Topic: ethminer-0.9.41-genoil-1.1 - page 93. (Read 397360 times)

https://docs.google.com/document/d/17zHy1SUlhgtCMbypO8cHgpWH73V5iUQKk_0rWvMqSNs/mobilebasic?hl=en_US&pli=1

There used to be a web page with a more straightforward explanation but it recently became unavailable.

Are there any specs for the stratum implementation ? I'll add it to eth.suprnova.cc too if possible

Some developer said before that it is quite difficult to squeeze the last few percentage of performance out.

Genoil, so are there any performance improvements with this miner from the original for AMD cards?

Nice! Thanks for posting this.

While waiting for your great next releases a make a bat file with loop. Works fine during a day )) But a lot of empty cmd windows after it ))))

run_loop.bat

:loop1
start coinotron.bat
timeout /T 3600

taskkill /F /IM ethminer.exe

timeout /T 25

goto loop1


coinotron.bat

setx GPU_FORCE_64BIT_PTR 0
setx GPU_MAX_HEAP_SIZE 100
setx GPU_USE_SYNC_OBJECTS 1
setx GPU_MAX_ALLOC_PERCENT 100

ethminer.exe -G -S coinotron.com:3344 -O user.worker:pass --cl-local-work 128 --cl-global-work 16384

Is the mining speed the same as the official 0.9.41 miner? Can I replace the QT miner with this miner on the Etherpool?

Genoil, thx 

When set --farm-recheck 3000.

Speed stable on 22 mh\s.... so it`s a real "speed" ?

Why pool recomend --farm-recheck 200 ? 

The fluctuation is caused by the way hashrate is calculated. Observe the image below.



The bar with squares represents your GPU miner thread. Each square is a GPU kernel run. Within each run it calculates --cl-global-work * --cl-local-work hashes (i.e 64 * 65536 = 4194304). The yellow bar represents the the getWork thread, which fetches new mining jobs from geth or pool and also displays hashrate. Most of the time it is a sleeping thread, with a period of --farm-recheck (default 500ms). So once in 500ms, it checks the amount of hashes the miner thread has done. But the miner thread only updates its total after having completed its batch. So the red squares aren't counted. In the sketch I quickly drafted there's only one interval with 3 runs and the rest with 4, but I hope you get the idea.

You can get a more stable hashrate figure by increasing the value of --farm-recheck (i.e. 3000), but don't do this if you're using getwork (-F flag), because then you'll also only get new jobs every 3 seconds. With the stratum miner it doesn't matter, it gets work pushed towards the miners in a separate thread. Another option is lowering --cl-global-work, but that leads to more CPU overhead, which is also not what you want.   

Radeon 7970. Why Mh\s so diff. ? 

What real Mh\s i Have ?

Well those are 2 different exceptions that shouldn't happen. I'll add a deadline timer of max. 30 seconds (~twice the blocktime) of working on the same job to band-aid the first one. The 2nd exception indicates a disconnect, after which the GPUs should stop working. There should be an automatic reconnect though. Not sure why that didn't happen, I'll see if I can take a different approach to that.

Thanks for your feedback. 

Hi, Genoil! Great work!
Just tried your stratum mining without proxy (beta 1.0.4b3) with coinotron.
First time after 2.5 hours of succesful work it stops to get new work and submit shares but gpu load was 100% and it was working on one old job I think.
After miner restart it was enable to work ok during next 2 hours but then miner stops - no gpu load, no new text output in console.

So I think the stratum support is in very early beta stage and not ready for release.


I used your 1.0.3 miner with dwarf proxy without any problems before so I think it is not only my setup issue.
I can provide more details if you want ...

What is ethminer-0.9.41-genoil-1.1.x?

Formerly known as Genoil's CUDA miner, ethminer-0.9.41-genoil-1.x.x is a fork of the stock ethminer version 0.9.41. While native CUDA support is its most significant difference, I wanted to bring some newly developed features to your attention, that are also interesting for AMD miners:

• on-GPU DAG generation
• stratum mining without proxy
• failover farm URL (getwork+stratum)
• realistic benchmarking against arbitrary epoch/DAG/blocknumber
• OpenCL devices picking
• cleaner and more useful log output

Source code: https://github.com/Genoil/cpp-ethereum
Releases: https://github.com/Genoil/cpp-ethereum/tree/master/releases


F.A.Q

1. Why is my hashrate with Nvidia cards on Windows 10 so low?
The new WDDM 2.0 driver on Windows 10 uses a different way of addressing the GPU. This is good for a lot of things, but not for ETH mining. There is a way of mining ETH at Win7/8/Linux speeds on Win10, by downgrading the GPU driver to a Win7 one (i.e 347.52) and using a special build which I will post soon. Credits to @Mo35 from the ETH forums for figuring that out.

2. And what about the GTX750Ti?
Unfortunately the issue is a bit more serious on the GTX750Ti, already causing suboptimal performance on Win7 and Linux. Apparently about 5MH/s can still be reached on Linux, which, depending on ETH price, could still be profitable, considering the extremely low power draw.

3. Are AMD cards also affected by this issue?
Only GCN 1.0 cards (HD 78x0, 79x0, R9 270, R9 280) and in a different way. While Nvidia cards have thresholds (i.e 2GB for 9x0 / Win7) of the DAG file size after which performance will drop steeply, on those AMD cards the hashrate also drops with increasing DAG size, but more in a linear pattern.

4. Can I mine ETH with my 2GB GPU?
2GB should be sufficient until Ethereum switches to PoS. I don't keep an exact list of all supported GPU's, but generally speaking the following cards should be ok:
AMD HD78xx, HD79xx, R9 2xx, R9 3xx, Fury.
Nvidia Geforce 5x0, 6x0, 7x0, 8x0, 9x0, TITAN
Quadro, Tesla & FirePro's with similar silicon should be fine too.

5. Can I buy a private kernel from you that hashes faster?
No.

6. What are the optimal launch parameters?
The default parameters are fine in most scenario's (CUDA). For OpenCL it varies a bit more. Just play around with the numbers and use powers of 2. GPU's like powers of 2.

7. Is your miner faster than the stock miner?
Yes. I don't know how much. It's more or less on par with CDM, depending on who you ask and how you measure.