[XMR] JCE Miner Cryptonight/forks, now with GPU! - page 104.

UnclWish

sr. member

Activity: 1484

Merit: 253

Quote from: JCE-Miner on May 08, 2018, 07:59:16 AM

i get it, but how non-assignment can help ?
If i've 8 cores for example, isn't it better to assign 4 cores to get the turbo, rather than having 8 cores running anywhere at 50% each, with no turbo ?

If use all 8 cores there is no way to use turbo and there is no effect on no assigning. And, of course, it's no matter.
I mean for these who not use all CPU to mining, like me ))))
Especially on CPU's on wich almost max speed can be achieved without using all cores, like my FX8320. Max speed is about 250 h/s. It's need to say that this result can be recieved with 8 single cores, or 6 single cores, due to small cache result between 8 threads and 6 differes not much.
200+ h/s can be achieved with 3 threads - thanks to double threads and turbo techno... But with assigning threads to cores turbo techno works not so efficient as without assignment.

Result is 70-80% of max hashspeed + only 40-50% of CPU usage + low power consumption + user can normal work with computer. Even play not hard games.

Iamtutut

full member

Activity: 1120

Merit: 131

Quote from: JCE-Miner on May 08, 2018, 12:40:41 PM

0.27 online - major update

Code:

Fix a possible crash when mining IPBC with assembly "generic"
New fork: XTL for Stellite, that's --variation 7
Fixed Loki wallet detection
Updated the documentation and examples in the .zip

What's the purpose of the variation for stellite ? I'm mining it with your 0.25 version.

JCE-Miner

member

Activity: 350

Merit: 22

0.27 online - major update

Code:

Fix a possible crash when mining IPBC with assembly "generic"
New fork: XTL for Stellite, that's --variation 7
Fixed Loki wallet detection
Updated the documentation and examples in the .zip

JCE-Miner

member

Activity: 350

Merit: 22

i get it, but how non-assignment can help ?
If i've 8 cores for example, isn't it better to assign 4 cores to get the turbo, rather than having 8 cores running anywhere at 50% each, with no turbo ?

edit: a user kindly reported me the speed increase on his Ryzen 1700 with JCE 0.26 compared with JCE 0.25 +5% on CN-Heavy, with optimal config

Code:

"cpu_threads_conf" :  
[  
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 1, "use_cache" : true },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 2, "use_cache" : false },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 3, "use_cache" : false },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 5, "use_cache" : true },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 6, "use_cache" : false },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 7, "use_cache" : false },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 9, "use_cache" : true },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 10, "use_cache" : false },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 11, "use_cache" : false },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 13, "use_cache" : true },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 14, "use_cache" : false },    
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 15, "use_cache" : false },    
]

UnclWish

sr. member

Activity: 1484

Merit: 253

Quote from: JCE-Miner on May 08, 2018, 06:56:49 AM

i tried, but always got a few h/s less than with affinity set. The autoconfig always assign one CPU.
Did you find a case where no assignment really gave better result ?

No assignment can rise hashrate only with enabled Turbo technology and when less or equal half of cores used for mining. If more than half cores used Turbo didn't work...

JCE-Miner

member

Activity: 350

Merit: 22

i tried, but always got a few h/s less than with affinity set. The autoconfig always assign one CPU.
Did you find a case where no assignment really gave better result ?

UnclWish

sr. member

Activity: 1484

Merit: 253

Did you tried to not set affinity for threads? But it can affect only when not all cores used...

JCE-Miner

member

Activity: 350

Merit: 22

sure, let me finish 0.27 with XTL fork (i'm writting assembly right now) and i give all my config i use for benchmark.
Again, i've no big Intel CPU so it may be possible JCE underperform against xmrig on some i7 or Xeon. Assembly requires fine adjustement, that's the game.
I could cheat by stealing xmrig code and compile it, but no. The perf difference of JCE against xmrig/stak, even when negative, is a proof i really write my own code

Note that on Ryzen, 0.26 with autoconfig --auto should already give the best config i found on Ryzen.
I reach 1850 on Turtle and, surprisingly, only 1755 on IPBC while there's only two extra ASM instructions between both. Undecided

best config for Ryzen 1600 (should apply to similar Ryzen too like 1600X)

CN-Light, Turtle, IPBC

Code:

"cpu_threads_conf" : 
[ 
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 0, "use_cache" : true, "multi_hash":2 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 1, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 2, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 3, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 4, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 5, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 6, "use_cache" : true, "multi_hash":2 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 7, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 8, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 9, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" :10, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" :11, "use_cache" : true, "multi_hash":1 },
]

CN-Classic, V7, Stellite

Code:

"cpu_threads_conf" : 
[ 
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 0, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 1, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 2, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 4, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 6, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 7, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 8, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" :10, "use_cache" : true, "multi_hash":1 },
]

CN-Heavy

Code:

"cpu_threads_conf" : 
[ 
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 0, "use_cache" : false, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 1, "use_cache" : false, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 2, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 4, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 6, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" : 8, "use_cache" : true, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" :10, "use_cache" : false, "multi_hash":1 },
     { "cpu_architecture" : "ryzen", "affine_to_cpu" :11, "use_cache" : false, "multi_hash":1 },
]

Not the "use_cache" : false that gives a boost for CN-Heavy, that's a new feature of 0.26+
On previous versions, no-cache was only for super-super-low power mode (one core @10% speed). Now it's also a speed enhancer.

s0ftcorn

newbie

Activity: 70

Merit: 0

Quote from: JCE-Miner on May 08, 2018, 02:44:06 AM

hi,

multi-hash: it's relevant only if you mine a CN-Light coin (turtle, aeon, ipbc...) or you want to use only one core of your cpu.

to use your cpu at max speed on CN-v7 i never found it better than simple hash.

the cpu i tested on are only core2 and Ryzen, i've a buldozzer but need repairs, not working yet. and no recent Intel at all Sad

i also observed speed better on my ryzen with 14M cache used with doublehash than the whole 16M with triplehash. but used alone, triple is faster. i may needto try more combination or register allocation, but did'nt have time yet.

are you willing to share your config for your ryzen?

JCE-Miner

member

Activity: 350

Merit: 22

hi,

multi-hash: it's relevant only if you mine a CN-Light coin (turtle, aeon, ipbc...) or you want to use only one core of your cpu.

to use your cpu at max speed on CN-v7 i never found it better than simple hash.

the cpu i tested on are only core2 and Ryzen, i've a buldozzer but need repairs, not working yet. and no recent Intel at all Sad

i also observed speed better on my ryzen with 14M cache used with doublehash than the whole 16M with triplehash. but used alone, triple is faster. i may needto try more combination or register allocation, but did'nt have time yet.

UnclWish

sr. member

Activity: 1484

Merit: 253

Quote from: JCE-Miner on May 07, 2018, 05:36:59 PM

0.26 online - major update

Code:

Triple, Quad, Penta and Hexa hash
Faster on IPBC and Cryptolight (tested on my Ryzen, don't know how it behaves on Intel)
Faster on CN-Heavy (ditto)
New coin: Gadcoin
New coin: Loki

Next tasks:
* Update the documentation
* Add Stellite variation (that will be --variation 7)

edit: bug reported about LOKI detection, workaround params: --any --variation 5
hotfix release to come soon

Thanks! Wee will see...

Did you compared triple, Quad, etc. speeds to XMRig?

Strange situation with my FX8320 (Vishera). It have 8Mb L3 cache shared between all cores, and 8Mb L2 cache shered 4x2Mb, each 2Mb per 2 cores.
So if I use 2 cores it can use 8Mb L3 and 2Mb L2 cache = 10Mb.
But if I setup miner to 2 double threads (8Mb need) speed is faster 10-15 h/s than if I set 3+2 multihash threads (10Mb need).
The same situation with any variants. Any variant with 8Mb cache is faster than 10Mb config... Not only 2 threads variants...

Did you have Vishera to check? Maybe miner didn't use L2 cache? Or something else... On XMRig the same situation...

s0ftcorn

newbie

Activity: 70

Merit: 0

since --auto does set multi hash to "no" on my ryzen 2700X, i assume there is no point in tinkering with it?

JCE-Miner

member

Activity: 350

Merit: 22

0.26 online - major update

Code:

Triple, Quad, Penta and Hexa hash
Faster on IPBC and Cryptolight (tested on my Ryzen, don't know how it behaves on Intel)
Faster on CN-Heavy (ditto)
New coin: Gadcoin
New coin: Loki

Next tasks:
* Update the documentation
* Add Stellite variation (that will be --variation 7)

edit: bug reported about LOKI detection, workaround params: --any --variation 5
hotfix release to come soon

JCE-Miner

member

Activity: 350

Merit: 22

proxy : i'm not fan of relaxing the proxy limitation, it would allow jce be embedded into botnets. i rather plan to allow several JCEs to work together, network linked. but not dev priority for now.

speed : i need your full xmrig config to understand the difference, even if i guess that's thanks to the triplehash that xmrig supports and not jce yet. i'll take a look at that cpu and provide a good jce config for it.

edit : that xeon should be at max speed with eight threads, 1M each. i'd be interrested by first lines of logs of both jce and xmrig.

i'm finalizing 0.26 which should give a boost on CN-Heavy

edit: i've benchmarked my Ryzen 1600 stock on Turtle, i pull 1850 H/s out of it on jce 0.26.
And 260 H/s on CN-Heavy. i couldn't reach those numbers on xmrig 2.6.2

However i've no big Intel CPU so cannot finetune for them Sad

arvonceda

newbie

Activity: 66

Merit: 0

i mine cryptnigth-lite (turtlecoin)
JCE 025 = 601 h/s
xmrig-2.6.0-beta3-gcc-win64 = 800h/s

mining with Procesor XEON E3-1245v2
what wrong ?

malhuis

newbie

Activity: 2

Merit: 0

Hi,

Have to say, great work with the miner.

See you busy with the new release, so just a quick question: any possibility to make this work with Xmr Proxy ? Most the new pools have starting difficulty that makes it senseless to mine with older to average CPU's, so with Xmr Proxy it really is a nice option to have all the lower hashrate CPU's mine to the proxy and let the proxy then connect to the pool as one worker and the difficulty then gets distributed in lower values to the CPU's on the miner's side. Even with the increased hashrate from your miner, still is the only way to make use of average/older CPU's on pools these days.

Thanks again for your work!!

JCE-Miner

member

Activity: 350

Merit: 22

Hi !

My rant of the day: avoid quoting my whole documentation, it makes the topic unreadable. And i update that doc, but your quote won't be updated...

Here are your answers:

* You can mine any Cryptonight-family coin, just add the --any parameter to unlock the coin list.
* Editing the example .bat is the simplest way to do, you're on the good way. You must replace the example wallet by yours, it can be a paper wallet or an online wallet, no problem.

but... UNIT is not a Cryptonight coin, it's a sha256 coin, so no you cannot mine it with JCE. However you can mine XMR for example.

To get you XMR wallet:
* Go to https://cn-wallet-generator.hashvault.pro/monero-wallet-generator.html
* Click Generate (yellow button)
* Then Note EVERYTHING on this page (needed to get your coins later)
* The Public Address that start by "4" is your XMR wallet.
* Replace my example wallet by this one in the .bat and you can start to mine.

finitel

newbie

Activity: 1

Merit: 0

JCE-Miner

member

Activity: 350

Merit: 22

no problem

Yeah, forget xmrig and stak with their 500-params config file. JCE is closer to Claymore CPU for config: all is a few command line parameters.

Config file exists, but is optional and for advanced CPU configuration only.

Good news for CN-Heavy miner : a user kindly reported that the obscure no-cache mode i advise not to use may give a boost when mining Heavy. My Ryzen 1600 jumps from 252 to 257, a welcome +2%. The general rule is than any unused physical core may give +1% perf.
That's for 64-bits AES only, but that's where JCE is the weakest compared to xmrig, so the boost is welcome.
Still testing, release soon. The binaries are now getting really huge... Cry

fishdogduck

newbie

Activity: 22

Merit: 3

Dumb question.... but how do I configure the pool/wallet/coin settings? I can't find the config file and also can't input anything except h r and q

Edit: Found it.... For those as dumb as me, edit the .bat file Cheesy

Topic: [XMR] JCE Miner Cryptonight/forks, now with GPU! - page 104. (Read 90815 times)