Pages:
Author

Topic: [XMR] JCE Miner Cryptonight/forks, now with GPU! - page 41. (Read 90814 times)

member
Activity: 350
Merit: 22
0.33b4 GPU is out

Hello i'm new into mining and i have one xeon skylake 12 cores, 24 threads, 56mb L3 cache and one xeon 10 cores, 20 threads with 56mb L3 cache too, in default settings i was able to get 1700h/s using cryptonight lite, is there any way to optimize my results? I was Reading about the multi hash but i don't know which config to use, and how much is the value for hugepages?

Huge pages should be enabled, on Windows 10 this is automatic if you run the miner as admin once, otherwise do it manually, here's a guide:
https://github.com/jceminer/cn_cpu_miner#large-pages

Do not mine without huge pages, the performance loss is very noticeable (like -15%)

For the threads, here's an example

Code:
"cpu_threads_conf" :
[
     { "cpu_architecture" : "auto", "affine_to_cpu" : 0, "use_cache" : true, "multi_hash":2 },
     { "cpu_architecture" : "auto", "affine_to_cpu" : 1, "use_cache" : true, "multi_hash":2 },
     ....
     { "cpu_architecture" : "auto", "affine_to_cpu" :23, "use_cache" : true, "multi_hash":2 },
]

For your 24-cpu xeon. Try it, and then apply this tweak:
* Change the first and last multi_hash (so, for CPU 0 and 23) to 1 and/or 3, and see if it's better.
* Then split the range in two (making two ranges: 0-11 and 12-23) and tweak again the respective 1st and last cpu: 0, 11, 12 and 23.
* Then split the ranges again in two (making four ranges) and again and again until you find the optimal setting.

The same applies to your 20-cpu xeon, and also the same if you mine CN-class algo (Monero, Stellite...) because you have a lot of cache
But config for Heavy would be very different, involving less threads but some no-cache mode.
member
Activity: 350
Merit: 22
TeamRed is new, and they experience the same problems as I got at the beginning: lack of Fan/Temp monitoring, missing info in the log, stability... They don't even have colors. But i know they will add them.

Fine if i beat SRB, and since Claymore is out of v8 support, now this is TeamRed vs. JCE Cool
I found what to do to get the +5% perfs on Vega. But not how.
My release happening one day after TR is not because of a hack (i always write my very own code) but not coincidencial, as I said, they proved there were some extra perf to pull from the AMD GPUs, and when you know something exists, so it's easier to find. But so far all my attempts on vega GPUs failed, probably because i lack some skills in Vega GCN.

I'm now a lot more confident I can fill the performance gap.

Enough GPU for now, i focus back on CPUs support.
newbie
Activity: 54
Merit: 0
Great job with the 0.33b4  Cool

GPU 0: RX 570 4G Samsung
GPU 1: RX 570 8G Micron

Code:
13:00:28 | Hashrate CPU Thread 0: 81.44 h/s
13:00:28 | Hashrate CPU Thread 1: 82.17 h/s
13:00:28 | Hashrate CPU Thread 2: 82.37 h/s
13:00:28 | Hashrate CPU Thread 3: 82.37 h/s
13:00:28 | Hashrate CPU Thread 4: 83.51 h/s
13:00:28 | Hashrate CPU Thread 5: 83.66 h/s
13:00:28 | Hashrate CPU Thread 6: 83.28 h/s
13:00:28 | Hashrate CPU Thread 7: 83.34 h/s - Total CPUs: 662.11 h/s
13:00:28 | Hashrate GPU Thread 8: 505.92 h/s
13:00:28 | Hashrate GPU Thread 9: 505.83 h/s - Total GPU 0: 1011.74 h/s
13:00:28 | Hashrate GPU Thread 10: 502.48 h/s
13:00:28 | Hashrate GPU Thread 11: 502.48 h/s - Total GPU 1: 1004.95 h/s

You're now far beyond SRB, and your miner works fine, unlike TRM that only manage to crash my graphic cards.
member
Activity: 350
Merit: 22
Quote
12:40:33 | Hashrate CPU Thread 0: 23.08 h/s
12:40:33 | Hashrate CPU Thread 1: 23.40 h/s
12:40:33 | Hashrate CPU Thread 2: 24.33 h/s
12:40:33 | Hashrate CPU Thread 3: 24.47 h/s - Total CPUs: 95.26 h/s
12:40:33 | Hashrate GPU Thread 4: 251.23 h/s
12:40:33 | Hashrate GPU Thread 5: 253.30 h/s - Total GPU 0: 504.53 h/s

That's my Core2Quad 2.666 and my RX560 2G with --auto on v8, looks like the b4 version burns like fire. It may be a false positive, but at first try, there's a noticeable perf boost even on RX. Feedbacks on other cards welcome.
member
Activity: 350
Merit: 22
Quote
please do not drop support for old maps on your miner i have better results
r7 370 2gb - monero v7 srb 1.6.9 --495h / s
 --monero v8 srb 1.6.9 --- 366h / s
                   jce 033b2 --- 422h / s
thank you very much for the miner for old video cards

Thanks, and so long, if TeamRed stay far above my miner on RX/Vega, that's what will happen, i'll focus on older cards and CPUs, were i'm the best. It may sound strange i advertise for a competitor in my own topic, but those guys deserve it, they did a really great job. When I came on v7 six month ago, there were an obsolete Claymore 11, Cast which were fast only because it was the first to use correctly the Compute Mode introduced by the beta blockchain, and SRB with perf not that better than the legacy stak/xmrig.

Here we're talking about +5% above SRB, which is already +2% mine. That's a real challenge.

0.33b4 under test, to be released soon.
Heavy: i may gain some +0.1% by backporting the v8 optims, but i'm running out of ideas to get extra speed here.

edit: i benched on my HD6950 and I get... +20% speed. Looks like the older is the card, the higher is my performance gap. It goes from +20% on HD6000 to -5% on Vega.
The 0.33b4 will probably worth to be benched on Vega, still, i may not have +20% but even a +2% would fill in the gap with SRB.
sr. member
Activity: 1484
Merit: 253
so you're mining turtle or similar? that's the hardest algo to configure, i take a look and provide some config you can try.

i should release today 0.33b4 with…. +5% performance on cnv8 Shocked
not a joke, but that's for Pitcairn. i benched on Tahiti and still got +3%, and +4% on Bonaire. i had no time to test on other gpus, and notably not vega.

no i didn't hack TeamRed, but i admit that gave me the clue there were a hardware +5% reserve in AMD gpus. i found it in old gpus, but maybe not in new ones.

Good news! What about heavy algo?
jr. member
Activity: 145
Merit: 1
so you're mining turtle or similar? that's the hardest algo to configure, i take a look and provide some config you can try.

i should release today 0.33b4 with…. +5% performance on cnv8 Shocked
not a joke, but that's for Pitcairn. i benched on Tahiti and still got +3%, and +4% on Bonaire. i had no time to test on other gpus, and notably not vega.

no i didn't hack TeamRed, but i admit that gave me the clue there were a hardware +5% reserve in AMD gpus. i found it in old gpus, but maybe not in new ones.


Oh yes hope you can find it for rx 580 !
member
Activity: 350
Merit: 22
so you're mining turtle or similar? that's the hardest algo to configure, i take a look and provide some config you can try.

i should release today 0.33b4 with…. +5% performance on cnv8 Shocked
not a joke, but that's for Pitcairn. i benched on Tahiti and still got +3%, and +4% on Bonaire. i had no time to test on other gpus, and notably not vega.

no i didn't hack TeamRed, but i admit that gave me the clue there were a hardware +5% reserve in AMD gpus. i found it in old gpus, but maybe not in new ones.
newbie
Activity: 52
Merit: 0
Hello i'm new into mining and i have one xeon skylake 12 cores, 24 threads, 56mb L3 cache and one xeon 10 cores, 20 threads with 56mb L3 cache too, in default settings i was able to get 1700h/s using cryptonight lite, is there any way to optimize my results? I was Reading about the multi hash but i don't know which config to use, and how much is the value for hugepages?
newbie
Activity: 26
Merit: 1
guys! i recently got a rx550, 2gb. Any idea if i can get more than 300h/s on cn-heavy? Cheesy
No, 2Gb too small amount to have normal speed on cn-heavy. You can increase speed only by use overclocking and editing memory straps in bios.

Yeah! I know, i have the bios modded and tried some different configurations of OC without luck...
Maybe someone had good ones Wink
Thanks man!
newbie
Activity: 8
Merit: 0

Quote
"Failed to allocate shared Large Page, using normal memory (slower). Failed to allocate 2MB Large Page Scratchpad Buffer, using normal memory (slower)."
That's a message about the CPU mining, because --auto also configures the CPU. To stay on GPU only, add parameter --no-cpu


I tried and with the --no-cpu in auto mode and with the command -c config.txt Error: CL_BUILD_PROGRAM_FAILURE Code: O-2.10 "
newbie
Activity: 33
Merit: 0
Hello, I am writing here for the first time
I have r7 370 2gb video cards in my farms
used to use srb miner, but lately support for old cards in it began to limp, when switching to v8, the hashrate dropped significantly, calculation errors appeared, wrote no changes to the srb miner topic
please do not drop support for old maps on your miner i have better results
r7 370 2gb - monero v7 srb 1.6.9 --495h / s
 --monero v8 srb 1.6.9 --- 366h / s
                   jce 033b2 --- 422h / s
thank you very much for the miner for old video cards
the rest concentrated their attention on rx / vega cards where big competition forgetting about other generations
you will be able to occupy your niche in the market of miners, about which others have forgotten
sr. member
Activity: 1106
Merit: 251
Thanks JCE Miner. It's great to mine cryptonights using JCE. It is just the best miner that I have tried. It is very easy to use that even the newbie miners will agree with me.

Keep up the good work man.
newbie
Activity: 4
Merit: 0
I can recognize when a dev is good, and they are.

They are two people, which is a important point for time. Look how i messed up with the CNv8 deadline because i was away at the wrong time. May have had a buddy dev, it wouldn't have happended. Two people specialized on cn-v8 for RX/vega against me, alone, having to support CPU aes, and non-aes, 32 and 64 bits, and GPU... I cannot win.

I may specialize in older cards they don't support, or just give up all GPUs. But +5% is a huge gap, i don't think I can fill it up.
Still need to bench the real hashrate, after all Claymore 11 was also +5% on screen, but -5% on pool. But beating ROCm with OpenCL is hardly possible.



I agree with you JCE  ..  the cpu miner is what put you on the map & even after the v8 fork i still use it as it give me the best results on my CPU miners even after a h/r reduction & extra power draw.
The v8 fork seems to have favoured the RX & VEGA GPU's as most are now getting v7 hash rates and in some cases even more with the various miners, tweaks & optimizations.

I don't see the point in spending time to give maybey the same or +/- 1%-3% hash rate in an already saturated GPU miner field. Everyone is chasing the RX & VEGAS and completely ignoring the lucrative
older CPU's which in some cased have taken a up to 25% drop in hash rate with  intel ,Xeons,AMD & opterons both 32 & 64 bit which i have experienced myself as i mine with all of them.


Many miners have gone offline as it's no longer profitable.

FYI:   i mine with RX 570's also and pre-fork claymore's miner was tops for me in hash rate/power draw & stability but on v8 i now use xmr-stak which when run dual thread is giving me a higher
h/r than i was getting on the claymore v7. Power draw is slightly up but its otherwise rock solid & stable so i personally would'nt bother to switch miner to get a 1% - 3% increase in hash rate.


I would hope you will find time to see if miner optimization can reduce the loss suffered by older the CPU's as in truth it is what i want to see now as there are already plenty of GPU miners to choose from.
newbie
Activity: 39
Merit: 0
member
Activity: 1558
Merit: 69
member
Activity: 1558
Merit: 69
Someone has good settings for MoneroV8 for the RX550? I can´t get a stable hashrate. I have  6 similiar RX550 with same bios. Hashrate is very unstable and mostly 1 card is slower then all other, and this change all 5min. Monero V7 works stable, and with SRB it works stable too, but  with srb the effective pool hashrate is 400h slower then the reported hashrate.
Memory is 2Mb?
Try this parameters:
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 8, "gamma" : 8, "delta" : 8, "epsilon" : 8, "zeta" : 8, "index" : 1, "multi_hash":464 },
{ "mode" : "GPU", "worksize" : 8, "alpha" : 64, "beta" : 8, "gamma" : 8, "delta" : 8, "epsilon" : 8, "zeta" : 8, "index" : 1, "multi_hash":464 }

Then try to increase both multi_hash parameters untill your card have enough memory or speed became lower even at start.

I have RX550 4GB. I tried worksize 8,12,16 and Multihash i tried 400,416,432,448,464,480,496 in many configs. In every Config it jumps up and down. Driver is 18.6.1. With SRB i get stable hashrate at around 2850 hashes.
I manage over 150 GPU´s, so i am not a noob ^^ Maybe it is a driver problem? On my other rigs with JCE Miner i have the beta blockchain driver installed and old pitcairn and tahiti cards, there it works perfect.
sr. member
Activity: 1484
Merit: 253
JCE, thanks for your hard work! Miner is great!

Is there some room for heavy algo optimizations? Especcially for 4Gb cards?
member
Activity: 350
Merit: 22
thanks!

i don't give up instantly, and i may keep some niche like Tube and older gpus. but they're really good and will probably add the missing algos. but it doesn't mean they will be faster on Tube, where i'm pretty good. just they added a big +5% on a new algo, and if they are better on both power and hashrate, fees deduced, i'll lose reason to continue my code.

i'm preparing an optimized version for rx560 and rx550, if i get a +2% i may fill the gap, if i get a pitty +0.1% i may giveup. i've more tests to do.

@vmozara: ok i trust you their hashrate is correct, i won't re-bench my side. i know what perf i need to achieve to stay in the race.
jr. member
Activity: 145
Merit: 1
Hello JCE i still use your miner in all my rigs because TR have not heavy algorithms and they are better than CNv8 ATM, on heavy your miner is far better than others ! (i got 1170h/S for 90W per rx 580).

Don't give away man they would launched heavy algos if they were already better than you.
Pages:
Jump to: