Topic: [XMR] JCE Miner Cryptonight/forks, now with GPU! - page 116. (Read 90815 times)

Hi

Just finish testing dualmining as you recommended, the good news, shares are accepted, the bad news hash decreased to 128 and system freezes after a couple minutes.

Gonna test 0.21 right now.

Thanks.

Version 0.21


Now JCE is as fast on CN-V7 than others on original CN, at least on my Ryzen : 502h/s

I just noticed, thanks to your post, that i use the term Huge instead of Large. To be fixed in doc and binaries.
I'm testing right now the 64-bits cryptonight heavy non-aes. I can lose against stak on some recent Intel cpu i haven't optimized for, but on non-aes.

edit: on my Xeon (autoconfig, core2 base, no aes, 64 bits)

Stak 2.4.2 (auto config, all Large Page enabled)

JCE is just... 50% faster on non-aes.
On Haswell... yeah i don't own a modern Intel cpu, only amd, hence why there's a Ryzen assembly, but no Haswell. Very possible Stak is still better on Haswell. I guarantee jce is faster on non-aes, 32-bits and on ryzen. Cannot be sure for others 64-bits AES. And even on Ryzen, i'm just 1% faster.

Good diagnostic, i found the bug, and yes, unrelated to the K8-K10 bug. That's a thread sync problem. So bad it can freeze the miner. I'm fixing, should be done rapidly.

Thanks.

Lost in translation, by windows italiano and me

How many cores and cache do you have ? It may be the doublehash that gives big boost to stak, it's good on high cache cpu, and xeon are.

seems you're talking about Pagefile. Huge pages are in memory, not related to Pagefile (aka swap)

If you run Windows 10 pro, run jce as admin once, reboot and it should be enabled. Otherwise look at this guide
https://technet.microsoft.com/en-us/library/ms190730(v=sql.105).aspx  

Jce tries to connect 5 times before it gives up. You can loop in a .bat if you want to try forever.

I note the dual thread bug, not as bad because it's rarely used, but needs to fix.

Ho, and the Linux version : Planned but not ready yet. All miners are detected as virus or Trojan, but they aren't, not jce nor any other.

Jce miner can't use Huge pages even if present, recognized by windows
Can be that doesn't like the fact that huge pages are on drive d: instead of c: ?

Are there builds for systems other than windows? ... For my windows work machine, they're very strict on apps (so this is reporting as malware/virus which I don't believe) ... So i want to try and run it in a linux virtual machine

Given that, I could not inspect the zip file ... but is there a way to provide a linux executable?

Addendum: the bug is fixed mining Aeon, but when mining TurtleCoin with dual-mining-feature, I still get rejected shares. Not using dual-mining, everything is fine even with TurtleCoin. So there must be a small bug somewhere

Edit: this time it has nothing to do with the K10. Lynnfield produces the same rejected shares as soon as I use dual-mining with TurtleCoin

What are your memory clock?

Ryzen likes memory at 2800+ and lowering the main timings (for 15-15-15-36 1T as example)

Why miner stops to connect?

12:30:50 | Pool sends a new Job.
12:31:43 | Pool sends a new Job.
12:32:32 | Pool sends a new Job.
12:33:05 | Pool sends a new Job.
12:33:12 | Nicehash session timeout, instant-reconnect.
12:33:12 | Connecting with SSL to mining pool cryptonightv7.eu.nicehash.com:33363 ...
12:33:12 | Connected to pool. Now logging in...
12:33:12 | Connection failed: Socket receive error: socket explicitly closed by the pool
12:33:12 | Connection interrupted, waiting 5s then retry.
12:33:17 | Connecting with SSL to mining pool cryptonightv7.eu.nicehash.com:33363 ...
12:33:18 | Connected to pool. Now logging in...
12:33:18 | Connection failed: Socket receive error: socket explicitly closed by the pool
12:33:18 | Connection interrupted, waiting 5s then retry.
12:33:22 | Connecting with SSL to mining pool cryptonightv7.eu.nicehash.com:33363 ...
12:33:23 | Connected to pool. Now logging in...
12:33:23 | Connection failed: Socket receive error: socket explicitly closed by the pool
12:33:23 | Connection interrupted, waiting 5s then retry.
12:33:28 | Connecting with SSL to mining pool cryptonightv7.eu.nicehash.com:33363 ...
12:34:44 | Hashrate Thread 0: 0.00 h/s
12:34:44 | Hashrate Thread 1: 0.00 h/s
12:34:44 | Hashrate Thread 2: 0.00 h/s
12:34:44 | Total: 0.00 h/s
12:39:38 | Hashrate Thread 0: 0.00 h/s
12:39:38 | Hashrate Thread 1: 0.00 h/s
12:39:38 | Hashrate Thread 2: 0.00 h/s
12:39:38 | Total: 0.00 h/s
12:44:31 | Hashrate Thread 0: 0.00 h/s
12:44:31 | Hashrate Thread 1: 0.00 h/s
12:44:31 | Hashrate Thread 2: 0.00 h/s
12:44:31 | Total: 0.00 h/s
12:49:25 | Hashrate Thread 0: 0.00 h/s
12:49:25 | Hashrate Thread 1: 0.00 h/s
12:49:25 | Hashrate Thread 2: 0.00 h/s
12:49:25 | Total: 0.00 h/s
12:54:18 | Hashrate Thread 0: 0.00 h/s
12:54:18 | Hashrate Thread 1: 0.00 h/s
12:54:18 | Hashrate Thread 2: 0.00 h/s
12:54:18 | Total: 0.00 h/s
12:59:12 | Hashrate Thread 0: 0.00 h/s
12:59:12 | Hashrate Thread 1: 0.00 h/s
12:59:12 | Hashrate Thread 2: 0.00 h/s
12:59:12 | Total: 0.00 h/s
13:04:05 | Hashrate Thread 0: 0.00 h/s
13:04:05 | Hashrate Thread 1: 0.00 h/s
13:04:05 | Hashrate Thread 2: 0.00 h/s
13:04:05 | Total: 0.00 h/s
13:08:59 | Hashrate Thread 0: 0.00 h/s
13:08:59 | Hashrate Thread 1: 0.00 h/s
13:08:59 | Hashrate Thread 2: 0.00 h/s
13:08:59 | Total: 0.00 h/s
13:13:53 | Hashrate Thread 0: 0.00 h/s
13:13:53 | Hashrate Thread 1: 0.00 h/s
13:13:53 | Hashrate Thread 2: 0.00 h/s
13:13:53 | Total: 0.00 h/s
13:18:46 | Hashrate Thread 0: 0.00 h/s
13:18:46 | Hashrate Thread 1: 0.00 h/s
13:18:46 | Hashrate Thread 2: 0.00 h/s
13:18:46 | Total: 0.00 h/s
13:22:13 | Stop signal received, Quitting...

Closing and launching again somtime fix connection... sometime not...

Was somebody already able to try it with a ryzen 5 2400G ?

i tried this on xeon haswell i got 155 hash for cryptonight heavy. xmrstak got 165 hash

I document minimum to be Seven, but I got reports 2008 Server and Vista work too.
XP and 2003 don't support Huge Pages, so they are not viable mining os

Yeeeees!  Great job! 

In the matter of OS Windows Vista / Server 2008 is min system requirement, isn't it? Any chance to get the code running on Windows XP / Server 2003? I have some of the very old boxes here with these operating systems.

Sorry i should have been more explicit : The idea was to set 6 threads on cpu 0-5 and make thread 5 (the last) dualmine with thread 4 (the one before). Period. And not use dualmine three times, which of course kill the perf.

But... Invalid hashes on dualmine on your k10... Yeah I made 0.20 so fast that I didn't even check the dualmine version of buggy k8-k10 asm, stupid. I'll go for another 0.21 that invalidate previous, remove all zip from github... Again.  

Will release asap. If you want a new unlisted coin by the way, ask here  

Config 1:


Start 1




Start 2




Config 2:


Start 1




EDIT

With a little OC to keep same CPU voltage (config1 and start1)

Good, and thanks for the screen
So that cpu has six cores and 3M of L2 + 6M of L3. Don't you have more perf with 6 threads ?
Or with my dual-thread feature, which is exactly made for that kind of cpu (no aes, lots of cores, low cache) ?
The config example in the zip would worth a look

32Bit it is fine, hashes about 110H/s with 5 threads.


Testing it now. Thanks.

Edit



 

K8-K10 bug fixed

please update, unless you use AES or 32-bits, in such case JCE 0.19 and 0.20 are the exact same.
Documentation on first post now explains the attrib trick.

My ryzen 1600 (6 cores, 16M cache, 8 mining threads, 3.2GHz, direct hardware (no VM)) peaks at 507, your performance sounds consistent, but i don't know how negative is the use of a VM.

Good news : I've setup an antique Athlon64 on Windows 7 x64 and I can reproduce the K8-K10 bug. Fixing, should be done in a few hours.

It invalidates version 0.17 to 0.19, switch to 0.20 once possible.
Then i update the documentation and remove the zip from github.

Sorry men for the bug