And the zcash difficulty is increasing rapidly...
|
|
|
Equihash has a Memory leak when running more than 1 thread pr gpu
Fixed in the sp-mod
|
|
|
here it is "best fake optimization in the world" The djezo kernel is full of easy pickings.
|
|
|
Testmining with the new kernel is looking good. Will deploy tomorrow on 100++ rigs...
The code is infected with gpl..
With nicehash $$$ backing and the angry french dude, I don't think I will release it.. I am the best optimizer in the world.
|
|
|
Amph, you are again speculating too much.
Sometimes it's important to ask your boss why you are doing what you are doing. If he can't give you a good enough answer you should change job...
|
|
|
My mod is best on 1060 3gb cards. 300MHASH with 80% tdp and oc. No crash, no bullshit. The 1070 is slow. 430 with 80% tdp.. the gtx 970 260 with 80% tdp
Pascal edition
My 3 card testrig is around 4.4% faster than djezo's excevator. (same config and clocks)
I think ccminer's thread handling is working bether on multicard rigs. With ewfb and the excevator hash is lost when you run with 6 cards.
My mod is running 4 threads pr gpu on the g1840 celeron
edit:
My kernel is performing bad on the pool with ccminer. Nheqminer is working ok.
|
|
|
the point is that he don't need theat money, he probably sitting on millions already, he is making 150k a month for your info
No I am not. I have alot of debt. Mining is risky, you need expensive equipment to mine with a profit. And GPU's are improving faster than moore's law...
|
|
|
not necessarily. maybe he did the same optimisations as you are doing now.
Could be.. but I don't think so. Let's see later.
|
|
|
Everyone wants shit for free. People bitch and moan so it's 'free'. Don't be afraid to charge for your work. Real support doesn't come from people who only give you kind words so they get free things.
SP I'm going to once again suggest using a dev fee as I did multiple times. Even in work that uses other peoples work you can have a dev fee, the source just needs to be provided. At 2% there aren't many people that are going to go out of their way to remove it, especially with readily available binaries. OR add it as a option that's enabled by default and can be disabled with -nofee or something along those lines. You can charge for work that uses others work as long as the source is available.
I will opensource my feecode. Good that you're improving equihash though.
9% faster than the opensource is still a few percent slower than EFWB. When I release it, he will probobly integrate my changes into his fork and continue to be the fastest..
|
|
|
any link for download ? couldn`t find it on github
It's not done yet.
|
|
|
I have already written the code. Epsylon never merge any pullrequest. So I guess I'll have to refork..
|
|
|
In my fork I have implemented a -g parameter to run more than one thread pr gpu, does your fork have a similar option?
Your fork only support 16 gpu threads.
So when I run equihash with 3 threads per card (18 threads) I get a crash.
ccminer (...) -d 0,0,0,1,1,1,2,2,2,3,3,3,4,4,4,5,5,5
can you increase the max_gpu and implement the "g" parameter please .
|
|
|
I am working on tvpruvot / DJezo's opensource Equihash kernel and managed to gain 9%. I am pretty sure my changes are different from excevator and ewfb.. I might opensource my work for free.. My mod works best with 4 threads pr gpu
|
|
|
Hello, I am trying to build a rig with that hardware: Asus PRIME H270-PRO soccket 1151 Celeron g3900, 1151 4GB ram DDR4 2X Cooler master masterwatt lite 700W 5X Sapphire Pulse radeon RX 580 8g oc w/bp (uefi) lite 5X USB 3.0 Pcie PCI-E Express 1x To 16x Extender Riser Card (Version 007). The parts are on the way, i am waiting the risers, ebay order.
I want to know how long this ring will be working without problems (theoretically) ? Presuming 24h/7days?
Thanks
I still have some reference 7970 cards mining z-cash Codenamed Tahiti, the Radeon HD 7900 series was announced on December 22, 2011. Products include the Radeon 7970 GHz Edition, Radeon HD 7970 and Radeon HD 7950. The Radeon HD 7970 features 2048 usable stream cores, (6.5 years old gpu) They produce bether hashrate than the rx 580, but use more power.
|
|
|
They have a cuda core that can do 4x4 matrix multiplication in 1 GPU cycle.
But few cryptocoin algos need fast matrix multiplications.. The only algos I can think of is M7 (cryptonite), primecoin, Nexus (prime solver) ..
|
|
|
Epsylon3 has been busy with alot of nice opensource commit's lately, but Custumer Satisfaction is not his priority right now...
|
|
|
Why do you want excavator? So you can use it for all your next GPL-breaking projects? ^^ You finally realized that ccminer base is shit and completely outdated for modern needs?
I thought that the point of making the excavator was that it doesn't use GPL code...? So if you decide to release it, it's not gpl...
|
|
|
I think you easily know which changes I did, no need to pretend something...
Did you implement the todo: in the code?
|
|
|
Otherwise, people would just sell open work with little or no modifications.
Nobody want's to buy if the speed isn't improved. When you optimize you often you need to try 10 different approaches to gain speed, so a few lines of code can take 100hours to make. because the 9 first attempts didn't work.
|
|
|
|