.53? Thats far too low. Can you provide more info: OS, driver and clocks?

That seems good
Ok, I have to admit: can not say why one card allows those settings and the other struggles. But seems your hash with slightly more clock and voltage is fine, then I would leave it at the more stable settings.

The problem about this line is the stratum+tcp://  ... just leave that out


that way it should work



Well the 1.05 at those settings seem ok ... but why it drops to 0.45 with the lower clocks is strange - can you tell me which hive image you run? (I would pick the beta one, since this has amdgpu-pro 19.30 driver. Also I do not know if hive already pushed 0.9.2 final to their repo)

Can you describe your problem more detailed?
There should be an example .bat for mining GRIN-31 included with the miner. Note that you will need Adrenalin 19.10.1 or newer on your sys. In case you still struggle feel free to dm me or join my Telegram channel https://t.me/lolMiner



I would update to 0.9.2 release just pushed - that should be even better then the beta See my msg below


Update to 0.9.2

I just pushed 0.9.2 - this one should stabilize the GRIN-31 code even better and speed it up again - by about 5% in worst case and 10% (Navi) in best case
Also we tested a lot with optimized settings for this algorithms (thanks a lot to Dobromir Dobrev, MMP OS) - it is possible to get V56 to 1.0 g/s at only 130W (1220/900/0.825v). Stock may be faster when you do not care the electricity cost, but with this setting (V64 similar is possible) C31 hits the top end of Vega profitability

Thats hard to debug from the distance. What exact cards are these and which settings do this cards have? Also what are you trying to mine with lolMiner (all the algos here are a bit different) and what lolMiner version and which driver you use. We need a bit more info here to help
 Most likely the undervolt is too low for the needed compute power... but thats only a very vague guess

That will be tough ^^ A bit better is possible, but not as much as the 8G Navis just did
Anyways, planing 0.9.2 next werk - some more tweeks to be tested before

Hey, just wanted yyou to know that I pushed a beta version to.my testing repo - not yet fully production ready, but would appreciate some testing. Here is the link: https://github.com/Lolliedieb/lolMiner-preview/releases/tag/0.9.2beta

Only changes so far is some slight improvement for C31 (while 4G solver is defect atm - will be fixed before release).

Testing results (v0.9.1 -> 0.9.2 beta)

Radeon VII (stock): 1.7 -> 1.76 (+3.5%)
Radeon VII (1400 / 1000 @ 0.825V): 1.35 -> 1.4 (+3.7%)
Vega 56 (1350 / 920 @ 0.9V): 1.0 -> 1.03 (+3%)
RX 5700 (Powercolor Red Dragon, Silent Bios): 0.78 -> 0.84 (+7.6%)

Let me know more or if there are issues with the beta. Release of official version some time next week. Want to go over more code sections

Ah, now we are talking about the right range

Glad you got it working

By the way I got a telegram channel now for support: https://t.me/lolMiner

Thats odd... is this 0.9 or 0.9.1? Please check!
I ask because even the 0.9.1 4G solver should be much quicker.

If its 0.9: then the 8G does not load because the Binary Kernel can not be loaded --> install Adrenalin 19.10.1 or newer in Windows or 19.30 in Linux. Also upgrade to 0.9.1, pool hash is better with that one.

Well ... hmm .. then WTM is off ^^
But even then it does not look too bad - I have to admit with 50 at 170 Ethash wins, next then is SUMO (CNR) and then next is Grin-31 ... so ok right, Ethash is in front but not that much - at least its good enough it should not be such a niche

A good idea! Will note that for near future features to be added.



Thanks John for pointing that out - I will prepare a miner for the fork





The problem with Cortex network is that its too small ... gminer was just earlier discovering it and now making a miner is kind of worthless. I mean... Cortex Cuckaroo-30 is an interesting algorithm to tune and it would be a lot of fun. Still I must ROI my time somehow. Just to drop a number: With current C31 code Vega 64 does 1.1 g/s at 180 W (settings: 1380 / 1000 @ 0.912V) making it the 2nd most profitable algorithm after CNR for V64 (!!!). Yet at my fee connector I get about 1 g/s meaning not even 100 of this cards mine this algorithm with my miner. (By the way: stale shares is ~2% and 0 invalids ... thats record breaking for this algo...)
So I am making maybe 35 cents each day from this algo - if it continues this way I can not expect that the time put into coding this ever ROIs for me. Then starting a new similar algorithm for a much smaller network would be just insane - If someone wants to fund the development things turn. For ExchangeCoin for example we had the same problem, but there the devs funded me (3 digit number, so not even too expensive) to build a proper miner, thats also why Exchange Coin was 0 fee in the first releases.

I only did a short run on an under-clocked one at about 0.9 g/s ... not sure what a normal could do - it should be around 1.05 to 1.1... I will get one soon to test that.


Not yet ... but will come once I get my own one. Till recently it was extremely hard to get this cards to do anything useful at all

I just pushed a hotfix. Can you check if that works with your card?
https://github.com/Lolliedieb/lolMiner-preview/releases/tag/0.9.1hotfix

Thanks for the report
Well the 290 is a 4G card - it has to run some time memory trade of to mine this algorithm at all - the 390 with 8G (but same chip) would be much quicker. This 4G is rather for grin enthusiasts




Thanks very much! Thats quite good. Please report if thats consistent with pool. Flagged Navi support as beta because was not able to fully verify it, because I still wait for my own. Currently sometimes test on a friends rig ^^




Let me check that. Seems like a bug in my GPU detection, the empty brackets indicate my miner believes it does not have the kernel for this. Will check and fix

Edit: Found the bug why Fiji is not loaded correctly. Code is already fixed, will release it asap.


Ps:
Wanted to mention that these 7mn AMDs as the VII and the Navi are undervolting monsters.
For example stock the VII gives 1.7 g/s at mine, but at 280W. Clocking down to 1450 at 0.82v costs a little performance, but the result is 1.43 g/s at only 130W - if you need to look to the energy bill thats the much better option Expect similar efficiency with Navi
(By the way the Sys where we tested V64 was running 1380 / 1000 @ 0.912V... approx 1.12 g/s at 170W).

Always welcome Guys

I pushed a new version (0.9.1) today adding support for 4G and 16G cards and experimental one for Navi

Also users of 0.9 should update, because in some situations it was possible that 0.9 gave lower pool hash for Grin-31 then it should give - this was fixed with 0.9.1

Have fun mining

We had that problem with qweb1 (there its solved by the way): You will need Adrenalin 19.10.1 WHQL or newer for the new 8G Kernels - on older drivers loading the Kernels will fail and thus the miner jumps to the next best option on same Algorithm (which is the old 4G solver).

This indeed can happen with Nvidias on this Algo - thus I never announced Nvidia support for this one
Also its loading the old 4G sources kernel and not the ones that are new in 0.9, because I did not create new Nvidia kernels.
I suggest you test with --devices 3 to pick the 580
For the Nvidias I would run bminer / nbminer along with lolMiner.

By the way: --workbatch is deprecated since ... 0.8 - it just has no effect any more

Hmm ... can you check if it really loaded the new kernel and not the old 4G one?  Because here on test system (Linux, 19.30 driver & Windows 10, Adrenalin 19.30) the V56 is at ~1.04 g/s. What I could imagine is that if you run an older driver or ROCm then the load of the binary kernal fails and it falls back to the backup 4G solver.

Here what the testsys (2x 56, 1x 64 in the middle, optimized clocks for performance/watts) just produced:

Edit:
In case its ROCm: will provide ROCm builds in next version

Yes, 5700 (XT) is currently a problem. Background is that one function compiles wrong and thus produces a low fidelity on the new code. I will try to find a way around that (else expected would be 0.9 g/s approx)

Other values from testing (0.9 Cuckatoo-31):
RX 580, 8G: 0.55 g/s
Vega 64: 1.15 g/s
Radeon VII: 1.53 g/s

Next days we will get a new 16G code (should push the VII over 1.75 g/s) and a new 4G as well (expected 0.4 g/s on RX 580 4G).

Yes indeed it could be OC settings. The problem when you run settings that are good for ETH is that those are pushing memory a lot and same time give very low voltage to the compute core. But both Equihash and also the Grin algorithms are much more compute intense (and the algorithms are more sensitive for memory or cache bits flipping), so I rather would suggest to start with stock settings, then reduce core clock a bit (e.g. 10%) and then start to up mem clock and undervolt the core as long as it remains stable.

Alternatively some mining OSed (e.g. MMP OS) have already OC / UV profiles adjusted for the combination of mining software and card. That helps maxing out performance / watts rather easily