Pages:
Author

Topic: excavator by NiceHash - multi-algorithm NVIDIA GPU miner [1.5.6a] - page 12. (Read 179232 times)

jr. member
Activity: 132
Merit: 7

what i'm doing wrong?


Please run excavator with -d 0 parameter (excavator.exe -c neo.json -d 0) and post the output here.
jr. member
Activity: 119
Merit: 3
@lyolyalya be sure neo.json is in same place with excavator like location it look like it load default one with what it came program, same be sure neo.json what you create is with that extension & not have txt extension Wink if you create with notepad
sr. member
Activity: 455
Merit: 250
always starting equihash with that json.
Code:
[
{"time":0,"commands":[
{"id":1,"method":"algorithm.add","params":["neoscrypt","yiimp.eu:4233","wallet:x"]}
]},
{"time":3,"commands":[
{"id":1,"method":"worker.add","params":["0","0"]}
]},
{"time":10,"loop":10,"commands":[
{"id":1,"method":"worker.print.speed","params":["0"]},
{"id":1,"method":"algorithm.print.speeds","params":["0"]}
]}
]



what i'm doing wrong?
jr. member
Activity: 132
Merit: 7
Woah guys, i just installed on my 1080Ti rig the latest 390.65 nvidia drivers, and i tried the latest 1.4.1 excavator compiled in cuda9.1, you really did awesome this time guys.

Neoscrypt from 7MH to 8.4MH (6x1080Ti - 150W per card (TDP 60%) - OC  +120/+200)

Keccak from 7GH to 7.5GH (6x1080Ti - 150W per card (TDP 60%) - OC  +120/+200)

going to test out more algos, cards model

EDIT1 : no increase on good old 1070 and 1070Ti, but 1080Ti gain is huge.

Thanks! Smiley Speedups on 1080Ti are quite good! Cheesy

Regarding 1070 cards, there should be some increase as well, but definitely not as big as on 1080Ti.

Here are the results of some quick benchmark on 1070 (stock settings):

Excavator version: 1.4.1a_nvidia
Algorithm: lyra2rev2
GeForce GTX 1070: speed = 42.22 MH/s, 0.00 H/s, power usage = 168.27 W
Algorithm: keccak
GeForce GTX 1070: speed = 803.26 MH/s, 0.00 H/s, power usage = 179.34 W

Excavator version: 1.4.1a_nvidia (CUDA 9.1)
Algorithm: lyra2rev2
GeForce GTX 1070: speed = 42.87 MH/s, 0.00 H/s, power usage = 170.12 W
Algorithm: keccak
GeForce GTX 1070: speed = 854.04 MH/s, 0.00 H/s, power usage = 179.22 W

Remark: Neoscrypt performs worse with CUDA 9.1 on 1070 cards - we already have a fix for it and hopefully a new version will be pushed out this week.
full member
Activity: 224
Merit: 100
CryptoLearner
Woah guys, i just installed on my 1080Ti rig the latest 390.65 nvidia drivers, and i tried the latest 1.4.1 excavator compiled in cuda9.1, you really did awesome this time guys.

Neoscrypt from 7MH to 8.4MH (6x1080Ti - 150W per card (TDP 60%) - OC  +120/+200)

Keccak from 7GH to 7.5GH (6x1080Ti - 150W per card (TDP 60%) - OC  +120/+200)

going to test out more algos, cards model

EDIT1 : no increase on good old 1070 and 1070Ti, but 1080Ti gain is huge.
jr. member
Activity: 132
Merit: 7
Unfortunately, you can't change the intensity/number of blocks for the equihash (the only algorithm where you can't do that).

Is that something that you're planning to add in a future update, or is it something that doesn't really apply to this algo?  Just wanting to know if I should keep an eye out for this, or figure something else out...

No, we're not planning to add it. As you said it can't really be applied to equihash (cause of its implementation) - it's quite unique algorithm. 
full member
Activity: 224
Merit: 100
CryptoLearner
since 1.3.8 i noticed a drop in keccak performance (7.1GH v1.3.7a) vs (6.6GH v1.3.8+) could you take a look ? (6x1080Ti rig)

I'll take a look into this. We haven't made any changes on keccak though.

I finally had some time to look at this.

Platform: Windows
Driver version: 390.65

Excavator version: 1.3.7a_nvidia | 3456 | 2017-11-17 13:26:36
GeForce GTX 1080: speed = 1.00 GH/s, 0.00 H/s, power usage = 179.39 W

Excavator version: 1.4.1a_nvidia | 4380 | 2018-01-22 10:16:39
GeForce GTX 1080: speed = 1.00 GH/s, 0.00 H/s, power usage = 179.11 W

It looks fine. I tested it on 1080 (not 1080 Ti) though, but it shouldn't be a difference.  Which OS are you using?

I use windows 10 enterprise, lemme recheck with 1.4.1 in both cuda 8 and 9

Yeah with 1.4.1 it's back to proper numbers, thanks for publishing this Smiley
hero member
Activity: 687
Merit: 511
Unfortunately, you can't change the intensity/number of blocks for the equihash (the only algorithm where you can't do that).

Is that something that you're planning to add in a future update, or is it something that doesn't really apply to this algo?  Just wanting to know if I should keep an eye out for this, or figure something else out...
jr. member
Activity: 132
Merit: 7
Which algorithm are you running? You could always try to change default startup parameters (e.g. number of blocks) to achieve better speeds.

Equihash is what I'm running - I don't see anything related to number of blocks in your API reference or commandline options, is this something undocumented?  What is the default, and in what way do you recommend changing it to improve performance?

Unfortunately, you can't change the intensity/number of blocks for the equihash (the only algorithm where you can't do that).

When running other algorithms you can change the default parameters (e.g number of blocks, thread per blocks, it depends on the algo - take a look here: https://github.com/nicehash/excavator/tree/master/nvidia ). You may reach higher speeds by experimenting with parameters. Even though we do tune parameters on different cards, your card may still get better results with changed parameters (due to different OC settings).
hero member
Activity: 687
Merit: 511
Which algorithm are you running? You could always try to change default startup parameters (e.g. number of blocks) to achieve better speeds.

Equihash is what I'm running - I don't see anything related to number of blocks in your API reference or commandline options, is this something undocumented?  What is the default, and in what way do you recommend changing it to improve performance?
jr. member
Activity: 132
Merit: 7
I noticed on one of my rigs the utilization percentage next to a couple of the CPU's is under 100% (like ~95%) - it seems to only be on machines with a couple of 1080 TI's on them.  I'm running Linux and all the other machines run solid 100%'s - is there something I should look at or a setting I can tweak to get the extra juice out of it?  Thanks!

Which algorithm are you running? You could always try to change default startup parameters (e.g. number of blocks) to achieve better speeds.
hero member
Activity: 687
Merit: 511
I noticed on one of my rigs the utilization percentage next to a couple of the CPU's is under 100% (like ~95%) - it seems to only be on machines with a couple of 1080 TI's on them.  I'm running Linux and all the other machines run solid 100%'s - is there something I should look at or a setting I can tweak to get the extra juice out of it?  Thanks!
jr. member
Activity: 132
Merit: 7
What about CUDA 9.1 on Linux?  Is the benefit strictly a Windows thing or is a 9.1 build coming in the future?

There will be a CUDA 9.1 build for Linux as well. We need to run some more tests first..
hero member
Activity: 687
Merit: 511
I only read the first three pages, sorry if it was mentioned after.

...why does Nicehash propose a feeless miner?

They don't explicitly say, but I would guess it allows them to help control the user experience (especially in the NiceHash Miner instance), and it builds awareness and presumably good will in the community.  Plus they charge fees on the pool-side, so that's probably more important to them than trying to make yet another fee'd up closed source miner.
hero member
Activity: 687
Merit: 511
What about CUDA 9.1 on Linux?  Is the benefit strictly a Windows thing or is a 9.1 build coming in the future?
newbie
Activity: 182
Merit: 0
I only read the first three pages, sorry if it was mentioned after.

...why does Nicehash propose a feeless miner?
jr. member
Activity: 132
Merit: 7
since 1.3.8 i noticed a drop in keccak performance (7.1GH v1.3.7a) vs (6.6GH v1.3.8+) could you take a look ? (6x1080Ti rig)

I'll take a look into this. We haven't made any changes on keccak though.

I finally had some time to look at this.

Platform: Windows
Driver version: 390.65

Excavator version: 1.3.7a_nvidia | 3456 | 2017-11-17 13:26:36
GeForce GTX 1080: speed = 1.00 GH/s, 0.00 H/s, power usage = 179.39 W

Excavator version: 1.4.1a_nvidia | 4380 | 2018-01-22 10:16:39
GeForce GTX 1080: speed = 1.00 GH/s, 0.00 H/s, power usage = 179.11 W

It looks fine. I tested it on 1080 (not 1080 Ti) though, but it shouldn't be a difference.  Which OS are you using?
jr. member
Activity: 132
Merit: 7
@dropky

We've just pushed out a new version:
https://github.com/nicehash/excavator/releases/tag/v1.4.1a

It includes a fix for issues with the re-connection to the NH stratum servers on Equihash, DaggerHashimoto and CryptoNight.

We uploaded the binary for v1.4.1a compiled with CUDA 9.1.  

Excavator compiled with CUDA 9.1 outperforms the one compiled with CUDA 8.0 on some algorithms.

Benchmark results can be found here:
https://github.com/nicehash/excavator/releases/tag/v1.4.1a

sorry i am confuse what's update ? 2 day agow 1.4.1.a > now you say new update but same version ? what's the catch ? & in other way 1.4.1.a Equihash still crash, stall & personaly i give up let a miner run & need to watch all the time because not know when it will crash is become ...

- excavator_v1.4.1a_NVIDIA_Win64.zip =  8.0, 9.0 ??!?!?!
- excavator_v1.4.1a_NVIDIA_Win64_CUDA_9.1.zip only with 9.1 ?!?!?

anyway hope you will fix it in a future

The difference between the excavator_v1.4.1a_NVIDIA_Win64.zip and excavator_v1.4.1a_NVIDIA_Win64_CUDA_9.1.zip is that the latter is compiled with CUDA 9.1. I uploaded both versions since the excavator compiled with CUDA 9.1 reaches higher hashrates on some algorithms compared to the version compiled with CUDA 8.0. These are just two different binaries for the same version.
jr. member
Activity: 119
Merit: 3
Test Excavator_v1.4.1a_NVIDIA_Win64_CUDA_9.1
Hardware Spec & setting:

GPU 0: Gigabyte Windforce 2 NVIDIA GeForce GTX 1060 3GB (Samsung) dedicate to mining
GPU 1: Gigabyte NVIDIA GeForce GTX 650 2GB only for own use, not for mining
OS: Windows 7 Service Pack 1
Video Driver: 390.65
MSI Afterburn: 4.4.2
Power limit: 72 => ~86 W
Temp limit: 80
Core Clock: +168
Memmory Clock: +650


A. Neoscrypt -  stell need much better tune till hit 700-720 kh/s aka see:
- Excavator 1.4.1.a / 9.1 = 628.654 kH/s
- Excavator 1.4.1.a = 614.909 kH/s
- 628.654 (with 9.1) - 614.909 (with out 9.1) = 13.745 kH/s
- Ccminer KlausT 8.19 -i 16 = 700 kH/s
- Hsrminer - 710 > 720 kH/s
- test run start on 2018-01-24 13:50:00 noo issue soo far beside speed & Web Api report see below

Question what is value or setting for json to set Excavator to run on equivalent of intensity of 16 - 18 ?

Web api report strange value 1.4.1.a with 9.1 & with out 9.1
- GPU Utilization 99%-100% when it suppose to be 100% all the time
- Power usage 79-94 W when it suppose to be min 84 W > max 86 W or in this range

p.s. after 1 hour & half run i see this on pool: a drop on round progress from 0.007% to 0.004% because speed & my 1 cent bet because from what i see miner it not adjusting pool difficulty stay at max 100% time vs Klaus or Hsrminer what they doo soo a lose of 0.003% /round loss but i know it need to spend more time to see if is real or not ...

B. Equihash pass on testing coz today i not have time to watch excavator if it will crash or not. excavator_v1.4.1a_NVIDIA_Win64.zip it was done that in numerous time & for that i give up is like playing rusion rulet now is working fine now it decide to crash with noo particulary reason way frak know ... what i know is i can't let Excavator to mine with out watching what a hell he doing on this algoritm at all Dstm, EWBF, Bminer they work with out a problem in any pool, with out be watch ....
sr. member
Activity: 390
Merit: 250
Could someone give me an example how to use this with suprnova?
Pages:
Jump to: