[ANN]: cpuminer-opt v3.8.8.1, open source optimized multi-algo CPU miner - page 107.

oldDIN

member

Activity: 85

Merit: 10

Again, it seems you got me all wrong, English is not my

W7-64-pro

MAC- AMD Intel
Git miner exit after 50 submits.
Git + code change in cpu-miner.c - work

Code:

if (rpc_pass && rpc_user)
     opt_stratum_stats &= (strstr(rpc_pass, "stats") != NULL) || (strcmp(rpc_user, "benchmark") == 0);

I collected everything under the window, not under linux (( Excuse me, I did not say immediately

felixbrucker

hero member

Activity: 700

Merit: 500

Quote from: joblo on February 02, 2017, 01:13:46 PM

A status summary just to make sure I keep my head straight.
Scrutinize my assertions to confirm you agree.

1. x11evo problem has been identified as an AMD only issue, still unresolved.

2. timetravel problem was believed to be fixed with 3.5.5-pre but felixbrucker
reports it fails on AMD/Win and works on Intel/Linux. oldDIN reports 3.5.5-pre
fails on Intel/Linux. This contradicts my results where it works on Intel/Linux.

The current discrepency is with timetravel using 3.5.5-pre on Intel/Linux.
Need to doublecheck those results as requested in previous post.

1.1 x11evo is reported broken in amd on windows, intel on windows
1.2 x11evo is reported working in intel on linux

2.1 timetravel exits on 3.5.5-pre and below (3.5.3 and 3.5.4 tested) on amd on windows
2.2 timetravel does not exit on intel on windows and intel on linux

joblo

legendary

Activity: 1470

Merit: 1114

A status summary just to make sure I keep my head straight.
Scrutinize my assertions to confirm you agree.

1. x11evo problem has been identified as an AMD only issue, still unresolved.

2. timetravel problem was believed to be fixed with 3.5.5-pre but felixbrucker
reports it fails on AMD/Win and works on Intel/Linux. oldDIN reports 3.5.5-pre
fails on Intel/Linux. This contradicts my results where it works on Intel/Linux.

The current discrepency is with timetravel using 3.5.5-pre on Intel/Linux.
Need to doublecheck those results as requested in previous post.

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: felixbrucker on February 02, 2017, 12:30:57 PM

i could be wrong, but i believe his test which was working was with the rpc user/pass stats/benchmark lines added to not enable the benchmark submit

Yes that's an invalid test because it avoids the problem by not running the stats. The stats need to run
at 50 submits and not exit for the test to be successful.

felixbrucker

hero member

Activity: 700

Merit: 500

i could be wrong, but i believe his test which was working was with the rpc user/pass stats/benchmark lines added to not enable the benchmark submit

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: felixbrucker on February 02, 2017, 12:12:41 PM

Quote from: joblo on February 02, 2017, 12:05:38 PM

Quote from: felixbrucker on February 02, 2017, 10:59:10 AM

edit2: also happening with sse2 bin

edit3: seems to be a windows issue, running fine on my linux server (xeon e3-1265lv2), not working on 3 amd/intel windows boxes

Awsome data. All failures are Windows on AMD CPUs, everything else works.
Now there's a delta. I can put my thinking cap back on.

I think this also relates to the other issue as it was also AMD related. In that case a macro was messing up timetravel
on AMD but it worked ok on Intel. There may be something similar going on with x11evo.

well its also broken for intel on windows for me (i5-3330) so it cant be specific to amd (x11evo)

i just tested the bin you sent me, it also exits on 50 accepted shares, is that expected?

No. OldDIN reported ok. Maybe I sent you the wrong code, Let me check.

Edit: I looked at the exact source I used to build your test load and it is correct.

I removed the macro that caused me problems with x11evo.

It seems if the bench stats collection runs on an AMD CPU mining timetravel the miner will exit.

The easy workaround is to disable stats, which I intend to do by default in the next release.

Finding root cause is impossible until I can reproduce it.

The swap macro failed for me on x11evo but worked on timetravel. I still don't understand why but it
led me to believe there is a problem using that macro in some situations, particularly on AMD.
I therefore removed the macro from timetravel in 3.5.5-pre. Now oldDIN reports testintg ok but you still report
miner exit after 50 submits.

oldDIN can you confirm cpu-miner.c is the same in 3.5.5-pre as 3.5.4?

Code:

diff cpuminer-opt-3.5.4/cpu-miner.c cpuminer-opt-3.5.5-pre/cpu-miner.c

In 3.5.5-pre opt_stratum_stats should be initialized to true and never changed, the same as it was
in 3.5.4. The only significant change is in timetravel.

Here's the diff from git...

https://github.com/JayDDee/cpuminer-opt/commit/977415a25bc13cf4fd871f50cf82a75eb8a5cde8

felixbrucker

hero member

Activity: 700

Merit: 500

Quote from: joblo on February 02, 2017, 12:05:38 PM

Quote from: felixbrucker on February 02, 2017, 10:59:10 AM

edit2: also happening with sse2 bin

edit3: seems to be a windows issue, running fine on my linux server (xeon e3-1265lv2), not working on 3 amd/intel windows boxes

Awsome data. All failures are Windows on AMD CPUs, everything else works.
Now there's a delta. I can put my thinking cap back on.

I think this also relates to the other issue as it was also AMD related. In that case a macro was messing up timetravel
on AMD but it worked ok on Intel. There may be something similar going on with x11evo.

well its also broken for intel on windows for me (i5-3330) so it cant be specific to amd (x11evo)

i just tested the bin you sent me, it also exits on 50 accepted shares, is that expected?

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: felixbrucker on February 02, 2017, 10:59:10 AM

edit2: also happening with sse2 bin

edit3: seems to be a windows issue, running fine on my linux server (xeon e3-1265lv2), not working on 3 amd/intel windows boxes

Awsome data. All failures are Windows on AMD CPUs, everything else works.
Now there's a delta. I can put my thinking cap back on.

Edit: ignore that I misread slightly.

You report failure on Windows Intel and AMD, working on Intel Linux.

OldDIN reports Linux fail on AMD, ok on Intel.

It looks more like all failures are on AMD. Your testing doesn't have Linux/AMD but oldDIN
filled that void.

Clearly an AMD issue and likely related to the timetravel issue where it failed only on AMD.
Now I got a lead to follow.

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: oldDIN on February 02, 2017, 11:46:07 AM

-MAC-
3.5.5-pre git corei7-avx - 50 off
3.5.5-pre with a change git code - works

-x11evo-
Rebuild 3.5.5-pre git
Intel - works
AMD - rejections after the first cache

Thanks. MAC looks good I will proceed with the full fix in the next release.

Now about x11evo, I'm sorry I doubted your testing but as you can see I had to be absolutely sure.
I need to find some way of reproducing it and as the previous post said, I'm stuck.

You can continue to use 3.5.5-pre for MAC until the next release, for x11evo use
3.5.4 if you can (it's faster), or 3.5.1 if not until the problem is solved.

oldDIN

member

Activity: 85

Merit: 10

-MAC-
3.5.5-pre git corei7-avx - 50 off
3.5.5-pre with a change git code - works

-x11evo-
Rebuild 3.5.5-pre git
Intel - works
AMD - rejections after the first cache

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: felixbrucker on February 02, 2017, 10:59:10 AM

well if no linux amd user chimes in a bin would be great

arch wise im using the corei7-avx, also for the last x11evo tests, should i try with sse2 bin (x11evo)?

~~the x11evo problem should be reproducable, or am i wrong?~~

edit: just saw your x11evo output, very strange, same capabilities

Yes very strange. You used bins I made so it's the same compiler version and I actually have the right
CPU to reproduce the problem.

- 3 users with same arch, 2 fail , one ok
- one fail on Linux, other on Win
- fail and ok in same pool

I can find no delta between the success and failures, I'm stuck.

While I'm stuck I'll build 3.5.5-pre-corei7-avx for you. Will PM link.

felixbrucker

hero member

Activity: 700

Merit: 500

well if no linux amd user chimes in a bin would be great

arch wise im using the corei7-avx, also for the last x11evo tests, should i try with sse2 bin (x11evo)?

~~the x11evo problem should be reproducable, or am i wrong?~~

edit: just saw your x11evo output, very strange, same capabilities

edit2: also happening with sse2 bin

edit3: seems to be a windows issue, running fine on my linux server (xeon e3-1265lv2), not working on 3 amd/intel windows boxes

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: felixbrucker on February 02, 2017, 10:48:47 AM

Quote from: joblo on February 02, 2017, 10:26:15 AM

Quote from: felixbrucker on February 02, 2017, 10:10:29 AM

i can test it if there is a full cmdline

i have somewhat lost track of all the different problems and version that should fix/be tested against, so a brief summary with the cmd to test would be great as well Cheesy

You can test x11evo with 3.5.4 at zpool with any btc address.

./cpuminer -a x11evo -o stratum+tcp://x11evo.mine.zpool.ca:3553 -u [btc addr] -p test

Testing timetravel needs the change in 3.5.5-pre (latest git). It is only for the purposes of testing timetravel on AMD CPUs
and is only available as source code.

If testing drags on I can build a bin of 3.5.5-pre specifically for you, just let me know which arch you need.

just tested x11evo on zpool, all rejects (at a very high speed)

first it takes about 30 sec and then there are 800+ instantly rejected shares and a lot of output ( i have configured my stdout/stderr log file parsing to be rotated every 50mb file size and it somehow crashed that Cheesy

)

starting with -D didnt seem to reveal anything special, i can upload the 38MB logfile of the first wave of rejects with -D to pastebin if neccessary

i have not used x11evo before, speed was shown as 10MH/s (not sure if thats normal)

also this test was on intel cpu, nothing amd related

What arch? Hash rate should be similar to my post above.

felixbrucker

hero member

Activity: 700

Merit: 500

Quote from: joblo on February 02, 2017, 10:26:15 AM

Quote from: felixbrucker on February 02, 2017, 10:10:29 AM

i can test it if there is a full cmdline

i have somewhat lost track of all the different problems and version that should fix/be tested against, so a brief summary with the cmd to test would be great as well Cheesy

You can test x11evo with 3.5.4 at zpool with any btc address.

./cpuminer -a x11evo -o stratum+tcp://x11evo.mine.zpool.ca:3553 -u [btc addr] -p test

Testing timetravel needs the change in 3.5.5-pre (latest git). It is only for the purposes of testing timetravel on AMD CPUs
and is only available as source code.

If testing drags on I can build a bin of 3.5.5-pre specifically for you, just let me know which arch you need.

just tested x11evo on zpool, all rejects (at a very high speed)

first it takes about 30 sec and then there are 800+ instantly rejected shares and a lot of output ( i have configured my stdout/stderr log file parsing to be rotated every 50mb file size and it somehow crashed that Cheesy

)

starting with -D didnt seem to reveal anything special, i can upload the 38MB logfile of the first wave of rejects with -D to pastebin if neccessary

i have not used x11evo before, speed was shown as 10MH/s (not sure if thats normal)

also this test was on intel cpu, nothing amd related

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: felixbrucker on February 02, 2017, 10:10:29 AM

i can test it if there is a full cmdline

i have somewhat lost track of all the different problems and version that should fix/be tested against, so a brief summary with the cmd to test would be great as well Cheesy

You can test x11evo with 3.5.4 at zpool with any btc address.

./cpuminer -a x11evo -o stratum+tcp://x11evo.mine.zpool.ca:3553 -u [btc addr] -p test

Testing timetravel needs the change in 3.5.5-pre (latest git). It is only for the purposes of testing timetravel on AMD CPUs
and is only available as source code.

If testing drags on I can build a bin of 3.5.5-pre specifically for you, just let me know which arch you need.

felixbrucker

hero member

Activity: 700

Merit: 500

i can test it if there is a full cmdline

i have somewhat lost track of all the different problems and version that should fix/be tested against, so a brief summary with the cmd to test would be great as well Cheesy

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: oldDIN on February 02, 2017, 06:19:27 AM

joblo
-"x11evo"
Checked on version 3.5.1 - it works on amd and intel well
version 3.5.5 - all rejects

-"MAC"
Yes, v 3.5.5 with code

Code:

if (rpc_pass && rpc_user)
opt_stratum_stats &= (strstr(rpc_pass, "stats") != NULL) || (strcmp(rpc_user, "benchmark") == 0);

works good

remind-fx-6300, i3-2120

Thanks for testing MAC but can you repeat the test with v3.5.5-pre exatcly as it exists in GIT? Unfotunately using the
two line above made your test invalid. v3.5.5-pre should be used only for testing timetravel on AMD.

As far as x11evo is concerned it works fine for me in 3.5.4 and I made no changes since so you should use 3.5.4.
Since x11evo works for me it may be a cpu architecture issue. Please show the miner's startup messages showing
which features are available to see if I can reproduce the problem. Maybe Felix can help out with different AMD rchitectures
now that binaries are available. Here is the output from my i5-2400, it is the same architecture as your i3-2130.

Code:

coin@sys16 ~/miners/cpu/cpuminer-opt/cpuminer-opt-3.5.4 $ ./cpuminer -a x11evo -o [...]

   ********** cpuminer-opt 3.5.4 ***********
   A CPU miner with multi algo support and optimized for CPUs
   with AES_NI and AVX extensions.
   BTC donation address: 12tdvfF7KmAsihBXQXynT6E6th2c2pByTT
   Forked from TPruvot's cpuminer-multi with credits
   to Lucas Jones, elmad, palmd, djm34, pooler, ig0tik3d,
   Wolf0, Jeff Garzik and Optiminer.

CPU: Intel(R) Core(TM) i5-2400 CPU @ 3.10GHz
CPU features: SSE2 AES AVX
SW built on Feb 2 2017 with GCC 4.8.4
SW features: SSE2 AES AVX
Algo features: SSE2 AES AVX AVX2
Start mining with SSE2 AES AVX

[2017-02-02 09:48:04] Starting Stratum on stratum+tcp://x11evo.mine.zpool.ca:3553
[2017-02-02 09:48:04] 4 miner threads started, using 'x11evo' algorithm.
[2017-02-02 09:48:05] x11evo block 357418, diff 0.741
[2017-02-02 09:48:12] x11evo block 357419, diff 0.918
[2017-02-02 09:48:12] CPU #0: 454.97 kH, 66.37 kH/s
[2017-02-02 09:48:12] CPU #2: 456.36 kH, 66.61 kH/s
[2017-02-02 09:48:12] CPU #3: 454.54 kH, 66.36 kH/s
[2017-02-02 09:48:12] CPU #1: 456.28 kH, 66.62 kH/s
[2017-02-02 09:49:05] CPU #1: 3575.55 kH, 67.11 kH/s
[2017-02-02 09:49:05] CPU #0: 3564.84 kH, 66.91 kH/s
[2017-02-02 09:49:05] CPU #3: 3560.21 kH, 66.82 kH/s
[2017-02-02 09:49:05] CPU #2: 3577.08 kH, 67.14 kH/s
[2017-02-02 09:49:08] x11evo block 357420, diff 0.741
[2017-02-02 09:49:08] CPU #0: 156.91 kH, 66.90 kH/s
[2017-02-02 09:49:08] CPU #3: 156.89 kH, 66.89 kH/s
[2017-02-02 09:49:08] CPU #1: 157.40 kH, 67.10 kH/s
[2017-02-02 09:49:08] CPU #2: 157.36 kH, 67.09 kH/s
[2017-02-02 09:49:35] CPU #1: 1819.90 kH, 67.11 kH/s
[2017-02-02 09:49:35] Accepted 1/1 (100%), 2291.07 kH, 267.99 kH/s, 68C
[2017-02-02 09:49:38] x11evo block 357421, diff 0.913
[2017-02-02 09:49:38] CPU #3: 2047.45 kH, 66.81 kH/s

Edit: also please include whether you compile yourself or use a windows binary and which
architecture you use.

oldDIN

member

Activity: 85

Merit: 10

joblo
-"x11evo"
Checked on version 3.5.1 - it works on amd and intel well
version 3.5.5 - all rejects

-"MAC"
Yes, v 3.5.5 with code

Code:

if (rpc_pass && rpc_user)
opt_stratum_stats &= (strstr(rpc_pass, "stats") != NULL) || (strcmp(rpc_user, "benchmark") == 0);

works good

remind-fx-6300, i3-2120

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: Epsylon3 on February 01, 2017, 11:59:24 PM

the default (to false) is here to highlight the fact its disabled by default... also -p is optional

http://yiimp.ccminer.org/site/benchmarks

OK I get it now. The password is how the user enables stats submission. I had the default set to true
which appears to be why you were getting so many stats reports. Fortunately this allowed the discovery
of a more serious issue.

I'll add your code as is and change the default to false in the next release.

Epsylon3

legendary

Activity: 1484

Merit: 1082

ccminer/cpuminer developer

the default (to false) is here to highlight the fact its disabled by default... also -p is optional

http://yiimp.ccminer.org/site/benchmarks

Topic: [ANN]: cpuminer-opt v3.8.8.1, open source optimized multi-algo CPU miner - page 107. (Read 444067 times)