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

[UnclWish]

I don't want to sound ungrateful or anything, but you seem to be wasting your talent on things that don't matter in mining. Forget about the GUI, it just brings bugs and its actually harder to work with. Forget the fancy auto overclocks and straps etc...

What we AMD miners really need is good kernels for profitable algos. Give us that and you will have more then enough money and users with a resonable fee for your miner. I tried using your miner a few times, but in the end I only have problems with the GUI and some obscure bugs because of that. And in the end other miners are simply faster and easier to use because of the command line.

Just my 2 cents, don't waste your potential.

P.S. I am sorry if I sounded hars, but I am always blunt and to the point, ubable to sugarcoat.

GGS allready many versions have command line version. GUI maybe used only for changing settings. You can not use overclocking, straps, etc. You can just setup algo, pool, wallet and that's all. After that you can close GUI and launch command line version. That's it.
zawawa spend many time not on GUI. He made a big work on tuning kernels. That's what you wrote AMD miners need. He works on it almost all time. So wait if you want to use it with very good kernels.

[1715269117]

1715269117

[1715269117]

1715269117

[zawawa]

I don't want to sound ungrateful or anything, but you seem to be wasting your talent on things that don't matter in mining. Forget about the GUI, it just brings bugs and its actually harder to work with. Forget the fancy auto overclocks and straps etc...

What we AMD miners really need is good kernels for profitable algos. Give us that and you will have more then enough money and users with a resonable fee for your miner. I tried using your miner a few times, but in the end I only have problems with the GUI and some obscure bugs because of that. And in the end other miners are simply faster and easier to use because of the command line.

Just my 2 cents, don't waste your potential.

P.S. I am sorry if I sounded hars, but I am always blunt and to the point, ubable to sugarcoat.

I don't mind the language at all. You probably don't know, however, that I did write the fastest kernels for some of the algorithms
before, both open-source and private. It is just that I personally prefer GUI over CUI and I don't like the user interface of other miners at all.
I will better document the command-line version and work on kernels.
In fact, you will see significantly improved Ethash kernels very soon.
I ended up choosing the 84 VGPR's version over 64 VGPR's version, and I am tweaking default settings right now.

[Lio78]

I don't want to sound ungrateful or anything, but you seem to be wasting your talent on things that don't matter in mining. Forget about the GUI, it just brings bugs and its actually harder to work with. Forget the fancy auto overclocks and straps etc...

What we AMD miners really need is good kernels for profitable algos. Give us that and you will have more then enough money and users with a resonable fee for your miner. I tried using your miner a few times, but in the end I only have problems with the GUI and some obscure bugs because of that. And in the end other miners are simply faster and easier to use because of the command line.

Just my 2 cents, don't waste your potential.

P.S. I am sorry if I sounded hars, but I am always blunt and to the point, ubable to sugarcoat.

I don't mind the language at all. You probably don't know, however, that I did write the fastest kernels for some of the algorithms
before, both open-source and private. It is just that I personally prefer GUI over CUI and I don't like the user interface of other miners at all.
I will better document the command-line version and work on kernels.
In fact, you will see significantly improved Ethash kernels very soon.
I ended up choosing the 84 VGPR's version over 64 VGPR's version, and I am tweaking default settings right now.

I think your GUI is very usefull and in CN7 your kernel is very good, maybe not the best but very very similar result.
I love the semplicity to set up voltage and clock.

I think is a great work from zawawa, obviously improvable, but great.

I hope to see your new work very soon, I work with 1.3.7 and 15.8.1 adrenaline driver because is most stable for me of 1.3.8.

[QuirkSilver]

I agree, the Sqlite is really slow. Unless you can't speed it up, i'd prefer something easier as well.

My opinion on the card settings:

- Radeon 550, the Baffin is slower on GGS than on xmr-stak. About 8-10%, Where xmr-stak gives a steady 560 H/s @1150/1950,  GGS is jumping between 510 and 540. Maybe the algo needs a closer look.

- Memory timings: I posted a elpida strap for cryptonight, before, did someone use it? I managed to squeeze a bit more out of it for the bad memory cards, giving now 920h/s with 1150/1750.
As soon i good some free time i will check the memory timings and hand them over...GGS deserves this support, really.

Maybe this helps.

Although this is old, but i couldn't PM you cause i'm a newbie i wish to speak to you about the straps, i mod my own (actually experimenting). I find your straps smart though but most importantly i also have loads of those 560s (same as 550) and intersted to know on what basis you changed the memory learning part?

[t3syncro]

I agree, the Sqlite is really slow. Unless you can't speed it up, i'd prefer something easier as well.

My opinion on the card settings:

- Radeon 550, the Baffin is slower on GGS than on xmr-stak. About 8-10%, Where xmr-stak gives a steady 560 H/s @1150/1950,  GGS is jumping between 510 and 540. Maybe the algo needs a closer look.

- Memory timings: I posted a elpida strap for cryptonight, before, did someone use it? I managed to squeeze a bit more out of it for the bad memory cards, giving now 920h/s with 1150/1750.
As soon i good some free time i will check the memory timings and hand them over...GGS deserves this support, really.

Maybe this helps.

...on what basis you changed the memory learning part?

Uh, i don't really understand what you mean by memory learning part. I couldn't investigate further as GGS memory timings and Bios memory straps don't "fit". This seems to be because of the AMD driver that is dynamically changing memory settings. In other words: The straps/settings you put into a bios mod will neither work nor be shown inevitably in GGS. The values there differ from the straps you put in the Bios mod.

[UnclWish]

...on what basis you changed the memory learning part?

Uh, i don't really understand what you mean by memory learning part. I couldn't investigate further as GGS memory timings and Bios memory straps don't "fit". This seems to be because of the AMD driver that is dynamically changing memory settings. In other words: The straps/settings you put into a bios mod will neither work nor be shown inevitably in GGS. The values there differ from the straps you put in the Bios mod.

I think that he really didn't know that ))))

[UnclWish]

zawawa, when we can try new version? What's the progress?

[zawawa]

The Ethash kernel for GCN3 is done.
Theoretically spraking, it cannot get any faster than as it is now with four active wavefronts
per CU.
I am sick of kooking at Ethash, though...

[heavyarms1912]

The Ethash kernel for GCN3 is done.
Theoretically spraking, it cannot get any faster than as it is now with four active wavefronts
per CU.
I am sick of kooking at Ethash, though...

and what sort of improvements do you see? 33-35 H/s is what should be an excellent achievement.

[UnclWish]

The Ethash kernel for GCN3 is done.
Theoretically spraking, it cannot get any faster than as it is now with four active wavefronts
per CU.
I am sick of kooking at Ethash, though...

and what sort of improvements do you see? 33-35 H/s is what should be an excellent achievement.

Maybe you wanted to say 33-35 MH/s?

[nerdralph]

The Ethash kernel for GCN3 is done.
Theoretically spraking, it cannot get any faster than as it is now with four active wavefronts
per CU.
I am sick of kooking at Ethash, though...

You don't need four wavefronts to saturate the VALU.  I explained that in an old blog post:
"Note that some sources state that full SIMD occupancy requires four waves, when it is technically possible with just one wave using only vector instructions."
http://nerdralph.blogspot.com/2017/02/inside-amd-gcn-code-execution.html

I realized a long time ago that the key to optimal ETHASH performance is not getting more waves, it's optimizing the pattern memory accesses across the CUs to avoid contention.

[heavyarms1912]

The Ethash kernel for GCN3 is done.
Theoretically spraking, it cannot get any faster than as it is now with four active wavefronts
per CU.
I am sick of kooking at Ethash, though...

and what sort of improvements do you see? 33-35 H/s is what should be an excellent achievement.

Maybe you wanted to say 33-35 MH/s?

Yup.  Typo. 33-35 MH/s is what i meant.

[xpander7]

Hi. Can I mine Lyra2z using Gateless Gate Miner? Im using simplemining.

When I set it up it shows activity in the pool, but the hash power shows 0 MH/s.

Thanks in advance.

[Lio78]

No more update? 

I hope to see new version...

[UnclWish]

zawawa, nicehash added X16R algo. Therefore please add in new version nicehash heavy and X16R algos.

[zawawa]

Sorry for the absence. I have been trying to get maximum performance with on-the-fly memory timing mods and the new Ethash kernel, but it turned out to be notoriously difficult to overclock the GPU to the limits with on-the-fly memory timing mods. With modded BIOS'es, I was able to get 32.5 MH/s with RX 580.

I am still communicating with AMD engineers through an official channel for memory timing mods for Vega. Even without altered timings, I was able to get 44.5 MH/s with Vega 64. The new kernels are already available on GitHub if you are really interested.

Now, the good new for you guys is that I am planning to drop the DEVFEE for Gateless Gate Sharp entirely as I am shifting focus to private
and customized miners. I will continue to work on Gateless Gate Sharp as a technical demonstration, though. We will see.

[Iamtutut]

I'd prefer a devfee and more frequent updates. But I guess you'll get more money with your new ways, so I wish you the very best.
32,5MH/s with a 480 is a huge performance ! What's the power draw please ? I think my best 570 gets 31,5MH/s while drawing 120W.

BTW, since you talk with AMD engineers, can you ask them to release a new soft like the "pixel clock patcher" https://www.monitortests.com/atikmdag-patcher-1.4.6.zip (source code: https://www.monitortests.com/end/atikmdag-patcher-1.4.6-src.zip) ? Since the windows 10 build 1803, it's impossible to use normally an IPG as primary display while computing with bios moded cards.
Trying to display with the 2400G vega cores full power while computing with the bios moded RX570s is not possible, windows always gets a code 43 error. So I have to shutdown the Ryzen 2400G Vega cores or get them recognized as a "basic microsoft video card".

I bought the 2400G especially to have a good display while mining, And I can't do that.

Thanks in advance

[zawawa]

I'd prefer a devfee and more frequent updates. But I guess you'll get more money with your new ways, so I wish you the very best.
32,5MH/s with a 480 is a huge performance ! What's the power draw please ? I think my best 570 gets 31,5MH/s while drawing 120W.

BTW, since you talk with AMD engineers, can you ask them to release a new soft like the "pixel clock patcher" https://www.monitortests.com/atikmdag-patcher-1.4.6.zip (source code: https://www.monitortests.com/end/atikmdag-patcher-1.4.6-src.zip) ? Since the windows 10 build 1803, it's impossible to use normally an IPG as primary display while computing with bios moded cards.
Trying to display with the 2400G vega cores full power while computing with the bios moded RX570s is not possible, windows always gets a code 43 error. So I have to shutdown the Ryzen 2400G Vega cores or get them recognized as a "basic microsoft video card".

I bought the 2400G especially to have a good display while mining, And I can't do that.

Thanks in advance

The power draw was 130-something without fine-tuning. I am under the impression that there is not much you can do with Ethash kernels once you achieve the 64 VGPR mark with an appropriate memory access pattern. Now that my kernel is open-sourced, I hope things would be fairer for miners.

As for switching business models, I would rather stick exclusively to open-source development myself, but I would have to sacrifice all of my waking hours that way, and my wife is not particularly happy about that. Hope you understand.

I doubt AMD would publish a hack like that officially. I will ask them if I have a chance.

[Iamtutut]

I wish you the very best with your new project(s).

Regarding the hack, it can be done unofficially on github

[UnclWish]

I didn't understand all. Will you make 1.3.9 version for us? Please finish it. Add nicehash heavy and x16r algos.