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

[coke15]

hi

vers 1.1.16     more stable for me but i have a prob: nicehash reject 100%.

when i try in "default pool" nicehash and my wallet,in log i have "nicehash mode is on",share accepted

but i put "custom pool" why the good adress,my worker,and password,in log  "nicehash mode is off" share rejected


in default mode i mine for me or you?

[OhMyHolyToken]

Now hashrate for Radeon 280/280X is much better.

4100 for 8 cards. Claymore 11.2 still faster (4500), but less stable. I thinks there is a room for improvement, my Asus 280X give 600 with Claymore 11.2, and only 530 with Gateless_Gate_Sharp_1.1.16. MSI 290 give 785 with Clay and 750 with Gateless.
https://image.ibb.co/fgJ0pw/116.jpg

RAW intensity for 280 is 56 and for 280X is 64. For 290 is 80. I guess it means 560, 640 and 800?

Claymore use 400 for 280/280X and 640 for 290. I tired that for GGS and result was much lower.

ps. Any plans for equihash?
pps: Any support of SSL pools? If yes - how?
ppps. Fan control must be disabled by default. 50% for all cards? People use Afterburner for that.

Thanks for hard work.

[nerdralph]

I will be focusing on the long due Equihash and optimizations for other algorithms for a week or two.
I really need to concentrate, so you may see me less often here. I will still fix critical bugs, though.

Have fun playing with the GDS.  :-)  I'm guessing you'll just be doing binary kernels?  If so, I'd still like to see your asm source.

[zawawa]

hi

vers 1.1.16     more stable for me but i have a prob: nicehash reject 100%.

when i try in "default pool" nicehash and my wallet,in log i have "nicehash mode is on",share accepted

but i put "custom pool" why the good adress,my worker,and password,in log  "nicehash mode is off" share rejected


in default mode i mine for me or you?

For you, of course! Just don't use NiceHash as a custom pool. Their protocols are just too different.

[zawawa]

RAW intensity for 280 is 56 and for 280X is 64. For 290 is 80. I guess it means 560, 640 and 800?

Claymore use 400 for 280/280X and 640 for 290. I tired that for GGS and result was much lower.

I really have to implement automated benchmarks, don't I? Let me see...

[treanski]

implementing oc settings would be awesome too
just kidding...i know u work hard

[zawawa]

Hello o7
GPU: Gigabyte Windforce 2 NVIDIA GeForce GTX 1060 3GB (Samsung)
OS: Windows 7 Service Pack 1
Video Driver:  388.71
MSI Afterburn: 4.4.2
Gateless_Gate_Sharp_1.1.16_alpha

Device #0: 252,01 Kh/s (NeoScrypt) vs other miner what can hit 550-650 Kh/s = very LOW speed
Device #0 (CryptoNight): 154,35 h/s vs ccminer.v.2.2.3 = 512 h/s = very LOW speed
Device #0 (Ethash): 23,16 Mh/s > is most close to a good speed 24 Mh/s = GOD speed

- this error was trow when Ethash i was test it:
[1] Exception: The given key was not present in the dictionary.   at System.ThrowHelper.ThrowKeyNotFoundException()
   at System.Collections.Generic.Dictionary`2.get_Item(TKey key)
   at GatelessGateSharp.MainForm.UpdateStatsWithLongPolling()

Comment:

1. You try to do to many thing at this point, in stead focus on optimize miner algo aka miner application it not suppose to OC, Control Fan or tune in any way GPU > GPU tune it need to be done before any miner application (what ever it is) it start to be be use & miner it need just to use setting what is already done ... for that it exist MSI Afterburn, for that ppls make update & tune video bios, so short version your application need to start & mine. IF in future you can add extra option what allow to make fine tune is ok BUT first focus on optimize mining algorithm ...

2. Application is very lagy start, stop, close, open, change thing it response very slow & something it have habit to freze or not response for 10-15 sec

3. Warning with Windows Swap size is bad one for a system with 8 G ram & 2 video gpu asking to set 24 G Swap O.o really ?!?!?! 8 - 16 G swap is more that inaff for mining or gaming or anything what is need to be done, add an option to remove, ignore that warning from beginning ...

cheers

Thank you for a detailed feedback.
I think there are some misunderstandings on your part, however.
First, GGS is intended not to be a bare-bone miner but to be a complete mining package.
Second, my current focus is not on NVIDIA but on AMD.
I have most of current NVIDIA cards as I worked for NiceHash before and I will eventually get to them, but I would like to achieve a top-notch performance with GGS and AMD cards first.

In any case, GGS'es design choices only make sense if you think of it more as a comprehensive platform than as a component part.

[zawawa]

I will be focusing on the long due Equihash and optimizations for other algorithms for a week or two.
I really need to concentrate, so you may see me less often here. I will still fix critical bugs, though.

Have fun playing with the GDS.  :-)  I'm guessing you'll just be doing binary kernels?  If so, I'd still like to see your asm source.

Hehe, I will. I am thinking about giving another shot at LLVM and the inline GCN assembly. We will see...

[zawawa]

(1) LLVM's AMDGPU backend is compatible with Mesa 3D.
(2) Mesa 3D is compatible with Windows.

Does this mean that I could use AMDGPU without modifications!?

[treanski]

(1) LLVM's AMDGPU backend is compatible with Mesa 3D.
(2) Mesa 3D is compatible with Windows.

Does this mean that I could use AMDGPU without modifications!?

i dont know what your are talking about but if you need testers here we are

[sir_blacks]

Thank you for a detailed feedback.
I think there are some misunderstandings on your part, however.
First, GGS is intended not to be a bare-bone miner but to be a complete mining package.
Second, my current focus is not on NVIDIA but on AMD.
I have most of current NVIDIA cards as I worked for NiceHash before and I will eventually get to them, but I would like to achieve a top-notch performance with GGS and AMD cards first.

In any case, GGS'es design choices only make sense if you think of it more as a comprehensive platform than as a component part.

Well i not undestend what i do "misunderstandings" but well it was a feedback & how i see actual situation.

- IF you want to make a complete mining package is your choose & is nice but i am just tell you how i see thing: a application/package what try to do to many, with untune algorithm (but part with tune is because see below). IF you want to implement something similar with MSI Afterburner inside application is your choose off course but at list give ppls option to use your OC system or not & say it in description, because if there are 2 app what try to OC in same time a device they will end fight & cause trouble Let ppls to chose what part it want to use from your package, it will be wise because it will make application more flexible ...

- IF "focus is not on NVIDIA but on AMD" then say it in first OP treath / program description, coz there are already ppls who ask how it work app from NVIDIA perspective

anyway anything what you choose to do is your & only your decision i & any person who try your GGS it will tell how it see thing from there side, perspective ... at list from my side everything is perfect clear & i undestend exactly what i do (my be my writing in eng is horrible i know i am eng butcher )

cheers & glad to see your work & special you was not stop work after nicehash experience end

[LinuxETC]

I there a decent if not thorough HowTo compile this for Linux (Ubuntu 16.04LTS preferred if possible) available out there?  Not finding much right off hand in the Github repo.

Thanks in advance.

[JJkod]

Something is wrong.

I did a test with a low difficulty coin, electronic dollar, cryptonight one.

Speed is great, over 950mhs with RX 580, but the number of accepted shares is wayyyyy off.

1 minute with gateless get 5 shares.
1 minute with cast xmr  gets 57 shares.

I tried in another rig and same results. So I'm not sure if it is a problem submitting shares or shares were sent but there is a problem counting them in the software.

[UnclWish]

Something is wrong.

I did a test with a low difficulty coin, electronic dollar, cryptonight one.

Speed is great, over 950mhs with RX 580, but the number of accepted shares is wayyyyy off.

1 minute with gateless get 5 shares.
1 minute with cast xmr  gets 57 shares.

I tried in another rig and same results. So I'm not sure if it is a problem submitting shares or shares were sent but there is a problem counting them in the software.

You sure that GGS and cast xmr connecting to the same port? Difficulty is the same? What pool show? Look at accepted speed.

[UnclWish]

zawawa, maybe it's time to update the source code for VS 2017? 2013 is too old allready...

[JJkod]

Something is wrong.

I did a test with a low difficulty coin, electronic dollar, cryptonight one.

Speed is great, over 950mhs with RX 580, but the number of accepted shares is wayyyyy off.

1 minute with gateless get 5 shares.
1 minute with cast xmr  gets 57 shares.

I tried in another rig and same results. So I'm not sure if it is a problem submitting shares or shares were sent but there is a problem counting them in the software.

You sure that GGS and cast xmr connecting to the same port? Difficulty is the same? What pool show? Look at accepted speed.

yeah, same settings and difficulty. Pool shows that the hashrate is fine, but doesn't show the number of submitted shares, so I don't know if it is submitting them or not.

[zawawa]

Something is wrong.

I did a test with a low difficulty coin, electronic dollar, cryptonight one.

Speed is great, over 950mhs with RX 580, but the number of accepted shares is wayyyyy off.

1 minute with gateless get 5 shares.
1 minute with cast xmr  gets 57 shares.

I tried in another rig and same results. So I'm not sure if it is a problem submitting shares or shares were sent but there is a problem counting them in the software.

The number of shares is dependent on the difficulty, so this kind of comparison is meaningless unless difficulties are exactly the same throughout.
I would compare hashrates reported by the pool.

[sp_]

Pool shows that the hashrate is fine, but doesn't show the number of submitted shares, so I don't know if it is submitting them or not.

Never heard of currupt pools?

submitted shares don't meen a shit unless they are verified...

[zawawa]

zawawa, maybe it's time to update the source code for VS 2017? 2013 is too old allready...

VS 2013 is much more desirable for compatibility reasons.
I won't upgrade VS unless it is absolutely necessary.

[nerdralph]

(1) LLVM's AMDGPU backend is compatible with Mesa 3D.
(2) Mesa 3D is compatible with Windows.

Does this mean that I could use AMDGPU without modifications!?

At least on the Linux side, as of amdgpu-pro 17, the default OpenCL compiler has been llvm (HSA/ROCm).  As you know, it's optimizer for GCN is shit compared to the old gcc version, hence my reason for using -legacy.  I haven't tried 17.50 yet, but 17.40 didn't recognize inline asm.  There's lots of commits going on, so you'll probably just have to go with 7.0.0/trunk unless the amdgpu updates get back ported to 6.0.  The docs are quite thorough now, if you haven't looked at them in the last few months:
https://llvm.org/docs/AMDGPUUsage.html

I think you'll have to write most of the code in straight asm, as it will be a long time before the optimizer gets good.  Some instruction intrinsics like alignbit have only been added in the last few months, and as far as I can tell, there are no optimization hints to indicate different instruction timing.
https://reviews.llvm.org/D34046