Excavator 1.4.4a: Testusers wanted

UselessGuru commented 6 years ago

Testusers wanted!

In an attempt to create AMD only and NVIDIA only miners I rewrote the Excavator miner files. This will fix https://github.com/MultiPoolMiner/MultiPoolMiner/issues/1075 Since I do not own AMD hardware I need some assistance.

How to do do this quick test:

You need to have a mixed rig with AMD and NVIDIA cards installed. Then

Manually update Excavator to the latest version
Delete all Stats\Excavator*.txt files (this will trigger benchmarking)
Move ALL miners files (except ExcavatorAmd1.ps1 from the zip file) to some other directory
Run MPM; It should begin mining on all AMD cards (check the miner screen), but it should NOT use any NVIDIA card. If this is true: Horray, step 1 done
Next: Stop MPM
Replace ExcavatorAmd1.ps1 with ExcavatorNvidia1.ps1 from the zip file
Start MPM; It should begin mining on all NVIDIA cards (check the miner screen), but it should NOT use any AMD card. If this is true: Horray, step 2 done. Mission completed!
Restore all other miner files (where did you put them in step 3???)
Report findings back here - thank you!

Manual Update of Excavator Version 1.4.4 alpha NVIDIA: https://github.com/nicehash/excavator/releases/tag/v1.4.4a

You have to update manually, because Excavator is a proprietary software by NiceHash and has a special EULA.

Unfortunately I cannot verify this behavior from the content in the log files - so uploading logs won't help me :-(

evanvz commented 6 years ago

Hi @UselessGuru

Not sure if you're still looking for this info or what info you require, but here we go.

Data.zip

Braintelligence commented 6 years ago

Ran a test on my 13x 1070 Ti rig now with these new .ps files and Excavator 1.4.1a Cuda 9.1 and 390.65 nvidia drivers:

All algos:

All 13 GPUs were recognized

Blake2s

52,28GH/s with EN1, significantly higher than TPruvot (therefore best)
EN2 seems to have failed
Retry with EN2 and interval 300, now it seems to not crash. Benchmarked with 58,23GH/s

Lbry

Silently crashes after a few seconds, not sure why. MPM tried it 4 times until declaring as failed.
3,83GH/s with EN1, just a tad worse than TPruvot
Retry Lbry with EN2 on interval 300, still silently crash after a few seconds, removed it from ps file

Neoscrypt

10,59MH/s with EN1, significantly lower than Klaust or PalginNvidia (very bad)
11,34MH/s with EN2, closer but still significantly lower than Klaust or PalginNvidia which are both above 13MH/s for me
I will try the EN2 one again with an interval of 300, since I heard it needs a bigger ramp-up time
Retry EN2 with interval 300 aaaaaaaand it crashed. sigh

Nist5

658,13MH/s with ExcavatorNvidia2, significantly higher than TPruvot (therefore best for me)
595,95 MH/s with ExcavatorNvidia1 (worse, don't know what this means)

Keccak

12,03 GH/s with ExcavatorNvidia2, significantly higher than Tpruvot
10,40 GH/s with EN1
MPM declared Keccak as failed, didn't catch why
Retried EN2 with interval 300, crashed as expected... MPM needed about 4 minutes to recover and start another algo

Equihash

MPM declared Equihash as failed at first, didn't catch why
6,45KH/s with EN2, significantly higher than DSTM (therefore best for me)
5,55KH/s with EN1, less than DSTM
Retried EN2 with interval 300, but it crashed... damnit! Waited over 5 minutes for MPM to react and either start the next miner or restart Excavator but nothing happened. Restarted start.bat
Another try again with interval 300 and very fast crash. Looks like I can't use EN2 for Equihash, which is a real bummer...

Ethash

Silently crashes after a few seconds. Same as Lbry.
348,49 MH/s with EN1, this is a lot worse than even Claymore Dual Mining ETH-performance
Retry EN2 Ethash with interval 300, crashes instantly like before, removing Ethash entirely from Excavator

Lyra2Re2

531,52MH/s with ExcavatorNvidia1, just a bit higher than CcminerLyra2re2 (therefore best for me)
EN2 seems to crash, MPM tries several times to invoke it but it crashes after a few seconds. This keeps on until MPM declares it as failed.
Now on sixth time of EN2 it didn't crash as soon, produced a hashpower of 583MH/s in the console but then crashed after about a minute. (Looks to me like the json config is being adapted on every try?)
7th time also crashed
8th time same as sixth time but it worked it's way to almost 600MH/s, still crash
9th time we reach 602MH/s, still crash
10th time (how often will this go on?) it crashes again (I really wish I had logs at this point...)
Didn't let it run for another time, restarted start.bat and it didn't even bother running lyra2re2 at this point

Pascal

15,10GH/s with ExcavatorNvidia2, no other miner did work with Single Pascal until yet, so that's a good thing I guess
EN1 kept running for several minutes in the benchmark, showed about 14,37GH/s - had to restart start.bat because it wouldn't kill the process by itself

Braintelligence commented 6 years ago

@UselessGuru Would love to hear your advice on how to find out why EN2 keeps crashing for so many algos.

UselessGuru commented 6 years ago

@evanvz

Thank you for the logs. But what I would like to know cannot be logged :-(

I have edited the test instructions to clarify the things to watch.

evanvz commented 6 years ago

Hi @UselessGuru

Noted.

Seems @Braintelligence it a betetr job.

Just trying to give my 2c helping you with your great program.

Cheers.

Braintelligence commented 6 years ago

@evanvz I'm sorry, I don't understand what you wrote when you referred to me ☹️. What I wrote sadly won't help @UselessGuru at all, because I don't have AMD cards on that rig.

I hope that my observations can help others and maybe @UselessGuru might drop a hint on what I could try to optimize my rig and make EN2 work, since it looks promising for Equihash.

UselessGuru commented 6 years ago

@Braintelligence

Would love to hear your advice on how to find out why EN2 keeps crashing for so many algos.

EN2 is running two treads per miner = double the amount of ram needed. This works for some algorithms, but not for others.

make EN2 work, since it looks promising for Equihash

No chance. Excavator does not have enough memory on the graphics card (happens for me too)

Braintelligence commented 6 years ago

@UselessGuru Are we talking VRAM (I mean: memory on GPU) or RAM (system memory on the mainboard)?

fonyo commented 6 years ago

@UselessGuru Can do some testing for you on Friday.

UselessGuru commented 6 years ago

Are we talking VRAM (I mean: memory on GPU)

VRAM assume (I have 32GB of RAM in my box, 3 GPUs and are still getting the same errors)

Braintelligence commented 6 years ago

@UselessGuru I see, thanks for clarifying! So it could work on GTX 1080Ti with their 11GB VRAM?

UselessGuru commented 6 years ago

So it could work on GTX 1080Ti with their 11GB VRAM?

Nope, some algogorithm use a percentage of available vram. If they take more than 50% then 2x 50% -> bang!

Braintelligence commented 6 years ago

@UselessGuru Makes sense, so this only works for some algorithms I guess. Crap. 😸

UselessGuru commented 6 years ago

@fonyo

Can do some testing for you on Friday.

Make sure you test the updated version

geckah commented 6 years ago

Tested 1.4.1a earlier with no crashes on my 3x1070ti,3x1080,1x1070 rig.

1.4.2a insta crashes. Will test more

added the direct path in the miner file for mpm to download and 1.4.2a is benchmarking now

1.4.2a is only using 2 of my 3 gpus (1x1080, 2x1070). Tried ccminer, and all 3 gpus were being used. Checked 1.4.1a again and found that only 2 gpus were used too

UselessGuru commented 6 years ago

added the direct path in the miner file for mpm to download and 1.4.2a is benchmarking now

You are supposed to do a manual DL and install. ;-) Due to EULA restrictions we are not allowed to support automatic DL.

geckah commented 6 years ago

I tried the manual way. Must have fat fingers, couldnt get it to run.

Am at a lost as to why only 2 out of 3 gpus are being used. Excavator picks up all 3 cards when initialising

UselessGuru commented 6 years ago

Please post the command line & json file

geckah commented 6 years ago

Got this from the excavator dev:

Ok, I see what's the problem. Wrong device ids in worker.add methods.

This

"method": "worker.add","params": ["0","1"] "method": "worker.add","params": ["0","2"] "method": "worker.add","params": ["0","3"] Should be:

"method": "worker.add","params": ["0","0"] "method": "worker.add","params": ["0","1"] "method": "worker.add","params": ["0","2"] Device ID of your first GPU is 0 not 1 and so on..

You're adding worker to a device with id 3 but that device doesn't exist, in the mean time your first GPU doesn't get a worker, that's the reason it's idle.

I will manually edit the json file and check

UselessGuru commented 6 years ago

@geckah

Could you please post the output of the following command: [OpenCl.Platform]::GetPlatformIDs() | ForEach-Object {[OpenCl.Device]::GetDeviceIDs($_, [OpenCl.DeviceType]::All)} Also post an example of the created json files. Thank you.

geckah commented 6 years ago

Here is the json file NiceHash_Nist5_1_Nvidia.txt

I'm sorry, but how do i get to output that command?

UselessGuru commented 6 years ago

I'm sorry, but how do i get to output that command?

Open a powershell session (pwsh.exe)
type '[OpenCl.Platform]::GetPlatformIDs() | ForEach-Object {[OpenCl.Device]::GetDeviceIDs($_, [OpenCl.DeviceType]::All)} > somefile.name' and press enter
upload somefile.name (where somefile.name) is a valid filename and path of your choice

geckah commented 6 years ago

getting an error with that command:

Unable to find type [OpenCl.Platform]. At line:1 char:1

[OpenCl.Platform]::GetPlatformIDs() | ForEach-Object {[OpenCl.Device] ...


+ CategoryInfo          : InvalidOperation: (OpenCl.Platform:TypeName) [], RuntimeException
+ FullyQualifiedErrorId : TypeNotFound

UselessGuru commented 6 years ago

OK, run pwsh from the MPM directory then type 'using module .\Include.psm1' and try again.

geckah commented 6 years ago

here you go somefile.txt

UselessGuru commented 6 years ago

somefile.txt

Thank you - I think I see the problem. Your CPU is also a GPU. I have updated the miner files: ExcavatorMiners.zip Can you update and test - thank you!

geckah commented 6 years ago

Thank you - I think I see the problem. Your CPU is also a GPU. I have updated the miner files: ExcavatorMiners.zip Can you update and test - thank you!

Sorry for the late reply. Testing now, all cards are being utilised now. Will update on the results then move on to test rigs with more gpus

geckah commented 6 years ago

nist5 alexis78cuda9 208MH/s EN2 157.59MH/s EN1 154.27MH/s

neoscrypt EN1 3.55MH/s klaust 3.18MH/s EN2 fails to run, couldn't catch the error

lyra2v2 EN2 135MH/s EN1 132MH/s polytimos 131.95MH/s

blake2s polytimos 13.02gh/s EN1 12.52gh/s EN2 12.4gh/s

UselessGuru commented 6 years ago

Testing now, all cards are being utilised now

Thank you. Good to hear!

EN2 fails to run, couldn't catch the error

That is somehow to be expected, some miners/algos eat up too much VRAM. Let MPM fail during benchmark, it will report 0 hash rate and will never try this miner/algo combination anymore.

UselessGuru commented 6 years ago

Obsolete with https://github.com/MultiPoolMiner/MultiPoolMiner/commit/a0857c0757ac2577df7c2631e31d746383d5a442

MultiPoolMiner / MultiPoolMiner