Gateless Gate Sharp 1.3.8: 30Mh/s (Ethash) on RX 480!

[UnclWish]

Hi, zawawa! How are you?

It's new algo - PHI1612. Can you add him to GGS?

[Littlegrand]

Quote from: Littlegrand on May 05, 2018, 03:40:39 PM


Quote

Quote from: Littlegrand on Today at 08:51:51 AM
Quote
It is all right. The log shows cryptonight heave loaded, the share submitted to the correct url, just rejected wrong algo...
Correct url didn't means that you select correct url for cryptonight heavy algo.
Most pools have several ports for different algos. If you chose port for cryptonightv7 but in miner you select CN heavy algo, miner will connect to pool via seleted port, and pool allows to connect, but miner will finds shares for heavy and send them on pool to port from wich pool wait shares for cryptonightV7, due to port.
Check port number for your pool. Is you right pointed it for CN heavy.

Yes it is correct port(though it does not show port in log).  The Haven address in log is correct as well.

It normal to say Switching to Monero 2 in log?

Any ideas?

Thanks

Please give me information about your pool, the coin you are mining, and your settings, and I will get to it right away

Haven

haven.ingest.cryptoknight.cc 5531  

Didnt optimize default settings:
Threads 1
Intensity 4 tried to 20
Work size 4
Strided index 1
Memory chunk 2

My laptop has NVIDIA geforce 740m.

Also tried it same settings on my 6 card AMD rig with 3 RX580 and 3 RX570, same thing

Log from laptop(same on rig)

2018-05-05 12:34:10.7939 [16] Running garbage collection...
2018-05-05 12:34:10.7939 [16] Memory used before collection: 14MB
2018-05-05 12:34:10.8900 [16] Memory used before collection: 12MB
2018-05-05 12:34:13.1940 [18] API Listener started.
2018-05-05 12:38:33.9272 [1] Launching miner(s) for Custom Pool 0...
2018-05-05 12:38:35.5019 [20] Miner thread for Device #0 started.
2018-05-05 12:38:35.5019 [20] NiceHash mode is off.
2018-05-05 12:38:35.6611 [20] Loaded Kernels\cryptonight_heavy.cl for Device #0.
2018-05-05 12:38:47.1088 [20] Built cryptonight_heavy program for Device #0.
2018-05-05 12:38:47.1088 [20] Build options:  -DSTRIDED_INDEX=1 -DMEMORY_CHUNK_SIZE=2  -IKernels -DWORKSIZE=4
2018-05-05 12:38:47.1695 [20] Switching to Monero Version 2
2018-05-05 12:38:57.8436 [20] Device #0 (CryptoNight): 30.11 h/s
2018-05-05 12:39:08.4715 [20] Device #0 (CryptoNight): 30.12 h/s
2018-05-05 12:39:08.8778 [12] Received new job: 800226972178483
2018-05-05 12:39:10.8926 [16] Running garbage collection...
2018-05-05 12:39:10.8926 [16] Memory used before collection: 16MB
2018-05-05 12:39:10.9617 [16] Memory used before collection: 15MB
2018-05-05 12:39:19.1095 [20] Device #0 submitted Share #0 to haven.ingest.cryptoknight.cc as (my have address taken out)
2018-05-05 12:39:19.1100 [20] Device #0 submitted a share to haven.ingest.cryptoknight.cc as  (my have address taken out)
2018-05-05 12:39:19.1100 [20] Device #0 (CryptoNight): 29.92 h/s
2018-05-05 12:39:19.7139 [12] Share #0 rejected: Wrong algo, use monero7 miner
2018-05-05 12:39:28.4014 [1] Stopping miners...
2018-05-05 12:39:28.4819 [1] Stopped miners.


Thanks,
--James

"2018-05-05 12:38:47.1695 [20] Switching to Monero Version 2"
Why is this crappy server not returning 7?
Does this mean I have to add "CryptoNight-HeavyV7"? What a mess...

Any luck on this?  Gonna clone your project in VS tonight see if I can help out  I have no experience in miners and the project is large sooooo;) Just a quick look on Github it appears that CNHeavy is broken into 3 and above and below 3, so I'm guessing this server should be returning 3 or higher so returning 2 is causing the wrong algo? Course could be way off:)

Oh and I tried Loki on the same pool, same result, its CNHeavy as well.   Stellite on a different pool however works fine(I know CN7 however haven't had time to try a diff Haven pool).

[zawawa]

I think I know the solution. The server is supposed to signal the V7 algorithm, but apparently it doesn't,
so all I have to do is to add "CryptoNight-HeavyV7" as a new algorithm, just like "CryptoNightV7".
I didn't want to clutter the user interface, but apparently there is no way around it.
I will make sure to fix the issue for the next version.

Quote from: Littlegrand on May 05, 2018, 03:40:39 PM


Quote

Quote from: Littlegrand on Today at 08:51:51 AM
Quote
It is all right. The log shows cryptonight heave loaded, the share submitted to the correct url, just rejected wrong algo...
Correct url didn't means that you select correct url for cryptonight heavy algo.
Most pools have several ports for different algos. If you chose port for cryptonightv7 but in miner you select CN heavy algo, miner will connect to pool via seleted port, and pool allows to connect, but miner will finds shares for heavy and send them on pool to port from wich pool wait shares for cryptonightV7, due to port.
Check port number for your pool. Is you right pointed it for CN heavy.

Yes it is correct port(though it does not show port in log).  The Haven address in log is correct as well.

It normal to say Switching to Monero 2 in log?

Any ideas?

Thanks

Please give me information about your pool, the coin you are mining, and your settings, and I will get to it right away

Haven

haven.ingest.cryptoknight.cc 5531  

Didnt optimize default settings:
Threads 1
Intensity 4 tried to 20
Work size 4
Strided index 1
Memory chunk 2

My laptop has NVIDIA geforce 740m.

Also tried it same settings on my 6 card AMD rig with 3 RX580 and 3 RX570, same thing

Log from laptop(same on rig)

2018-05-05 12:34:10.7939 [16] Running garbage collection...
2018-05-05 12:34:10.7939 [16] Memory used before collection: 14MB
2018-05-05 12:34:10.8900 [16] Memory used before collection: 12MB
2018-05-05 12:34:13.1940 [18] API Listener started.
2018-05-05 12:38:33.9272 [1] Launching miner(s) for Custom Pool 0...
2018-05-05 12:38:35.5019 [20] Miner thread for Device #0 started.
2018-05-05 12:38:35.5019 [20] NiceHash mode is off.
2018-05-05 12:38:35.6611 [20] Loaded Kernels\cryptonight_heavy.cl for Device #0.
2018-05-05 12:38:47.1088 [20] Built cryptonight_heavy program for Device #0.
2018-05-05 12:38:47.1088 [20] Build options:  -DSTRIDED_INDEX=1 -DMEMORY_CHUNK_SIZE=2  -IKernels -DWORKSIZE=4
2018-05-05 12:38:47.1695 [20] Switching to Monero Version 2
2018-05-05 12:38:57.8436 [20] Device #0 (CryptoNight): 30.11 h/s
2018-05-05 12:39:08.4715 [20] Device #0 (CryptoNight): 30.12 h/s
2018-05-05 12:39:08.8778 [12] Received new job: 800226972178483
2018-05-05 12:39:10.8926 [16] Running garbage collection...
2018-05-05 12:39:10.8926 [16] Memory used before collection: 16MB
2018-05-05 12:39:10.9617 [16] Memory used before collection: 15MB
2018-05-05 12:39:19.1095 [20] Device #0 submitted Share #0 to haven.ingest.cryptoknight.cc as (my have address taken out)
2018-05-05 12:39:19.1100 [20] Device #0 submitted a share to haven.ingest.cryptoknight.cc as  (my have address taken out)
2018-05-05 12:39:19.1100 [20] Device #0 (CryptoNight): 29.92 h/s
2018-05-05 12:39:19.7139 [12] Share #0 rejected: Wrong algo, use monero7 miner
2018-05-05 12:39:28.4014 [1] Stopping miners...
2018-05-05 12:39:28.4819 [1] Stopped miners.


Thanks,
--James

"2018-05-05 12:38:47.1695 [20] Switching to Monero Version 2"
Why is this crappy server not returning 7?
Does this mean I have to add "CryptoNight-HeavyV7"? What a mess...

Any luck on this?  Gonna clone your project in VS tonight see if I can help out  I have no experience in miners and the project is large sooooo;) Just a quick look on Github it appears that CNHeavy is broken into 3 and above and below 3, so I'm guessing this server should be returning 3 or higher so returning 2 is causing the wrong algo? Course could be way off:)

Oh and I tried Loki on the same pool, same result, its CNHeavy as well.   Stellite on a different pool however works fine(I know CN7 however haven't had time to try a diff Haven pool).

[UnclWish]

What about PHI1612?

[zawawa]

What about PHI1612?

I will take a look later. Still debugging the code optimizer now...

While I am waiting for the optimizer to complete, I finally started working on documentation.
I guess the time is about right as I am quite happy with the current user-interface.
After I am done with documentation and optimizations for the cards I have,
I will contact pool operators to ask them to put GGS on their "Getting Started" pages.
We will see...

[liomojo1]

There are timings, in full view, on this very forum for MJR that give 1200h/s on cn heavy at 1244/2125 570 nitro+ 8g.
Just keep looking.  

This forum is very big. And I can looking for years in search of right info. Maybe you share it to save our time?

Its no secret, 999000000000000022339D00CEDD6B4470551414B80D960B0040C400750314204A8900A00200000 018123745B92B3A16 Doktor put it in SRBMiner thread, and its really good compared to the one click strap.

Zawawa, some issues in latest version:
*Rx 560 2gb doesn't want to get the timing optimizations, neither do the R9 390
*GPU current clocks aren't shown in overclocking, instead are the last copied ones (copied from similar device etc.)
*Work size (4) in cryptonight should be enabled, if possible (for some odd GPUs i believe).. for some reason R9 390 on V7 gives similar results to claymore with two threads and a 740-820 intensity and most importantly 4 work size rather than 8.

In Old GG some GPUs of 2gb size were able to fit 448 intensityX2 threads, those gave noticeable higher gains in speed, not sure what rules, but, read somewhere that elpiads are different than samsung and hynix etc. in amount of memory available.

And you with doctor's timings and 1244/2125 clocks recieve 1200 h/s in GGS? What intensity? Can you give screen?

Man these timings are longer than the normal ones , are sure aboit them!

[heavyarms1912]

There are timings, in full view, on this very forum for MJR that give 1200h/s on cn heavy at 1244/2125 570 nitro+ 8g.
Just keep looking.  

This forum is very big. And I can looking for years in search of right info. Maybe you share it to save our time?

Its no secret, 999000000000000022339D00CEDD6B4470551414B80D960B0040C400750314204A8900A00200000 018123745B92B3A16 Doktor put it in SRBMiner thread, and its really good compared to the one click strap.

Zawawa, some issues in latest version:
*Rx 560 2gb doesn't want to get the timing optimizations, neither do the R9 390
*GPU current clocks aren't shown in overclocking, instead are the last copied ones (copied from similar device etc.)
*Work size (4) in cryptonight should be enabled, if possible (for some odd GPUs i believe).. for some reason R9 390 on V7 gives similar results to claymore with two threads and a 740-820 intensity and most importantly 4 work size rather than 8.

In Old GG some GPUs of 2gb size were able to fit 448 intensityX2 threads, those gave noticeable higher gains in speed, not sure what rules, but, read somewhere that elpiads are different than samsung and hynix etc. in amount of memory available.

And you with doctor's timings and 1244/2125 clocks recieve 1200 h/s in GGS? What intensity? Can you give screen?

Man these timings are longer than the normal ones , are sure aboit them!

It's not longer.  There is a extra space in the timing strap.

[liomojo1]

Yep , complete idiot I am 
do you have for samsung too?
these work like charm for my rx580 8gb hynix

[manotroll]

I downloaded but I do not know where to configure
Is there no demonstration of how to do this?

[joaocha]

What about PHI1612?

I will take a look later. Still debugging the code optimizer now...

While I am waiting for the optimizer to complete, I finally started working on documentation.
I guess the time is about right as I am quite happy with the current user-interface.
After I am done with documentation and optimizations for the cards I have,
I will contact pool operators to ask them to put GGS on their "Getting Started" pages.
We will see...

There is no good miner for Phi1612 yet, and lux is getting bigger everyday, you should give it a chance,

and release a linux version of your miner, most of the users  are on linux ^^

we need a simple version for linux, few parameters just to begin.

[hexagram]

Hi zawawa, everyone.

I have 12x RX 560 4GB Pulse, Windows 10 with the latest Adrenaline drivers (18.4.1) and I'm getting these sad hashes for Cryptonight-Heavy, otherwise the system would freeze:



These are my settings:



Any thoughts for improvement? 

[Mrsmoker]

sir, im trying upgrade from 1.3.5 to 1.3.7, a lot of error and my rig stop working, so i go back to 1.3.5.

[zawawa]

sir, im trying upgrade from 1.3.5 to 1.3.7, a lot of error and my rig stop working, so i go back to 1.3.5.

Could you be more specific about errors? You could try to install .NET Framework 4.7.1 and
Visual C++ Runtime 2017 manually.
It seems like there is a problem with the current installer, so I decided to switch to WiX for the next version.

[zawawa]

Hi zawawa, everyone.

I have 12x RX 560 4GB Pulse, Windows 10 with the latest Adrenaline drivers (18.4.1) and I'm getting these sad hashes for Cryptonight-Heavy, otherwise the system would freeze:



These are my settings:



Any thoughts for improvement? 

I haven't tested the miner with RX 560 yet. I will keep you posted.

[Littlegrand]

I think I know the solution. The server is supposed to signal the V7 algorithm, but apparently it doesn't,
so all I have to do is to add "CryptoNight-HeavyV7" as a new algorithm, just like "CryptoNightV7".
I didn't want to clutter the user interface, but apparently there is no way around it.
I will make sure to fix the issue for the next version.

Quote from: Littlegrand on May 06, 2018, 12:05:35 PM
Quote
Quote from: Littlegrand on May 05, 2018, 03:40:39 PM


Quote

Quote from: Littlegrand on Today at 08:51:51 AM
Quote
It is all right. The log shows cryptonight heave loaded, the share submitted to the correct url, just rejected wrong algo...
Correct url didn't means that you select correct url for cryptonight heavy algo.
Most pools have several ports for different algos. If you chose port for cryptonightv7 but in miner you select CN heavy algo, miner will connect to pool via seleted port, and pool allows to connect, but miner will finds shares for heavy and send them on pool to port from wich pool wait shares for cryptonightV7, due to port.
Check port number for your pool. Is you right pointed it for CN heavy.

Yes it is correct port(though it does not show port in log).  The Haven address in log is correct as well.

It normal to say Switching to Monero 2 in log?

Any ideas?

Thanks

Please give me information about your pool, the coin you are mining, and your settings, and I will get to it right away

Haven

haven.ingest.cryptoknight.cc 5531 

Didnt optimize default settings:
Threads 1
Intensity 4 tried to 20
Work size 4
Strided index 1
Memory chunk 2

My laptop has NVIDIA geforce 740m.

Also tried it same settings on my 6 card AMD rig with 3 RX580 and 3 RX570, same thing

Log from laptop(same on rig)

2018-05-05 12:34:10.7939 [16] Running garbage collection...
2018-05-05 12:34:10.7939 [16] Memory used before collection: 14MB
2018-05-05 12:34:10.8900 [16] Memory used before collection: 12MB
2018-05-05 12:34:13.1940 [18] API Listener started.
2018-05-05 12:38:33.9272 [1] Launching miner(s) for Custom Pool 0...
2018-05-05 12:38:35.5019 [20] Miner thread for Device #0 started.
2018-05-05 12:38:35.5019 [20] NiceHash mode is off.
2018-05-05 12:38:35.6611 [20] Loaded Kernels\cryptonight_heavy.cl for Device #0.
2018-05-05 12:38:47.1088 [20] Built cryptonight_heavy program for Device #0.
2018-05-05 12:38:47.1088 [20] Build options:  -DSTRIDED_INDEX=1 -DMEMORY_CHUNK_SIZE=2  -IKernels -DWORKSIZE=4
2018-05-05 12:38:47.1695 [20] Switching to Monero Version 2
2018-05-05 12:38:57.8436 [20] Device #0 (CryptoNight): 30.11 h/s
2018-05-05 12:39:08.4715 [20] Device #0 (CryptoNight): 30.12 h/s
2018-05-05 12:39:08.8778 [12] Received new job: 800226972178483
2018-05-05 12:39:10.8926 [16] Running garbage collection...
2018-05-05 12:39:10.8926 [16] Memory used before collection: 16MB
2018-05-05 12:39:10.9617 [16] Memory used before collection: 15MB
2018-05-05 12:39:19.1095 [20] Device #0 submitted Share #0 to haven.ingest.cryptoknight.cc as (my have address taken out)
2018-05-05 12:39:19.1100 [20] Device #0 submitted a share to haven.ingest.cryptoknight.cc as  (my have address taken out)
2018-05-05 12:39:19.1100 [20] Device #0 (CryptoNight): 29.92 h/s
2018-05-05 12:39:19.7139 [12] Share #0 rejected: Wrong algo, use monero7 miner
2018-05-05 12:39:28.4014 [1] Stopping miners...
2018-05-05 12:39:28.4819 [1] Stopped miners.


Thanks,
--James

"2018-05-05 12:38:47.1695 [20] Switching to Monero Version 2"
Why is this crappy server not returning 7?
Does this mean I have to add "CryptoNight-HeavyV7"? What a mess...

Any luck on this?  Gonna clone your project in VS tonight see if I can help out Wink  I have no experience in miners and the project is large sooooo;) Just a quick look on Github it appears that CNHeavy is broken into 3 and above and below 3, so I'm guessing this server should be returning 3 or higher so returning 2 is causing the wrong algo? Course could be way off:)

Oh and I tried Loki on the same pool, same result, its CNHeavy as well.   Stellite on a different pool however works fine(I know CN7 however haven't had time to try a diff Haven pool).

I tried ggs on Haven on another pool. Still switches to Monero version 2 and pool rejects share with message incorrect share. It surprises me that no one else is having this problem?

If anyone is mining Haven with GGS, what pool are you using?

[skablast]

I tested CryptoNight-Heavy with default pools again and didn't see any problems.
I will look into this.
[/quote



SO, to answer you.

yes, it was highlighted some time ago

No, it wasn't sorted out

I didn't bother keep chasing because of 3 factor:

Version before last one were simply not working in Mixed rigs with 8gb card &Nvda
and:

The speed that was show from Zawawa on the continuation of that quote was 1/2 of the average speed I was getting aniway, as I have 1100 h/s from RX 580 8gb. While on CN7 normal GGS speed is already at the top.
So CN-heavy was clearly a rush implementation, and other CN7- variant light or heavy (IPBC, Alloy,LOKI,ARTO) has been 50 % more profitable for me  than heavy but not supported by GGS

I did advise again later

you need an 8gb rx580, because CN heavy is different.
]


If you have to mine  Haven, aniway the best pool is :
haven.miner.rocks, by far (12,5 %).
cryptoKnight.cc is reporting less speed and share while being  good for other CN coin.
Don't ask me why, I am not a tech guy but I am used to compare $ to $ each day, each Rig, each miner to decide WTM  the day after and where mine it so...

[zawawa]

I remember I reached 1k with CryptNight-Heavy on RX 480, so I think it's a matter of adjusting parameters.
I am done debugging the code optimizer for now, so I should be able to get to it the first thing tomorrow morning.

[zawawa]

This is like a dream for GPGPU programmers, methinks.
What do you guys think?

[skablast]

you know I'm not a dev and not english native.

But I can read:


The size of the SIMD VGPR files introduce an important limit: VGPRs of the SIMD are evenly divided between threads of the active waves. If a shader requires more VGPRs than are available, the SIMD will not be able to execute the optimal number of waves. Occupancy, a measure of the parallel work that the GPU could perform at a given time, will suffer as a result.

https://gpuopen.com/optimizing-gpu-occupancy-resource-usage-large-thread-groups/

So, in my limited understanding, the result will be a better occupancy thus an increase in performance from 2 % up to 10 % depends of the kernel.

and because of this:

GCN was very VGPR limited, I managed to get 3 out of a maximum of 10
threads to fit by isolating bits of code in dummy branches. Since the
algorithm is memory heavy any improvements here would help a lot, but
unfortunately although I think VGPR use could be substantially lower, the
AMD compiler likes to go crazy and use as many as possible.

I think you basically rewrite a part of the AMD compiler ?

Do not forget to query for number of (unlocked) shaders and not refer at the standard number ,as most of the Baffin series can be unlocked

[hexagram]

Here's another rig I have running Cryptonight-Heavy with GGS 1.3.7 alpha. This rig has 10x RX 570 8GB Nitro+ but somehow Windows 10 detects them as RX 580 after upgrading to Adrenaline 18.4.1



and my GGS settings:



Those churning ~820 h/s are Samsung DDRs and the rest at ~750 h/s are Hynix

I haven't toyed with the Memory Timings, kind of scared to fuck up my cards.

Anyone can help me improve my hashrate please? 

Will be happy to give a tip (0.005 BTC) if someone can help me get them to ~900-1000+ with stability.