Pages:
Author

Topic: excavator by NiceHash - multi-algorithm NVIDIA GPU miner [1.5.6a] - page 62. (Read 179300 times)

sr. member
Activity: 703
Merit: 272
On equihash, what are folks doing to match ewbf speeds or close to them?  I have not tested my 1070's yet.  I ran excavator with 2 threads for each card.


gtx         ewbf     excavator
980         330         300
titan xp   720         650


Thanks

on my 1070
gtx       ewbf      excavator
1070      430        400
full member
Activity: 181
Merit: 100
On equihash, what are folks doing to match ewbf speeds or close to them?  I have not tested my 1070's yet.  I ran excavator with 2 threads for each card.


gtx         ewbf     excavator
980         330         300
titan xp   720         650


Thanks
legendary
Activity: 1134
Merit: 1001
Yes , with new driver , is ok Smiley .
legendary
Activity: 3248
Merit: 1070
I get a bug in the GH/W calculation:

0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
00000000000000000000000000000.#INF00 GH/W
[13:10:07][0x00000f00][info] net | Share #18 accepted


000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000.#INF00 GH/W
3:10:21][0x00000f00][info] net | Share #19 accepted
3:10:29][0x00000f00][info] net | Share #20 accepted
3:10:33][0x000009f4][info] =====> 215.264541 H/s
3:10:33][0x000009f4][info] gpuinfo | GPU#0    TEMP: 56 C      USAGE: 100 %
MCTRL: 74 %   POWER: -2 W
3:10:33][0x000009f4][info] =====> Total power: 0 W    Efficiency: 100000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000

1060 3gb is only doing 215 with 2 threads.

My private fork is doing 275 on the same hardware, clocks and tdp. and 3 threads.



not a bug, you need the last nvidia drivers or 378.xx version to display the wattage correctly
full member
Activity: 400
Merit: 100
Hello,
how can I set intencity?
hero member
Activity: 588
Merit: 501
sp, one post before I said to copy your own nvml or install latest drivers.

Regarding speed; 250 sol for 3gb version with 60% TDP limit.
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
I get a bug in the GH/W calculation:

0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
0000000000000000000000000000000000000000000000000000000000000000000000000000000 0
00000000000000000000000000000.#INF00 GH/W
[13:10:07][0x00000f00][info] net | Share #18 accepted


000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000.#INF00 GH/W
3:10:21][0x00000f00][info] net | Share #19 accepted
3:10:29][0x00000f00][info] net | Share #20 accepted
3:10:33][0x000009f4][info] =====> 215.264541 H/s
3:10:33][0x000009f4][info] gpuinfo | GPU#0    TEMP: 56 C      USAGE: 100 %
MCTRL: 74 %   POWER: -2 W
3:10:33][0x000009f4][info] =====> Total power: 0 W    Efficiency: 100000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000

1060 3gb is only doing 215 with 2 threads.

My private fork is doing 275 on the same hardware, clocks and tdp. and 3 threads.

legendary
Activity: 2408
Merit: 1004
can used these miner with r9 fury card ati???
hero member
Activity: 588
Merit: 501
Good to know regarding nvml... Will work on some solution to fix this issue in future versions... For now, whoever wants to have accurate readings, just install latest NVIDIA drivers or copy your own nvml.dll.
legendary
Activity: 1901
Merit: 1024
yeah new driver fixed the reading

also tested old driver DLL file (on 2nd rig) and now reading are fine too, maybe in next miner version use the driver system DLL?

also speed now match ewbf miner... if you deduct 2% from his fee
hero member
Activity: 588
Merit: 501
Could you try and upgrade drivers to 378.78 ?

We have various cards and all show wattage correctly. Also, the reading fails at function nvmlDeviceGetPowerUsage, for which there is no reason to fail if supplied with proper device handle. And since nvidia-smi does show you a value, your card does support power readings.

EDIT: alternatively, you can put your own nvml.dll into excavator folder (copy it out of program files, nvidia).
legendary
Activity: 1901
Merit: 1024
Quote
21:01:42][0x00000f34][debug] device#9: 213.077631 H/s
[21:01:42][0x00000f34][debug] gpuinfo | GPU#0   PCIBUSID: 0000:01:00.0
[21:01:42][0x00000f34][info] gpuinfo | GPU#0    TEMP: 48 C      USAGE: 100 %    MEMCTRL: 82 %   POWER: -2 W
[21:01:43][0x00000f34][debug] gpuinfo | GPU#1   PCIBUSID: 0000:03:00.0
[21:01:43][0x00000f34][info] gpuinfo | GPU#1    TEMP: 59 C      USAGE: 99 %     MEMCTRL: 81 %   POWER: -2 W
[21:01:43][0x00000f34][debug] gpuinfo | GPU#2   PCIBUSID: 0000:05:00.0
[21:01:43][0x0000161c][debug] device#3 | Found nonce: 070000000000000000000000000000000000000000000079060000 diff=194.967
[21:01:43][0x00000f34][info] gpuinfo | GPU#2    TEMP: 54 C      USAGE: 99 %     MEMCTRL: 79 %   POWER: -2 W
[21:01:43][0x0000161c][debug] net | Sending share #25
[21:01:43][0x00000f34][debug] gpuinfo | GPU#3   PCIBUSID: 0000:06:00.0
[21:01:43][0x00000f34][info] gpuinfo | GPU#3    TEMP: 54 C      USAGE: 98 %     MEMCTRL: 81 %   POWER: -2 W
[21:01:43][0x00000f34][debug] gpuinfo | GPU#4   PCIBUSID: 0000:07:00.0
[21:01:43][0x000013b0][info] net | Share #25 accepted
[21:01:43][0x00000f34][info] gpuinfo | GPU#4    TEMP: 49 C      USAGE: 99 %     MEMCTRL: 80 %   POWER: -2 W
[21:01:43][0x00000f34][info] =====> Total power: 0 W    Efficiency: 1000000000000000000000000000000000000000000000000000000000000000000000000000000 0000000000000000000000000000000000000000000000000000000000000000000000000000000 0000000000000000000000000000000000000000000000000000000000000000000000000000000 0000000000000000000000000000000000000000000000000000000000000000000000000000000 0000000000000000000000000000000000000000000000000000000000000000000000000000000 0000000000000000000000000000000000000000000000000000000000000000000000000000000 0000000000000000000000000000000000000000000000000000000000

Win10 x64 drv 378.57 5x 1070 Gainward phoenix (intel GPU is enebled but no monitor connected at any)
hero member
Activity: 588
Merit: 501
OK, here it is, new version as promised: http://nicehash.com/tools/excavator_v1.1.3a_Win64.zip

Changes:

Quote
v1.1.3a
- slight equihash speed improvement with same TDP
- added support for GTX 1080 Ti
- added option '-od' to set custom overclock delay
- bug fixes and improvements

Regarding overclocking:

Quote
If you overclock, we suggest you to set '-or' to reset clocks after miner quits. This may prevent driver crashes or freezes due to high clocks. With parameter '-od' you can adjust overclock delay and apply new clocks when mining is already happening. This can also prevent crashes or freezes due to high clocks.

For all, that Wattage report isn't working correctly, run under debug mode ('-d 1' or '-f 1') and paste log here. We would like to get to know whether the issue is because of pcibusid string or something else.
hero member
Activity: 588
Merit: 501
Interesting stuff , im actually even more surprised by the running temps, they're really low  Shocked

Yes, temps are not so high if you set low TDP, and not so loud. But with 100% TDP... things really get hot. All algos actually, by stock clocks, go over 100% TDP. Other cards reach more than 100% TDP with overclocking. This card gets quickly power throttled or thermal throttled.
full member
Activity: 224
Merit: 100
CryptoLearner
Interesting stuff , im actually even more surprised by the running temps, they're really low  Shocked
hero member
Activity: 588
Merit: 501
It is an interesting thing we are noticing with these cards. Due to so big memory bandwidth, it is completely irrelevant to overclock memory if you run lower TDP - it does not help, instead, it can actually make less speed. We also noticed that downclocking memory can help slightly and giving some 10-20 sols more.  Shocked

This is probably efficiency world record for mining equihash (4.2 Sol per Watt).
Quote
[17:47:53][0x000026b0][info] =====> 533.249844 H/s
[17:47:53][0x000026b0][debug] device#0: 266.807375 H/s
[17:47:53][0x000026b0][debug] device#1: 266.442468 H/s
[17:47:53][0x000026b0][debug] gpuinfo | GPU#0   PCIBUSID: 0000:04:00.0
[17:47:53][0x000026b0][info] gpuinfo | GPU#0    TEMP: 49 C      USAGE: 99 %     MEMCTRL: 71 %   POWER: 125.271 W
[17:47:53][0x000026b0][info] =====> Total power: 125.271 W      Efficiency: 4.256770 H/W

Here we downclocked memory -400. Notice memory controller load - it is 71% which is below 80%. 80% is where memory starts to be major factor for equihash slowdown.
hero member
Activity: 588
Merit: 501
Limited to 150 W (60% TDP), the efficiency is much better (are there more efficient cards out there?):

Quote
[17:33:45][0x000025fc][info] =====> 590.691606 H/s
[17:33:45][0x000025fc][debug] device#0: 295.852459 H/s
[17:33:45][0x000025fc][debug] device#1: 294.839147 H/s
[17:33:45][0x000025fc][debug] gpuinfo | GPU#0   PCIBUSID: 0000:04:00.0
[17:33:45][0x000025fc][info] gpuinfo | GPU#0    TEMP: 48 C      USAGE: 100 %    MEMCTRL: 62 %   POWER: 151.544 W
[17:33:45][0x000025fc][info] =====> Total power: 151.544 W      Efficiency: 3.897822 H/W
hero member
Activity: 588
Merit: 501
We got 1080 Ti in our labs and are now adding support in excavator and will be releasing new alpha version soon.

Equihash:
Quote
[17:23:30][0x00002638][info] =====> 730.418495 H/s
[17:23:30][0x00002638][debug] device#0: 369.054074 H/s
[17:23:30][0x00002638][debug] device#1: 361.364421 H/s
[17:23:30][0x00002638][debug] gpuinfo | GPU#0   PCIBUSID: 0000:04:00.0
[17:23:30][0x00002638][info] gpuinfo | GPU#0    TEMP: 52 C      USAGE: 99 %     MEMCTRL: 77 %   POWER: 287.721 W
[17:23:30][0x00002638][info] =====> Total power: 287.721 W      Efficiency: 2.538635 H/W

These do run very hot. Need to see performance under water.
legendary
Activity: 3248
Merit: 1070
ah well without tdp limitation i can reach 440 i think with that crazy mem, but with 60% it's simply not feasible

actually my core is reaching 2000 with 60%...nevermind it's 1840 i confused with another algo
sr. member
Activity: 463
Merit: 250
i'm not making at all the 440 reported here, but only 1650(412 each) with 4 gpu, 60% tdp +100 core +600 ram

Make sure you are using at least 2 threads. If only 1, then your GPU is not 100% loaded. 1 thread solution is good if you plan to do other stuff with your card. Imagine it as being an intensity switch.

We reached 440 with 115W TDP limited card that has OC set to 2120 core and 4700 mem. With next version, it is about 445 with these same settings.

Another example is 126W TDP limited card, OCed to 2110 core and 4400 mem (micron with BIOS patch) reaching 440 sol.

You can view how far is your card OCed when you launch excavator with overclocking set.

yeah i'm running it with -ca -ca , three thread result in a crash

how can you consume only 115 watt with core 2120? it's impossible to set the gpu at 2120 core with 60% tdp, without having a crash

what you have set on the core? the only thing may be the mem, your is at +900 mine only 600

I believe they mean the core settings without the TDP limitation. Once limited by TDP it will be downclocked for sure. This would mean roughly +220 OC for the non OC cards?
Pages:
Jump to: