Claymore's Dual Ethereum AMD+NVIDIA GPU Miner v15.0 (Windows/Linux)

[Claymore]

v11.5:

- added "-oldkernels" option (AMD GPUs only), you can try "-oldkernels 1" to use old-style GPU kernels from v10, they can be more stable for hard OC and custom BIOSes.
- a bit reduced stale shares for Nvidia GPUs.

[1715607962]

1715607962

[1715607962]

1715607962

[1715607962]

1715607962

[1715607962]

1715607962

[1715607962]

1715607962

[Claymore]

Let me explain: some people apply OC and memory straps to get best hashrate. They spend a lot of time to find best values, and finally it works. But it works only for some specific GPU kernel, it may fail as soon as you try to start some different miner or even same miner but with updated GPU kernel. After that you can blame the miner of course, but the reason is in your settings, you need to readjust them for new kernel.
I added "-oldkernels 1" option so now you can use old kernels in latest version, hope it will help.

[Digital_Seytan]

Just ran a 6 hour experiment on the exact same 1060 rigs.

Average shares per hour:

Claymore 11.5: 152
Phoenix 2.6: 174

Extrapolated over the month, Phoenix provides 0.08 more ETH over Claymore.

phoenixminer 2.7c crashes every so often for me, my virus program deleted the executable so I've gone back to 2.6.

[Raja_MBZ]

Let me explain: some people apply OC and memory straps to get best hashrate. They spend a lot of time to find best values, and finally it works. But it works only for some specific GPU kernel, it may fail as soon as you try to start some different miner or even same miner but with updated GPU kernel. After that you can blame the miner of course, but the reason is in your settings, you need to readjust them for new kernel.
I added "-oldkernels 1" option so now you can use old kernels in latest version, hope it will help.

Exactly. With each newer version, you've to find out the best overclocking/etc. values once again. I figured it out and now loving your newer version of miner, with better hash-rate & stability (I just upgraded to your v11.5 from v7, that's an upgrade after a long time).

[Marvell2]

I see some people have problems with latest versions because they use hard OC and/or custom bioses created for old versions. They work fine with old GPU kernels, but I modified kernels in v11.2 and you need to readjust your settings.
For those who cannot do it (or don't want to do it) I can revert some changes and release a version with "old-style" binaries. So I will release v11.5 within a few hours and it will be the only change in this version, so don't upgrade if v11.4 works fine for you.

does it mean 11.5 will have issue with stale shares again? I like 3% stale instead of 8% for 11.1, but a couple of rigs began very rare crashing since updating to 11.2.

No, stale shares will be same as for 11.4 (i.e. almost zero) because it is not related to GPU kernels directly.

stale shares have never been near zero with claymore, I get 3% stale shares
on ethermine no matter what version i use

[Prelude]

v11.5:

- added "-oldkernels" option (AMD GPUs only), you can try "-oldkernels 1" to use old-style GPU kernels from v10, they can be more stable for hard OC and custom BIOSes.
- a bit reduced stale shares for Nvidia GPUs.

I've tested the old kernel on my Hawaii rigs for the last 16 hours or so, and it solves the issue introduced by 11.x with Hawaii GPUs. 

[big al]

Exactly. With each newer version, you've to find out the best overclocking/etc. values once again. I figured it out and now loving your newer version of miner, with better hash-rate & stability (I just upgraded to your v11.5 from v7, that's an upgrade after a long time).

When you toss in ram straps then it gets really complicated. I'm guessing the guys will have to come up with ones for the new kernel

[atecubanoS]

I see some people have problems with latest versions because they use hard OC and/or custom bioses created for old versions. They work fine with old GPU kernels, but I modified kernels in v11.2 and you need to readjust your settings.
For those who cannot do it (or don't want to do it) I can revert some changes and release a version with "old-style" binaries. So I will release v11.5 within a few hours and it will be the only change in this version, so don't upgrade if v11.4 works fine for you.

So what would work better on an rx580:

- stock bios with 11.4
- modified bios with 11.5

or is it just a matter of testing? Is it no longer worth strapping the ram?

The 11.5 should be slightly slower than 11.4

Why? How can you state that?

Should I stick with v11.4 with my Nvidia rig?

[peterboy1]

v11.5:

- added "-oldkernels" option (AMD GPUs only), you can try "-oldkernels 1" to use old-style GPU kernels from v10, they can be more stable for hard OC and custom BIOSes.
- a bit reduced stale shares for Nvidia GPUs.

is there a conflict if i have -asm 2 along with -oldkernels 1?

[Claymore]

v11.5:

- added "-oldkernels" option (AMD GPUs only), you can try "-oldkernels 1" to use old-style GPU kernels from v10, they can be more stable for hard OC and custom BIOSes.
- a bit reduced stale shares for Nvidia GPUs.

is there a conflict if i have -asm 2 along with -oldkernels 1?

No conflicts. -oldkernels selects source of binaries, -asm selects binary.

[Claymore]

Should I stick with v11.4 with my Nvidia rig?

If you have enough rigs to see good average stats you will notice about 1% less stale shares in v11.5 on nvidia.

[leomcr]

I'm still on the 10.6 version, which gives me over 196mhs on my 13 rx560 cards (using linux/hiveos).
From the 11.x version, I lost about 0.800 per card, which means that I lost a big 10mhs for my rig.
Using the 11.4 version, things got a little bit better, but still under 10.6 version. With this version I got 194mhs.
Not tested yet the 11.5 version, but probably the -oldkernels switch will bring me back the best performance I had with claymore.

[UnixBird]

Please address this.

Bug report:
Somewhere around version v11.0, the miner's reboot mechanism has stopped to work on Linux. It just idles doing nothing.
miner writes to log:
WATCHDOG: GPU 5 hangs in OpenCL call, exit
Rebooting
Quit, please wait...

It does not call reboot.sh.
Kernel writes "INFO: task ethdcrminer64:1054 blocked..."
Syslog writes "hung task".
The only way of recovery is to reboot the rig.

Observation:
Somewhere between v10.1 and v10.6 the percentage of stale hashes increased from 3% to 8% on ethermine.

[peterboy1]

v11.5:

- added "-oldkernels" option (AMD GPUs only), you can try "-oldkernels 1" to use old-style GPU kernels from v10, they can be more stable for hard OC and custom BIOSes.
- a bit reduced stale shares for Nvidia GPUs.

is there a conflict if i have -asm 2 along with -oldkernels 1?

No conflicts. -oldkernels selects source of binaries, -asm selects binary.

i noticed. thanks.

gosh, i was spoiled by higher hashrate already with the new kernels. using old kernels, hashrate swings are noticeable (i dont know if its because my oc has been messed up) and of course the little decrease in hashrate.

guess what, i came back deleting -oldkernels parameter and enjoy higher hashrate, i'll just readjust mclocks whenever incorrect share is logged.

cant go back to old hashrates/kernels anymore.  

atleast we now have options to revert back to old kernels without downgrading.

thanks sir

[j2james]

If you have enough rigs to see good average stats you will notice about 1% less stale shares in v11.5 on nvidia.

Can you explain how CUDA runtime version affects mining? My driver CUDA version is 9.1. Should 9.1 runtime be better for me then 8.0 runtime?

[Pirator]

I'm still on the 10.6 version, which gives me over 196mhs on my 13 rx560 cards (using linux/hiveos).
From the 11.x version, I lost about 0.800 per card, which means that I lost a big 10mhs for my rig.
Using the 11.4 version, things got a little bit better, but still under 10.6 version. With this version I got 194mhs.
Not tested yet the 11.5 version, but probably the -oldkernels switch will bring me back the best performance I had with claymore.

that is right. But the stale share is less. So you might end more valid shares.

[Claymore]

I'm still on the 10.6 version, which gives me over 196mhs on my 13 rx560 cards (using linux/hiveos).
From the 11.x version, I lost about 0.800 per card, which means that I lost a big 10mhs for my rig.
Using the 11.4 version, things got a little bit better, but still under 10.6 version. With this version I got 194mhs.
Not tested yet the 11.5 version, but probably the -oldkernels switch will bring me back the best performance I had with claymore.

that is right. But the stale share is less. So you might end more valid shares.

I have similar hashrate for v10.6 and v11.5, but you can try "-oldkernels 1" if you see bad speed. Also you can try "-asm 2" and "z" key to find best -dcri for ETH-only mode.

[big al]

i noticed. thanks.

gosh, i was spoiled by higher hashrate already with the new kernels. using old kernels, hashrate swings are noticeable (i dont know if its because my oc has been messed up) and of course the little decrease in hashrate.

guess what, i came back deleting -oldkernels parameter and enjoy higher hashrate, i'll just readjust mclocks whenever incorrect share is logged.

cant go back to old hashrates/kernels anymore.  

atleast we now have options to revert back to old kernels without downgrading.

thanks sir

You may not get improvement with lower clocks. Do you have modified ram timing straps? When I get home I will compare stock bios with new kernel vs modified bios with old kernel in 11.5

[peterboy1]

i noticed. thanks.

gosh, i was spoiled by higher hashrate already with the new kernels. using old kernels, hashrate swings are noticeable (i dont know if its because my oc has been messed up) and of course the little decrease in hashrate.

guess what, i came back deleting -oldkernels parameter and enjoy higher hashrate, i'll just readjust mclocks whenever incorrect share is logged.

cant go back to old hashrates/kernels anymore.  

atleast we now have options to revert back to old kernels without downgrading.

thanks sir

You may not get improvement with lower clocks. Do you have modified ram timing straps? When I get home I will compare stock bios with new kernel vs modified bios with old kernel in 11.5

really? isn't it the reported hashrate is an indication of speed? as long as you get no hwinfo errors and incorrect shares?

[dzap]

I have the same amount of ETH incorrect shares dual mining with v11.4 as before over two rigs, at most 1 or 2 per 1000 shares on Nanopool. If you're getting more incorrect shares with v11.4, if probably means your settings were too high to begin with or a Bios mod with too aggressive timings. Use HwInfo to check for memory errors and adjust the memory clock down by 50 MHz until there are no more errors.

Whats ur 2nd coin you mine using keccak if dont mind me asking? I was mining Pascal but it is going down the drain...