Gateless Gate Sharp 1.3.8: 30Mh/s (Ethash) on RX 480! - page 167.

toptek

legendary

Activity: 1274

Merit: 1000

Quote from: zawawa on February 09, 2017, 07:46:00 PM

Don't get me wrong. I truly admire Claymore's coding skills. To be very honest, I am rarely impressed by other programmers, but he is definitely top notch.

All I am saying is that he studies his competitions, be it potential or actual, and he studies them really well. This is not an automatic assumption but an educated, informed observation of his work as a long-time user. Also, it is often the case that different developers come up with different ideas for optimizations, so we are constantly learning from each other.

Anyway, all this talk is meaningless if I cannot come up with a faster miner than Claymore's. We shall see very, very soon.

I hope you can at least match it , either way to keep you at it, i plan to keep donating like i said in a PM

. for a while ....

zawawa

sr. member

Activity: 728

Merit: 304

Miner Developer

Don't get me wrong. I truly admire Claymore's coding skills. To be very honest, I am rarely impressed by other programmers, but he is definitely top notch.

All I am saying is that he studies his competitions, be it potential or actual, and he studies them really well. This is not an automatic assumption but an educated, informed observation of his work as a long-time user. Also, it is often the case that different developers come up with different ideas for optimizations, so we are constantly learning from each other.

Anyway, all this talk is meaningless if I cannot come up with a faster miner than Claymore's. We shall see very, very soon.

zzzzzzzzzz

full member

Activity: 150

Merit: 100

Quote from: xeridea on February 09, 2017, 06:58:34 PM

Quote from: zawawa on February 09, 2017, 05:53:34 PM

Quote from: jstefanop on February 09, 2017, 04:40:20 PM

Seems like claymore is in the same spot...wouldnt be surprised if his recent "eureka" moment is from the discussion happening in this thread.

Oh, I'm now pretty sure he reads this thread. It's really funny in an ironic sense, but it does make things a bit more challenging for me. It's like playing poker with you being the only player with open hands. Oh well.

I am not entirely sure if I would take his current approach, though. I'm not going to feed my competition this time around, so he has to figure things out himself...

I find these assumptions a bit odd, he has been fastest 95% of the time since launch, and people assume he uses methods from slower miners. I think recent improvements are just to keep up with Optiminer, and timing is a coincidence. He already had ASM working well many weeks ago for all cards. I can't know for sure, I just find it odd that every time there is a CM speedup, some assume automatically it is from others.

Keep up the good work, just posting my thoughts.

So, if CM had "ASM working well many weeks ago for all cards", why did he post this 6 days ago:

https://bitcointalksearch.org/topic/m.17725004

And no, I'm not implying, in any way, that CM is doing anything unethical. I'm just saying, according to CM himself, your statement is completely false.

FFI2013

hero member

Activity: 906

Merit: 507

Quote from: xeridea on February 09, 2017, 06:58:34 PM

Quote from: zawawa on February 09, 2017, 05:53:34 PM

Quote from: jstefanop on February 09, 2017, 04:40:20 PM

Seems like claymore is in the same spot...wouldnt be surprised if his recent "eureka" moment is from the discussion happening in this thread.

Oh, I'm now pretty sure he reads this thread. It's really funny in an ironic sense, but it does make things a bit more challenging for me. It's like playing poker with you being the only player with open hands. Oh well.

I am not entirely sure if I would take his current approach, though. I'm not going to feed my competition this time around, so he has to figure things out himself...

I find these assumptions a bit odd, he has been fastest 95% of the time since launch, and people assume he uses methods from slower miners. I think recent improvements are just to keep up with Optiminer, and timing is a coincidence. He already had ASM working well many weeks ago for all cards. I can't know for sure, I just find it odd that every time there is a CM speedup, some assume automatically it is from others.

Keep up the good work, just posting my thoughts.

yes but zawawa was talking asm than suddenly claymore implemented it that's just weird timing but could be a coincidence

xeridea

sr. member

Activity: 449

Merit: 251

Quote from: zawawa on February 09, 2017, 05:53:34 PM

Quote from: jstefanop on February 09, 2017, 04:40:20 PM

Seems like claymore is in the same spot...wouldnt be surprised if his recent "eureka" moment is from the discussion happening in this thread.

Oh, I'm now pretty sure he reads this thread. It's really funny in an ironic sense, but it does make things a bit more challenging for me. It's like playing poker with you being the only player with open hands. Oh well.

I am not entirely sure if I would take his current approach, though. I'm not going to feed my competition this time around, so he has to figure things out himself...

I find these assumptions a bit odd, he has been fastest 95% of the time since launch, and people assume he uses methods from slower miners. I think recent improvements are just to keep up with Optiminer, and timing is a coincidence. He already had ASM working well many weeks ago for all cards. I can't know for sure, I just find it odd that every time there is a CM speedup, some assume automatically it is from others.

Keep up the good work, just posting my thoughts.

fr4nkthetank

legendary

Activity: 2294

Merit: 1182

Now the money is free, and so the people will be

Quote from: zawawa on February 09, 2017, 06:26:35 PM

Hallelujah! My GCN compiler is finally working!
Well, it is a Frankenstein of LLVM/Clang and CLRadeonExtender, but as long as it works...

Cant wait to test out your next update !!! I just can't get enough of em flashy colors

m1n1ngP4d4w4n

full member

Activity: 224

Merit: 100

CryptoLearner

Quote from: zawawa on February 09, 2017, 06:26:35 PM

Hallelujah! My GCN compiler is finally working!
Well, it is a Frankenstein of LLVM/Clang and CLRadeonExtender, but as long as it works...

Good work man, now dazzle us Wink

zawawa

sr. member

Activity: 728

Merit: 304

Miner Developer

Hallelujah! My GCN compiler is finally working!
Well, it is a Frankenstein of LLVM/Clang and CLRadeonExtender, but as long as it works...

zawawa

sr. member

Activity: 728

Merit: 304

Miner Developer

Quote from: jstefanop on February 09, 2017, 04:40:20 PM

Seems like claymore is in the same spot...wouldnt be surprised if his recent "eureka" moment is from the discussion happening in this thread.

Oh, I'm now pretty sure he reads this thread. It's really funny in an ironic sense, but it does make things a bit more challenging for me. It's like playing poker with you being the only player with open hands. Oh well.

I am not entirely sure if I would take his current approach, though. I'm not going to feed my competition this time around, so he has to figure things out himself...

jstefanop

legendary

Activity: 2174

Merit: 1401

Seems like claymore is in the same spot...wouldnt be surprised if his recent "eureka" moment is from the discussion happening in this thread.

zawawa

sr. member

Activity: 728

Merit: 304

Miner Developer

Quote from: hannusolo on February 08, 2017, 01:14:43 PM

But I think you shouldn't bother with comfort functions like this only if you really have notjhng to do Cheesy

I know, I know... I'm still debugging my custom LLVM, but it is coming along.
I stripped away most of the unnecessary targets and tools to make it relatively lightweight.
I am taking my time here because it is crucial for miner development to have inline GCN assembly and a set target as opposed to ever-changing AMD drivers.
This may seem like a detour, but it is actually a necessary step for me to continue miner development without losing sanity.

h311m4n

sr. member

Activity: 487

Merit: 266

Quote from: VaynardW on February 08, 2017, 07:47:04 AM

I had the same exact issue at first, it also happens to one of my rig,
It turns out that rig still using Catalyst 16.9, after a DDU uninstall and upgrading to 16.11, the problem seems to disappear

Alright thanks for the tip!

hannusolo

sr. member

Activity: 434

Merit: 257

Quote from: zawawa on February 08, 2017, 09:59:28 AM

Building LLVM is taking forever, so I have to entertain myself with something else...
So I have been thinking for quite some time about adding a nice graphical user interface/remote monitor to Gateless Gate, and I think I want to go with MultiMiner as somebody suggested here a while ago:

https://github.com/nwoolls/MultiMiner

It's open-source, compatible with sgminer, and seems to be feature-complete with automatic currency-switching.
What do you guys think?

This is a good idea, it would make it more user friendly, which would help the average people who might have a good gpu but is using it only for gaming and doesn't want to mess with command windows.

But I think you shouldn't bother with comfort functions like this only if you really have notjhng to do Cheesy

zawawa

sr. member

Activity: 728

Merit: 304

Miner Developer

Quote from: nerdralph on February 08, 2017, 11:27:27 AM

Quote from: zawawa on February 07, 2017, 04:27:55 PM

Quote from: nerdralph on February 07, 2017, 01:46:30 PM

Quote from: zawawa on February 07, 2017, 04:31:48 AM

I finally tracked down the last bug and was able to confirm that LLVM/Clang can be used to generate binaries for AMD's proprietary OpenCL drivers. Since LLVM is only capable of generating ROCm binaries, I had to insert a thin ABI translation layer at the beginning of the kernel.

According to a source at AMD, the ROCm ABI is the same as the HSAIL/brig ABI. When I used CodeXL to generate a kernel binary on a system with Crimson 16.10 drivers, the binary contained HSAIL as well as brig symbols. Since AMDs drivers can handle HSA/brig kernels, I would think they should also handle ROCm binaries.

What I found out was that AMD's drivers add extra hidden arguments to kernels, which completely messes up the order of arguments. Therefore, modifications to LLVM are necessary in order to use it for AMD's proprietary drivers.

Interesting. I started to reverse-engineer the fglrx drivers but only got part-way through it. It appears most of the clang/llvm mods by AMD are open source, but it's rather difficult to figure out which branch with which patches are used in which driver release. Here's an example of how AMD has been reasonably transparent about what they've been doing with llvm:
http://llvm.org/devmtg/2013-11/slides/Stellard-R600.pdf

I don't know... They just moved implicit arguments at the end of the argument list to to the beginning. This does not make any sense to me, and I did not find any traces or mentions of this seemingly arbitrary change. Things like this make me believe that open-source tools are crucial for GPGPU and AMD is not doing enough.

In any case, it doesn't matter to me because I'm almost done with my custom LLVM. Can't wait to actually use it...

zawawa

sr. member

Activity: 728

Merit: 304

Miner Developer

Quote from: YIz on February 08, 2017, 11:51:24 AM

Thanks for the good work zawawa and nerdalph, seems to be very close to Claymore's on my hardware and without the fee, but I'm unsuccessful in trying to make it start with Windows.

My crystal ball has been abused with assembly programming, so I need to know the following:

(1) your OS,
(2) the coin you tried to mine, and
(3) the display driver you are using.

YIz

hero member

Activity: 686

Merit: 502

Thanks for the good work zawawa and nerdalph, seems to be very close to Claymore's on my hardware and without the fee, but I'm unsuccessful in trying to make it start with Windows.

nerdralph

sr. member

Activity: 588

Merit: 251

Quote from: zawawa on February 07, 2017, 04:27:55 PM

Quote from: nerdralph on February 07, 2017, 01:46:30 PM

Quote from: zawawa on February 07, 2017, 04:31:48 AM

I finally tracked down the last bug and was able to confirm that LLVM/Clang can be used to generate binaries for AMD's proprietary OpenCL drivers. Since LLVM is only capable of generating ROCm binaries, I had to insert a thin ABI translation layer at the beginning of the kernel.

According to a source at AMD, the ROCm ABI is the same as the HSAIL/brig ABI. When I used CodeXL to generate a kernel binary on a system with Crimson 16.10 drivers, the binary contained HSAIL as well as brig symbols. Since AMDs drivers can handle HSA/brig kernels, I would think they should also handle ROCm binaries.

What I found out was that AMD's drivers add extra hidden arguments to kernels, which completely messes up the order of arguments. Therefore, modifications to LLVM are necessary in order to use it for AMD's proprietary drivers.

Interesting. I started to reverse-engineer the fglrx drivers but only got part-way through it. It appears most of the clang/llvm mods by AMD are open source, but it's rather difficult to figure out which branch with which patches are used in which driver release. Here's an example of how AMD has been reasonably transparent about what they've been doing with llvm:
http://llvm.org/devmtg/2013-11/slides/Stellard-R600.pdf

zawawa

sr. member

Activity: 728

Merit: 304

Miner Developer

Oh, I forgot to add that MultiMiner is cross-platform as well.
Actually, it seems to be working quite well with GG...

zawawa

sr. member

Activity: 728

Merit: 304

Miner Developer

Building LLVM is taking forever, so I have to entertain myself with something else...
So I have been thinking for quite some time about adding a nice graphical user interface/remote monitor to Gateless Gate, and I think I want to go with MultiMiner as somebody suggested here a while ago:

https://github.com/nwoolls/MultiMiner

It's open-source, compatible with sgminer, and seems to be feature-complete with automatic currency-switching.
What do you guys think?

fr4nkthetank

legendary

Activity: 2294

Merit: 1182

Now the money is free, and so the people will be

Quote from: VaynardW on February 08, 2017, 07:47:04 AM

I had the same exact issue at first, it also happens to one of my rig,
It turns out that rig still using Catalyst 16.9, after a DDU uninstall and upgrading to 16.11, the problem seems to disappear

Yeah get the 16.11 drivers and I would recommend a custom fan profile with MSI afterburner. Unless you are on linux I guess. It gets the job done well.

Topic: Gateless Gate Sharp 1.3.8: 30Mh/s (Ethash) on RX 480! - page 167. (Read 214431 times)