased my eth by ~1mhs. I prefer no sacrifice to eth maintaining 46mhs+/per card. lhr dual ratio defaults to 9 or 10 and on average I return 145mhs ALPH. Could get more ALPH with more watts but this is opposite my goal. Locking core on my 3060tis requires 1400+ which pumps out many more watts to get 46mhs+. Efficiency optimal setting core offset such that core hovers around 1000 at average cost of 133w when fans 85% and average lhr is 75 on drivers 472.12. I do not set pl. Results will vary for everyone.

LADGRES commented 2 years ago

I can't seem to get the --dual-algo-mode to change any default ratio settings ... trying both a12:r15 and a12:h95 - I will try yours and see what I get. I didn't even think the r0 would be an option. Anything to get every GPU to have the same ratio would be a win

LADGRES commented 2 years ago

Worked well for my 2x gpu test setup, but when I tried it out on the 6x 3060ti rig, it didn't change anything - still got wildly different ratios (8,5,4,0,0,0 and 14,14,15,7,11,9) ... hmmmm, guess I will keep working on it. edit: Looks like the ratio is only being applied to the first 3 GPUs . The ratios are 0,0,0,8,5,4 and 8,11,9,14,14,14 ... weird that it moved them around.

LADGRES commented 2 years ago

The --dual-algo-mode does not seem to work in linux (HiveOS) - it does work on my single gpu in Win10.

EDIT: I'm an idiot - works fine in HiveOS - Please remember to remove the double hyphen when using HiveOS. ie "dual-algo-mode" not "--dual-algo-mode" ...

tjayz commented 2 years ago

https://github.com/trexminer/T-Rex/wiki/Dual-mining

After some more testing past few days, decided hynix memory r1 best, samsung r2.

LADGRES commented 2 years ago

My best overall results were at r2 and gave me much more consistent numbers across all the 3060tis :-)

Test2