[ANN] coolMiner - Lyra2z Miner for nVidia [Maxwell\Pascal\Volta\RTX]

[thebd]

After 7h:42m of running 1.2, 2 out of 4 rigs crashed when the stratum was changed for dev's fee (Illegal memory access..)

overall gain from 2.2.5 was 3%

going to take 1.3 for a spin and coming back with feedback

What card?

1080ti

[jebidia]

After 7h:42m of running 1.2, 2 out of 4 rigs crashed when the stratum was changed for dev's fee (Illegal memory access..)

overall gain from 2.2.5 was 3%

going to take 1.3 for a spin and coming back with feedback

What card?

1080ti

Strange...I'm seeing 5-6% boost from 2.2.5 with the settings I posted earlier with my 3 rigs of 1080ti's EVGA and Gigabyte.  No crashes yet...What is your core at?  Do you keep it consistent through all the cards?

I do see the stratum change to 2, I'm assuming that's when the devfee goes in.  Seems it should be possible to not have it shift the stratum when switching to dev fee...?   Maybe 1.3?   Haven't tried yet...I use a set stratum.

[thebd]

the 7h average hashrate was 28.7mh/s for 8*1080ti and i was doing 27.8mh/s before with 2.2.5, so overall real increase is 3%

the time needed to reach full speed back from stratum switch is affecting overall performace

i suggest a discord channel to compile some direct feedback... @Ryzen

[Ryzen2700x]

"(Illegal memory access..)" can be caused by many reasons, many cases found when searching ccminer +Illegal memory access, most of the time is stebility issu with OC/power/Drivers etc, maybe you are on the edge of the stability and then the pool switch come and I see this codes are before/after :""// reset default mem offset before idle.." " // reapply mem offset if needed ", I don't know if this related, you should start isolating the problem maybe 1 GPU makes all the trouble.

Also- what did you mean by that:

the 7h average hashrate was 28.7mh/s for 8*1080ti and i was doing 27.8mh/s before with 2.2.5, so overall real increase is 3%

You mean the speed you saw on 7th excepted share? if so it's not accurate, min is 30. and I recommend testing it on version  coolMiner-v1.3  since it's start @full load much faster.

Also l think that on lyra2z coins that the block/hour is high, it's better to use lower intensity (~19) versus 20~21 because it increases the chance that new bock found and the nonce you find after get discarded (hashing takes too long). and coolMiner has lower penalty when lowering the intensity versus ccminer 2.2.5. (i have posted bench results few page back).
Edit- found it:
https://bitcointalk.org/index.php?topic=498930.msg5492268#msg5492268

GPU operate in a "batch" mode.  They start hashing and attempt so many hashes before stopping to return results.  The higher the intensity the more hashes the card attempts in one batch but also the longer the batch takes.  The card only returns results at the end of the batch.  The higher the intensity (and longer the batch) the more shares which are computed but never "delivered" before the block changes and they become stale.  On the other hand the lower the intensity the more of the GPU time spent setting up and taking down batches which means a lower raw hashrate

You get paid for shares (or work) accepted so that is what you should be looking to maximize.  It is possible the accepted rate is still higher even with a higher rejected rate because the gross (raw) hashrate is higher.

Simplified example
100 shares gross w/ 1% rejected = 99 shares accepted
120 shares gross w/ 10% rejected = 102 shares accepted.

The later is more profitable.  Since you get "paid" for shares accepted that is the number you want to maximize.

This is what I have wrote few pages back:

Edit:I have tested it:
-i 16 vs -i 21: ccminer ~8% penalty vs coolMiner ~3% penalty .
-i 17 vs -i 21: ccminer ~4% penalty vs coolMiner ~1.5% penalty .
You can test it yourself.

[thebd]

1.3 is in the grinder right now, will post progress at 12-24-36h

[HashAuger]

Thanks for your efforts on releasing this new miner.  It looks like it an competitive alternative to existing Lyra2z miners. I’m working on including it my software, Hash Auger, but I was disappointed to see that it requires Cuda SDK 9.2.  Considering all the issues surrounding the new Nvidia drivers on Windows, many people are still reluctant to update from 391.35.  For example, 398.11 has a known issue with multiple GPUs and certain motherboard chipsets. Any plans to release a version of CoolMiner compatible with 9.1 or does it depend on features in the new SDK?  Also, since most devs of mining software recommend 32-bit versions, is there any particular reason why CoolMiner is 64-bit?

[Ryzen2700x]

Thank you,
I was asked if I could recompile with cuda 9.1 and I did try but I saw that the some of the performance gains that I had with my mods got lost and the end results wasn't justifying the added fee so there is no benefit for the miners to use it, so it's only cuda9.2.
Regarding 64bit vs 32bit, I know that before cuda 9 the 32bit had better performance but since then 64bit is better or equal, if there is a demand for it I can try to compile 32 bit and test it if the speed is same.
 

[HashAuger]

Thank you,
I was asked if I could recompile with cuda 9.1 and I did try but I saw that the some of the performance gains that I had with my mods got lost and the end results wasn't justifying the added fee so there is no benefit for the miners to use it, so it's only cuda9.2.
Regarding 64bit vs 32bit, I know that before cuda 9 the 32bit had better performance but since then 64bit is better or equal, if there is a demand for it I can try to compile 32 bit and test it if the speed is same.

Thank you for the prompt reply.  Based on what I have read about Cuda 9.2, there are some optimizations not found in 9.1, so it is understandable that it may not run as fast with the earlier SDK. Unfortunately it is a bit late in my own development cycle to upgrade my device drivers to support 9.2, but after my next release I will do so to test integrating your software into Hash Auger.

Regarding 64-bit vs 32-bit, I am not sure if Lyra2z benefits from  the extra CPU precision that the 64-bit code provides. If not, the smaller memory footprint and file size of a 32-bit version may benefit those users with limited RAM.

[CryptoSifus]

Trying on 398.11 and I have one of 6 GPU losing power about once every 7-8hrs.
Not the same one and doesn't happen on 2.2.5

Also w/ v1.3 the load time to take it to max hash is still about 30-40mins.
That's a long time to recover if the miner disconnects or resets.

What times are you guys seeing to get to top avg hash?

[m1hyka]

Im using the 397.31, stable whit this after 7 hours.

[OhMyHolyToken]

What is the most profitable coin for this algo?

Thanks

[soggyminer]

What driver supports this.

Getting a "your system does not support cuda 9.2 api" error when launching the miner.
I downloaded and installed everything but the driver directly from nvidia.
Currently running nvidia 390.77

For some reason my rig crashes when I install the latest drivers. Won't accept more than one GPU when I run them.
Help!

I've been running 398.11 since yesterday without any issues (on multiple multi-gpu rigs). I believe 396.xx is the minimum for CUDA 9.2 support. As HashAuger stated earlier, some people have reported issues with 398.11 and multiple GPUs.

You could always try running DDU before installing the new drivers if nothing else is working: http://www.guru3d.com/files-details/display-driver-uninstaller-download.html

[dkbit98]

I tried with my edited BAT file,
but it is not working, and mining window is closing.

Anyone have the .bat template
or instructions ?

Thank you !

There are many now many coins with Lyra2z algo,
but just few of them are profitable to mine.
Miners can only hope for future price rise

[CryptoSifus]

What driver supports this.

Getting a "your system does not support cuda 9.2 api" error when launching the miner.
I downloaded and installed everything but the driver directly from nvidia.
Currently running nvidia 390.77

For some reason my rig crashes when I install the latest drivers. Won't accept more than one GPU when I run them.
Help!

I've been running 398.11 since yesterday without any issues (on multiple multi-gpu rigs). I believe 396.xx is the minimum for CUDA 9.2 support. As HashAuger stated earlier, some people have reported issues with 398.11 and multiple GPUs.

You could always try running DDU before installing the new drivers if nothing else is working: http://www.guru3d.com/files-details/display-driver-uninstaller-download.html

Thanks for reply, I did try for a fresh install but I can't get the driver to complete install on a 4 card rig.
Crashes 3/4 of the way through and puts the computer into startup repair mode.

If I disconnect 3/4 cards I can get it to load again, but if I add new ones it does the same thing. Crash, reboot, startup repair mode...

[z0mb13m4n]

I'm getting 3.5 at most on 1080 ti's using the OC shown. At 1050mv i can get 3.5mh/s on every 1080ti, hence the one card in the picture showing ~3.45mh/s. All the others are locked at 950. Other cards seem to improve slightly too. This was after running it for over an hour. No clocks over 1975 mhz for core. Mem untouched. P0 state on all cards.

https://pasteboard.co/HoSijoA.jpg

Edit: Was able to get 3.56mh/s on all 4 1080 ti's with clocks pushed to 2050mhz core at 1075mv or lower. 1080 up to 2.5mh/s with 1050mv and 2050mhz core. 1060 to almost 1.3mh/s same settings. 2k on 1070 with core upped to 2150mhz at 1050mv. Was on -i 17

[soggyminer]

I'll go ahead and throw up my numbers as I have been running this since yesterday:

EVGA FTW3 1080 ti - 3760 kH, +115 core (2088)
EVGA FTW 1070 ti - 2448 kH, +150 core (2012)
EVGA SC2 1070 - 1985 KH, +150 core (2062)
EVGA FTW 1060 3GB - 1160 kH , +115 core (2025)

Edit: -i 19

[jebidia]

Edit: Was able to get 3.56mh/s on all 4 1080 ti's with clocks pushed to 2050mhz core at 1075mv or lower. 1080 up to 2.5mh/s with 1050mv and 2050mhz core. 1060 to almost 1.3mh/s same settings. 2k on 1070 with core upped to 2150mhz at 1050mv.

Update your nvidia driver.  -i 20    2012 mhz @ 1050 mv will get you over 3.6 on a 1080ti

[soggyminer]

What driver supports this.

Getting a "your system does not support cuda 9.2 api" error when launching the miner.
I downloaded and installed everything but the driver directly from nvidia.
Currently running nvidia 390.77

For some reason my rig crashes when I install the latest drivers. Won't accept more than one GPU when I run them.
Help!

I've been running 398.11 since yesterday without any issues (on multiple multi-gpu rigs). I believe 396.xx is the minimum for CUDA 9.2 support. As HashAuger stated earlier, some people have reported issues with 398.11 and multiple GPUs.

You could always try running DDU before installing the new drivers if nothing else is working: http://www.guru3d.com/files-details/display-driver-uninstaller-download.html

Thanks for reply, I did try for a fresh install but I can't get the driver to complete install on a 4 card rig.
Crashes 3/4 of the way through and puts the computer into startup repair mode.

If I disconnect 3/4 cards I can get it to load again, but if I add new ones it does the same thing. Crash, reboot, startup repair mode...

Have you tried switching pci-e slots for your risers? I have 1 weird rig that won't boot period if certain pci-e slots are used (1,2,4,6....if i use 3 or 5 the rig doesn't boot)

[z0mb13m4n]

Thanks jebidia

[jebidia]

Thanks jebidia

No worries,.  That help?