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

[Annabelle75]

Well, Vega getting closer, at the least cards are recognized, but it can't compile OpenCL:

19:14:49:580   68a6e740   GPU #0: gfx900, 16368 MB available, 64 compute units
19:14:49:580   68a6e740   GPU #0 recognized as Vega
19:14:49:580   68a6e740   POOL/SOLO version
19:14:49:580   68a6e740   b391
19:14:49:580   68a6e740   Platform: Linux
19:14:49:696   68a6e740   start building OpenCL program for GPU 0...
19:14:49:697   68a6e740   done
19:14:49:797   68a6e740   Error: AMD HSA Code Object loading failed: HSA_STATUS_ERROR_INVALID_CODE_OBJECT: The
                                                 code object is invalid.
19:14:49:797   68a6e740   Cannot build OpenCL program for GPU 0
19:14:49:900   68a6e740   GPU #0: algorithm ASM


so it mines along at 0.00.

modinfo amdgpu | more
filename:       /lib/modules/4.9.0-kfd-compute-rocm-rel-1.6-77/kernel/drivers/gpu/drm/amd/amdgpu/amdgpu.ko
version:        17.20.5
....
firmware:       amdgpu/vega10_smc.bin
firmware:       amdgpu/vega10_asd.bin
firmware:       amdgpu/vega10_sos.bin
firmware:       amdgpu/vega10_rlc.bin
firmware:       amdgpu/vega10_mec2.bin
firmware:       amdgpu/vega10_mec.bin
firmware:       amdgpu/vega10_me.bin
firmware:       amdgpu/vega10_pfp.bin
firmware:       amdgpu/vega10_ce.bin
firmware:       amdgpu/vega10_sdma1.bin
firmware:       amdgpu/vega10_sdma.bin
firmware:       amdgpu/vega10_uvd.bin
firmware:       amdgpu/vega10_vce.b

What's interesting is Radeon Open Compute now supports Vega with their in-kernel module (but claymore still hashes at zero)
filename:       /lib/modules/4.11.0-kfd-compute-rocm-rel-1.6-127/kernel/drivers/gpu/drm/amd/amdgpu/amdgpu.ko
license:        GPL and additional rights
description:    AMD GPU
author:         AMD linux driver team

Made sure the HSA Foundation libraries are installed and in $PATH, and the RocM OpenCL implementation is saying 2.0, although it seems to be 1.2 with 2.0 features and not fully compliant.

strace is showing claymore is finding the hsa libraries:
open("/opt/rocm/hsa/lib/libhsa-runtime64.so.1", O_RDONLY|O_CLOEXEC) = 5

If anyone is interested, Vega with Genoil compiled under gcc-6 and/or gcc-7 with all the fun -0MG_FAST -march-onward tricks (PGO'd even) is wobbling around 7-11 Mh/s. Which is up from 6.8, so that's some goods news...

This would probably be better posted to the RocM boards... I should look for a claymore slack I guess... anyway definitely open to suggestions as I'm sure it's something silly I'm too tired now to notice.  If anyone has already fixed this please PM me! :-)

[1714037383]

1714037383

[age4rage]

Just wanted to give a shout out to the DEV and everyone else supporting this awesome miner. This got me into crypto! Wish there was a claymore miner for other algos, but this is awesome!

[NavyCuda]

Well, Vega getting closer, at the least cards are recognized, but it can't compile OpenCL:

19:14:49:580   68a6e740   GPU #0: gfx900, 16368 MB available, 64 compute units
19:14:49:580   68a6e740   GPU #0 recognized as Vega
19:14:49:580   68a6e740   POOL/SOLO version
19:14:49:580   68a6e740   b391
19:14:49:580   68a6e740   Platform: Linux
19:14:49:696   68a6e740   start building OpenCL program for GPU 0...
19:14:49:697   68a6e740   done
19:14:49:797   68a6e740   Error: AMD HSA Code Object loading failed: HSA_STATUS_ERROR_INVALID_CODE_OBJECT: The
                                                 code object is invalid.
19:14:49:797   68a6e740   Cannot build OpenCL program for GPU 0
19:14:49:900   68a6e740   GPU #0: algorithm ASM


so it mines along at 0.00.

modinfo amdgpu | more
filename:       /lib/modules/4.9.0-kfd-compute-rocm-rel-1.6-77/kernel/drivers/gpu/drm/amd/amdgpu/amdgpu.ko
version:        17.20.5
....
firmware:       amdgpu/vega10_smc.bin
firmware:       amdgpu/vega10_asd.bin
firmware:       amdgpu/vega10_sos.bin
firmware:       amdgpu/vega10_rlc.bin
firmware:       amdgpu/vega10_mec2.bin
firmware:       amdgpu/vega10_mec.bin
firmware:       amdgpu/vega10_me.bin
firmware:       amdgpu/vega10_pfp.bin
firmware:       amdgpu/vega10_ce.bin
firmware:       amdgpu/vega10_sdma1.bin
firmware:       amdgpu/vega10_sdma.bin
firmware:       amdgpu/vega10_uvd.bin
firmware:       amdgpu/vega10_vce.b

What's interesting is Radeon Open Compute now supports Vega with their in-kernel module (but claymore still hashes at zero)
filename:       /lib/modules/4.11.0-kfd-compute-rocm-rel-1.6-127/kernel/drivers/gpu/drm/amd/amdgpu/amdgpu.ko
license:        GPL and additional rights
description:    AMD GPU
author:         AMD linux driver team

Made sure the HSA Foundation libraries are installed and in $PATH, and the RocM OpenCL implementation is saying 2.0, although it seems to be 1.2 with 2.0 features and not fully compliant.

strace is showing claymore is finding the hsa libraries:
open("/opt/rocm/hsa/lib/libhsa-runtime64.so.1", O_RDONLY|O_CLOEXEC) = 5

If anyone is interested, Vega with Genoil compiled under gcc-6 and/or gcc-7 with all the fun -0MG_FAST -march-onward tricks (PGO'd even) is wobbling around 7-11 Mh/s. Which is up from 6.8, so that's some goods news...

This would probably be better posted to the RocM boards... I should look for a claymore slack I guess... anyway definitely open to suggestions as I'm sure it's something silly I'm too tired now to notice.  If anyone has already fixed this please PM me! :-)

I read a tweet the other day about this specific problem.  It's being worked on but I don't know the details.  Vega will eventually perform on linux as well as it does on windows.

[Annabelle75]

"I read a tweet the other day about this specific problem.  It's being worked on but I don't know the details.  Vega will eventually perform on linux as well as it does on windows."

I haven't found one... ...if your memory jogs PM me a link please? Perhaps google is too difficult for me, as it already points a search on this error back to this post here, and I don't know the answer yet... Seems like "if you have to ask, you don't understand" circular logic...

the problem seems the amdgn-amdhsa target of RocMs' llvm, but I'm on a super slow internet connection and the miner w/Vega is a G3930 so even slower compiling progress...

Anyway, it's amazing how quickly anything on bitcointalk goes up the google indexing, so while most here aren't interested, when someone smart figures it out and sends me the answer, I'll put reference it here.

FWIW, there's a truly staggering volume of complex code in the ROC project, truly amazing work... ...I'd send them a BeerCoin if I could figure that out ;-)

[NoOneLt]

The other alternative is the open source ethminer program originally made by the eth foundation and improved by Genoil. Its free and exact same speed as claymore.

However it sometimes freezes, and if 1 GPU crashes it stops the entire program so no GPU is hashing until it gets restarted.

Every calculation with my pool/wall measurements i make shows dual more profitable, even with 2+1% +1% DCR fees, at least on my green rigs switching dual/single makes little to no change to ETH MH/s choosing right intensity.

Maybe i am miscalculating though...

[Trimegistus]

Just wanted to give a shout out to the DEV and everyone else supporting this awesome miner. This got me into crypto! Wish there was a claymore miner for other algos, but this is awesome!

Check this out:

https://bitcointalk.org/index.php?topic=638915.new;topicseen#new

and this:

https://bitcointalk.org/index.php?topic=1670733.new;topicseen#new

 

[nub miner]

I'm having issues with Sapphire PULSE RX580 8GB cards.

I'm using these command line options:

Code:

EthDcrMiner64.exe -epool eu1.ethermine.org:4444 -ewal WALLET.WORKER -epsw x -tstop 85 -tt -55 -cclock 1130 -mclock 2150 -cvddc 850 -mvddc 850

For some reason the mclock on all cards keep jumping up and down between 1000 MHz and 2150 MHz about once every second, lowering the mining rate significantly.

If I leave out the -mclock 2150 setting, and set it manually in Wattman, the problem does not occur. I've already tried to increase the Power Limit to +15%, but the behavior still exists.

Anyone?

remove latest amd drive 17.7.2 i believe. instal older...

[trio]

I recently registered at the forum.
Perhaps this question was asked earlier, but he is very interested in me:

1) Is it possible to hide the text for dual mode, for the second coin.
2) It is possible to deduce the speed of a hasht only total (not for each map as it is now).

Thank you for your help.

[notbatman]

Hey Claymore, it looks like the Linux console doesn't support ANSI codes� Also could you add share difficulty info, I tried to solo mine ETC for a week and got nothing but hot air and enough DCR to cover the power bill.

[bardacuda]

Here you go:

http://etcstats.net/

86 Trillion. No sweat.

[Claymore]

Hey Claymore, it looks like the Linux console doesn't support ANSI codes� Also could you add share difficulty info, I tried to solo mine ETC for a week and got nothing but hot air and enough DCR to cover the power bill.

Please explain why do you need the diff for every share found. Also note that you can see current target diff by pressing "s" key. Before answering, make sure that you know how solo mining works. I did not add this info because I try to show only useful information in the miner and I don't know how this information can help in solo mining or pool mining.

[LMN0PY]

Claymore if you see this, someone alerts you to this, OR this has already been hashed out and fixed (my apologies I haven't read through the 700+ pages yet, working on it).

Cards with the 9000gbps OC'd memory, notably the 1060 Gaming X+ (I don't know if AMD cards have the same issue), seem to hash slower than their peers with the non OC'd mem.,

Any reasons to why they are under-performing would be greatly appreciated! Some speculation was that the OC'd mem isn't able to be processed/accessed properly whether by the mining software or OS, etc.
Also with most users they report GPU load at 58%, 60%, 80%, but never more - even when power limit increased to 110%.

Thanks in advance!

[doktor83]

Claymore if you see this, someone alerts you to this, OR this has already been hashed out and fixed (my apologies I haven't read through the 700+ pages yet, working on it).

Cards with the 9000gbps OC'd memory, notably the 1060 Gaming X+ (I don't know if AMD cards have the same issue), seem to hash slower than their peers with the non OC'd mem.,

Any reasons to why they are under-performing would be greatly appreciated! Some speculation was that the OC'd mem isn't able to be processed/accessed properly whether by the mining software or OS, etc.
Also with most users they report GPU load at 58%, 60%, 80%, but never more - even when power limit increased to 110%.

Thanks in advance!

try making the card go into p0 state by turning off p2 state forcing on driver level.

[notbatman]

... QQ

Please explain why do you need the diff for every share found. Also note that you can see current target diff by pressing "s" key. Before answering, make sure that you know how solo mining works. I did not add this info because I try to show only useful information in the miner and I don't know how this information can help in solo mining or pool mining.

Thanks for your reply,

   what I really need is the best share found since the miner started displayed when "s" is pressed so I can know if I can hit the current network difficulty or not solo within moderate timeframe (12h -24h). Displaying the share difficulty along with the share accepted output would also be helpful in determining what the best shares found are with a short timeframe (5m - 15m). There's a sweet spot solo mining but if the bar is set too high I end up in a sour one.

Bottom line is if I can see the difficulty of the shares I'm finding along with the best one found then I can know if I'm able to hit a block or many blocks solo or not.

Regards.

[ritizen]

NVIDIA released CUDA 9.0 recently
and 384.94 is running that driver

Here is the extracted part of CUDA 9.0.103 runtime from CUDA 9 RC, or you can download the full package on official developer website.
https://mega.nz/#F!sZMCyIAS!1AZqkgvW5BUtZJlgQx2waQ


Besides, CUDA 9.0 said "PERFORM 2X FASTER OUT OF THE BOX WITH PASCAL GPUS".
I expect have slight improvement in mining although it may improve the 3D performance only.

[NavyCuda]

"I read a tweet the other day about this specific problem.  It's being worked on but I don't know the details.  Vega will eventually perform on linux as well as it does on windows."

I haven't found one... ...if your memory jogs PM me a link please? Perhaps google is too difficult for me, as it already points a search on this error back to this post here, and I don't know the answer yet... Seems like "if you have to ask, you don't understand" circular logic...

the problem seems the amdgn-amdhsa target of RocMs' llvm, but I'm on a super slow internet connection and the miner w/Vega is a G3930 so even slower compiling progress...

Anyway, it's amazing how quickly anything on bitcointalk goes up the google indexing, so while most here aren't interested, when someone smart figures it out and sends me the answer, I'll put reference it here.

FWIW, there's a truly staggering volume of complex code in the ROC project, truly amazing work... ...I'd send them a BeerCoin if I could figure that out ;-)

Is this relevant to you?

http://www.phoronix.com/scan.php?page=news_item&px=Radeon-Vega-FE-More-OpenCL

[gt_addict]

Now this has probably been answered before but I have just noticed on the Zcash mining version that the gpu lists match the msi afterburner list, is it possible to implement it on this version?

[Ursul0]

Sell your rig, mining is not for you.

You are very helpful. Thank you! Keep doing this!
After this:
****TRASHY HACKER LINK REMOVED****
I am ok again.

You're a newbie with 2 posts recommending a post from a newbie with 13 posts who is offering a hacked version of AMD drivers.

RISK level 1000%

Don't download software from unknown sources. Don't listen to newbies with only a few posts.  Chances are you are going to download a trojan or virus.

There's nothing wrong with the DAG patch that he was referring to. What's your problem? heard word "hacker" too many times for your own good?

[fittsy]

Has anyone confirmed the alleged ridiculous hashrates for Vega?

You are confused. Keep doing your research. The Vega with ridiculous hash rate has not been sold yet.

Yes, VEGA 64 isn't sold until tomorrow, HOWEVER, it was sent out to media last week.  A Media person that is a miner could have some results.  Or they could have sold it to a friend.

On another note, looks like there is a bug in AMD drivers for VEGA on Linux. I expect whatever we see as hash rates next week will just be the beginning.  We'll see improvements to AMD drivers and we'll see people like Claymore who learn to tweak their software to get more out of the VEGA cards.

---

And yes Linux drivers for VEGA are bugged.

[fittsy]

c.
Also with most users they report GPU load at 58%, 60%, 80%, but never more - even when power limit increased to 110%.

Ethash algorithm of encryption is more memory intensive than CPU. That means that Ethereum mining will hit a memory limit before the CPU. 

Other algorithms, like Zcash, which use the Equihash algorithm are more CPU based and less memory based.