[JCE]Fast & stable CN/v8/Heavy/Tube/XHV miner, CPU+GPU, Vega56 1800+ RX580 1200+

[livada]

xmrig is no longer a clone of Wolf0 miner, they really provide their own code with own optimizations.
Including some i found 6 months ago, but not all, so i don't think they hacked my miner, just they knew where to look for performance, as TeamRed hinted me, and i hinted SRB. Experienced devs facing the same problem find the same solution, normal.

Online is the 0.33b-y

* A lot less tested than a regular version, i hope there's no bugs out there
* Fees lowered back to 0.9% on all algos, as I'm no longer 20% faster than the reference code
* More hybrid, so should eat a bit more CPU
* Bad shares on Heavy fixed, at least i had zero during my 5-hour test session
* Light optim ~2% for HD7000 is backported to HD6000
* It was more stable on my rig, but i'm doubting of such observations

That's the standalone .exe on Github, as for the previous b-x

This version block my 6*vega 56 nitro rig  after 1 min  in ARTOcashe and XTL
Tube have ony 9000HR- last version have 11600HR.
Setup is identical.

[1714646003]

1714646003

[1714646003]

1714646003

[1714646003]

1714646003

[JCE-Miner]

thanks all for your feedbacks.

i expected a speed regression on big cards, the b-x/y is rather for the old cards, but is the varying hashrate fixed in such case ?

there's nothing i can do about your browser or antivirus blocking the download, you have to add exceptions or disable security temporarily. jce contains no malicious code, it's just a miner.

there should be a little boost for Hawaii but didn't have time to decently bench.


i take note of the display bug in the report, i've also to make the reconnect to reset the hash counter

[UnclWish]

On 588 cards speed on b-y version is a bit lower and a little more stable, but still start to fluctuate after some time... Bad shares became rare...

[pbfarmer]

thanks all for your feedbacks.

i expected a speed regression on big cards, the b-x/y is rather for the old cards, but is the varying hashrate fixed in such case ?

there's nothing i can do about your browser or antivirus blocking the download, you have to add exceptions or disable security temporarily. jce contains no malicious code, it's just a miner.

there should be a little boost for Hawaii but didn't have time to decently bench.


i take note of the display bug in the report, i've also to make the reconnect to reset the hash counter

It's looking to me like 33 in general is a regression for heavy on 8g polaris.   I am able to get to ~1175h/s on 32q after 15min or so, and it pretty much sticks there.  On 33b-y, with same exact settings (both miner config and gpu) I can only get to 1025-1050, though it does get to stable faster it seems.  In a 12 hr test, it did touch 1165 for a single reporting cycle (hours into the run,) but immediately returned to 1035.  I am now running 33b5, and while it's only been up ~30min, it does not appear to be any different from 33b-y.

EDIT: Looks like i was wrong - about 1.25 hours in on my 33b5 test, h/r ramped up to 1250.  Will let it continue running to test stability.

[pbfarmer]

@JCE-Miner - this brings up a feature ask...  Can we get a configurable watchdog delay?  The hour-long warmups aren't terribly bothersome (to me) as long as the miner is stable, but the 5-min watchdog is useless for these scenarios.

[DXInfinity]

Hello JCE-Miner, can you please look into temperature/fan monitoring? It does not work for me at all. Temp and fan are always zero...
I'm using 18.6.1 driver version. Windows 10 1709 and 1803.
Miner version 0.33b6 (tried previous versions - same result).
Cards are different - mix of RX470/RX480/RX570/RX580 - Asus and Powercolor.
This is reproduced on all my rigs with different hardware and even platform (intel and amd).

[UnclWish]

I think that 1st what need to do - is stabilizing heavy algo speed, 2nd - lowering power consumption for v8.

[Boss_One]

Online is the 0.33b-y

* A lot less tested than a regular version, i hope there's no bugs out there
* Fees lowered back to 0.9% on all algos, as I'm no longer 20% faster than the reference code
* More hybrid, so should eat a bit more CPU
* Bad shares on Heavy fixed, at least i had zero during my 5-hour test session
* Light optim ~2% for HD7000 is backported to HD6000
* It was more stable on my rig, but i'm doubting of such observations

That's the standalone .exe on Github, as for the previous b-x

I'm become crazy , i've 3 PC and ALL Crash, with 14.4 driver and 6950HD, i've allready post my config.
Could u help me ? i'm very glad to fee you my hash

I must use this 14.4 driver with your miner ?  i'm the only one who have problem with 6950.

[UnclWish]

Online is the 0.33b-y

* A lot less tested than a regular version, i hope there's no bugs out there
* Fees lowered back to 0.9% on all algos, as I'm no longer 20% faster than the reference code
* More hybrid, so should eat a bit more CPU
* Bad shares on Heavy fixed, at least i had zero during my 5-hour test session
* Light optim ~2% for HD7000 is backported to HD6000
* It was more stable on my rig, but i'm doubting of such observations

That's the standalone .exe on Github, as for the previous b-x

I'm become crazy , i've 3 PC and ALL Crash, with 14.4 driver and 6950HD, i've allready post my config.
Could u help me ? i'm very glad to fee you my hash

I must use this 14.4 driver with your miner ?  i'm the only one who have problem with 6950.

You must have Windows 8.1+. What config you're using? What amount of video memory your cards have? And what algo you what to mine?

[UnclWish]

JCE, can you make 33b5 version with devfee 0.9%? b5 looks best on heavy for my 588 cards... better b-y version...
On b5 speed fluctuations some time drops speed - and some time up it. On b-y version after 10-15 minutes speed srops and never up...

[JCE-Miner]

i'll release a b-z (the last experimental since i'm running out of letters) with restored speed on big RX/Vega and then an official b7, both with 0.9% fees.

zero fanspeed/temp : i got this bug reported several time but i couldn't reproduce it yet, so it's hard to fix. I'll take a new look for a quick win in case it's easy.

stuck hd6950: if it gets stuck with any miner, it may be a PSU problem, the PSU not affording the needed power.

speed on HD6000: if we do the maths, speed ~280/hs for a multi_hash of 864 means each round takes a huge ~3.5s, so you can get a lot of stale shares. The CPU otherwise does each round in a split second, so it may be normal you get a lower effective hashrate in such case. Try with a  very low fixed diff and/or a pool with less aggressive job renewal (xmrpool.eu may be a good candidate, and Nicehash the worst one).

[Bulitt]

JCE CPU miner crashed on me last night for the first time ever. I have no idea why either no logging!

[JCE-Miner]

the logs are disabled by default, but in such case there would probably be zero hint in the log.

Online is the 0.33 GPU B-Z

Like b-y but more of the same:

* Still more hybrid
* Mix of the b5 and the by
* Fees still at 0.9%

[christiano88]

the logs are disabled by default, but in such case there would probably be zero hint in the log.

Online is the 0.33 GPU B-Z

Like b-y but more of the same:

* Still more hybrid
* Mix of the b5 and the by
* Fees still at 0.9%

So speed on big RX is not restored?

[UnclWish]

the logs are disabled by default, but in such case there would probably be zero hint in the log.

Online is the 0.33 GPU B-Z

Like b-y but more of the same:

* Still more hybrid
* Mix of the b5 and the by
* Fees still at 0.9%

So speed on big RX is not restored?

Speed on RX 588 is high and not stable. But from all previous versions speed is maximum(like b5) and most stable when mining lasts long time...
Who used b5 version - recommend to change it to b-z.

[christiano88]

the logs are disabled by default, but in such case there would probably be zero hint in the log.

Online is the 0.33 GPU B-Z

Like b-y but more of the same:

* Still more hybrid
* Mix of the b5 and the by
* Fees still at 0.9%

So speed on big RX is not restored?

Speed on RX 588 is high and not stable. But from all previous versions speed is maximum(like b5) and most stable when mining lasts long time...
Who used b5 version - recommend to change it to b-z.

b5 was never able to get higher effective hashrate than b4 and q versions on the long term, I don't think you did your testing right. Since we're on this subject of bx versions, since b4, stability degraded. On b4 I found each morning 1 or 2 rigs would go offline, those work right with q version for example.

[JCE-Miner]

the b5 and above are to be faster than the b4, but suffer from the hashrate instability problem. It really depends on your config. UnclWish is my most annoying user, so when it says there's an improvement, there is.
I'm preparing the b7 with (yet another) stability fix for Heavy hashrate, but i'll release a new CPU version before.

Unclwish report on b-z is what I expected: restored speed on RX/Vega but the stability bug remains. I still advise it as it has lower fees than b5 and b6. It also has a light improvement over b-y on Tahiti and Pitcairn GPU on v8 (~0.4%).

so next will be 0.33g CPU with:

* more MoneroOcean features
* Cryptonight FreeHaven

Cryptonight-Fest happens to be the exact same as Cryptonight-Arto, that i support for a long time. I'll just update the doc, but there's no new code out there.

[pbfarmer]

the b5 and above are to be faster than the b4, but suffer from the hashrate instability problem. It really depends on your config. UnclWish is my most annoying user, so when it says there's an improvement, there is.
I'm preparing the b7 with (yet another) stability fix for Heavy hashrate, but i'll release a new CPU version before.

Unclwish report on b-z is what I expected: restored speed on RX/Vega but the stability bug remains. I still advise it as it has lower fees than b5 and b6. It also has a light improvement over b-y on Tahiti and Pitcairn GPU on v8 (~0.4%).

so next will be 0.33g CPU with:

* more MoneroOcean features
* Cryptonight FreeHaven

Cryptonight-Fest happens to be the exact same as Cryptonight-Arto, that i support for a long time. I'll just update the doc, but there's no new code out there.

@JCE-Miner - I honestly have been experiencing these h/r inconsistencies at startup since before 33, but the watchdog is a great way to deal w/ them.  A configurable delay on the watchdog (from startup) would be a great intermediate solution to these problems (5 min is too short in many cases)

[JCE-Miner]

it's an idea, but as a workaround. I really want to have the hashrate just stable, with no tricks involved.
Once i'm done with the new CPU version i go back to this problem as top priority.

[danspasiva]

it's an idea, but as a workaround. I really want to have the hashrate just stable, with no tricks involved.
Once i'm done with the new CPU version i go back to this problem as top priority.

Hey JCE

I figured out a manual fix to get those rx 580 cards to hash at max speeds on Heavy and be stable, any time i see one that has lost the max speed, i pause/unpause 4 times each (8 button pushes) in quick succession and the card comes back to maximum speed.  May be something like this can be coded into the miner?