Topic: [XMR] JCE Miner Cryptonight/forks, now with GPU! - page 29. (Read 90807 times)

i'll put back the q version online, otherwise use a git client on github to get all historical versions.
version b-x to be released soon. it hashes solid on my RX, but i already said such for the b5...

i still use most of my dev time for the cpu version.
TeamRed are great, i think i know how i can achieve low power, but i need time to test. at least i somehow reached equivalent speed, and i've less fees on v8 (0.9 vs 2.5)
but counting the power saving, they're the best.

Tested TeamRed cn8 miner... Speed is almost the same as JCE, only one GPU a bit faster... But his miner uses real lower power on cn8 than JCE... Interesting, how they can do that?

4x rx570 4g  3400hs at 440w at the wall mining Tube
Stable as for a few days now on b5 version.
Took a bit of fiddling to get it stable but were else can you get 850-900hs on heavy with a 4g card
B5 version is great. Thanks
Also works great with my vegas hashing at 1800-1950hs on heavy.
Verified the same on the pool side too.

This is an 8x nitro 580 8gb rig - different mem mfgs - on 32q mining saber (TUBE). ~785W (98W/GPU):  

EDIT: power above is ex-idle.  Including idle, would be ~905W total, 113W/GPU


uptime 10 days currently - doesn't really ever go down except for power outages, updates, or major network issues.  heavy/haven do better on same settings, for the most part.

haven't tried 33b[anything] yet

35 hours (and counting) mining XHV:
2170H/s with 3 RX570 4G, one being my display GPU.

That's around 10% more than using the 0.32G version. with the 0.32G, both core and mem clock were higher than now (currently 1230 / 1230 / 1220 core; 2100 / 2010 / 2020 mem).

Power draw including the CPU mining TRTL: 360W, steady at the wall.

JCE, make x-version. I'll test it on 270X Pitcairn and RX 580 8Gb cards.

An IGP is a GPU, even if physicaly located Inside your CPU.
Both of you, the problem is the Intel IGP/GPU, to fix:

* run JCE with --probe
* Note the index of your Intel GPU, probably GPU 0
* run jce with param -g 1,2,3,4,5 (that's an example! make the list of all your GPU but the Intel one. So my example is when you have one Intel at index 0 and five AMD), there's no space around the commas in the list.
* or, use the manual config and don't set any config for the Intel GPU

The b5/b6 have great potential, and work fine on some Vega rigs and old cards. But there's that big hashrate drop issue, i know. For some config it happens always, for some, never...
I experimented, for science, the mix of the Hybrid miner for HD7000 on my RX and the result is... surprising
I got a stable and pretty high hashrate, yet maybe lower than the b6 when it works, at the cost of a few invalid shares, because the sync is not adapted to the RX memory controller.

So I plan to release a version b-x, x for eXperimental, targetted to HD7000 owner to experience the perf boost from Hybrid (which is quite solid) and to the RX owners who have the drop problem, in case it hashes better. You will have a few bad shares, but if i get feedback the performances are good, so i'll fix the shares and release it as b7.

That's on purpose, the netcode continues to run during a pause, but the devices are idle.

The b4, sure. The b5/b6, depends if you're lucky

A tiny bug: the miner keeps getting new jobs from the pool while everything is paused:

Abort was called at 304 line in file:
d:\qb\workspace\19045\src\vpg-compute-neo\runtime\os_interface\windows\wddm\wddm.cpp


How to solve this problem？

GPU:8X470 4G  18.6.1

windows10 2019ltsc

Ok guys i have a solution to make it clear, let's talk about hashrates and power consumption.

My testing Rigs (2 rigs with exactly same config) :

• 6*Rx580 Msi Armor 8Gb (Samsung memory) with celeron and other low pow shits.
  

With 33b2 here are my rates on heavy algos :

• 6960 Kh/s for 750W at wall

Very stable, it can run days without crash or fluctuating hashrate. On pool side

What about you ?

I tried again and again with b5 and b6. Never got same exact hashrate for hours and thebest i obtained was 7200hs, so with the higher fee it is better to get the stable 6960hs. I hate those versions (too long time testing for nothing )

It's strange... All new versions was better than previous, except b5 and b6. On my RX 580 8Gb b4 is better than all others...

On b6 - if mining goes long time - 20+ hours than speed can became max and didn't drop. But not always...

After trying all versions, I think 032q is the best for RX 8GB - speed & stability.

Why not 33b4?

Absolutely,

But not with the latest version because the hashrate is very unstable and the fee more than 2 times higher.

I still mine with v.033b2.

So does this mine any better with the rx580 8gb cards than xmr-stak?

Hi all!

Things are going fast this month!

So far the status is:
* I'm done with my hybrid miner, quite disapointed. It works but is barely faster. I'll release it for old GPU, i benched it to be ~3% faster on HD7000, nice to have. But the breaking change i expected won't come and the hybrid version will be just a regular version, yet faster. At least i could make it run with negligible impact on CPU, even non-aes ones.

* I expected TeamRed to release a fast CN-Heavy version, but that's SRB who came, and it's pretty solid. I benched it to be slightly slower than mine, but with more stable hashrate. So on cards with the hashrate bug, it provides higher yield, otherwise i keep the lead on extreme cases: well tuned vegas like vmozara's ones, or very old cards, but the bug is very frequent on hi-mem RX, the most common cards.

* I'm still working on the CPU version to add a better MoneroOcean support, and add CN-Fest

* Otherwise the next GPU will get bakports of CPU, the Hybrid for HD7000 and, if i success, the stabilized hashrate.

I still keep an eye on TeamRed, if they release a CN-Heavy as good as their CN-v8, they will take the lead once for all. Really those guys did a geat job.

What are the log lines logged before the crash?
Do you have an IGP in your computer (AMD or Intel) ?

I'll take a look, probably a case i didn't think about. It's a cosmetical bug, but needs to be fixed of course.

630 h/s, this is the best stable result I was able to get from srb.
I tried many options but max I got from your miner was 610h/s but after some time this speed droped to 560-580h/s.

JCE, there is a small problem in the statistics of stakjson. When the miner reconnects to the pool, Uptime is reset, but Hashes is not reset to zero. Because of this, an effective hashrate is mistakenly considered. We need to reset the Hashes as well.

Hi JCE any news on rx580  8Gb cards bad hashrate problems ?