Pages:
Author

Topic: HSRMINER Neoscrypt Fork by Justaminer - High Hashrate, API, all GPUs supported! - page 19. (Read 13383 times)

newbie
Activity: 45
Merit: 0
asus mining p104-4g it's not working
 INFO : [02:20:03] : GPU #3: Asus P104-100, flags: 0, 0, 0



can you make it work?
full member
Activity: 206
Merit: 100
So now with forced P0 state and higher mem and core clocks I'm averaging 2030 kH/s at 120% power limit. Note that in order to achieve these rates I still have to specify high cpu priority via "-c 5" even when using the _hp build.

full member
Activity: 206
Merit: 100
I forced p=0 video Precision Xoc utilizing the KBoost button which can be seen highlighted in the benchmarks using your algo.

I didn't do this until half way through bc I wanted to make sure that yours was at p0 when it started and wasn't as concerned with the others.

As for the flags I'm not sure, but I know upon entering benching with your algo p0 was definitely enabled and clock was pinged at 2100.  It only dropped slightly under load and settled at the shown 2050.  Now, this is the same level achieved across the board pretty much even with p0 not enabled.

Remember as well I'm using a 1080ti, not 1070 for these benches. I'm not sure I can p0 it under full load with +400-500mem without fragging instantly or losing performance significantly.  On both my FTW3 air, SC2 Air and SC2 Hybrid, I can get a max of about +115-120core and +300, maybe stretching to +350 but for zero sum gains.

You should just check P state with "nvidia-smi -q -d PERFORMANCE" during mining with ccminer or my fork to be sure.

Also problem is not with CORE clocks (2100 is already very high), but with memory clocks, they drop with P2 state. User SCSI2 also has similar EVGA 1080 Ti SC, you can check his settings for core and mem clocks and see hashrate - rig with 9 cards shows 17+ mh/s,  ~1.9 mh/s per card.


Here are my current settings on that rig. NVsmi reports P2 on all 9 GPUs, so I will try to force P0 and see what it gives... All cards are the original SC Black.

jr. member
Activity: 325
Merit: 2
I forced p=0 video Precision Xoc utilizing the KBoost button which can be seen highlighted in the benchmarks using your algo.

I didn't do this until half way through bc I wanted to make sure that yours was at p0 when it started and wasn't as concerned with the others.

As for the flags I'm not sure, but I know upon entering benching with your algo p0 was definitely enabled and clock was pinged at 2100.  It only dropped slightly under load and settled at the shown 2050.  Now, this is the same level achieved across the board pretty much even with p0 not enabled.

Remember as well I'm using a 1080ti, not 1070 for these benches. I'm not sure I can p0 it under full load with +400-500mem without fragging instantly or losing performance significantly.  On both my FTW3 air, SC2 Air and SC2 Hybrid, I can get a max of about +115-120core and +300, maybe stretching to +350 but for zero sum gains.

You should just check P state with "nvidia-smi -q -d PERFORMANCE" during mining with ccminer or my fork to be sure.

Also problem is not with CORE clocks (2100 is already very high), but with memory clocks, they drop with P2 state. User SCSI2 also has similar EVGA 1080 Ti SC, you can check his settings for core and mem clocks and see hashrate - rig with 9 cards shows 17+ mh/s,  ~1.9 mh/s per card.
newbie
Activity: 5
Merit: 0
I forced p=0 video Precision Xoc utilizing the KBoost button which can be seen highlighted in the benchmarks using your algo.

I didn't do this until half way through bc I wanted to make sure that yours was at p0 when it started and wasn't as concerned with the others.

As for the flags I'm not sure, but I know upon entering benching with your algo p0 was definitely enabled and clock was pinged at 2100.  It only dropped slightly under load and settled at the shown 2050.  Now, this is the same level achieved across the board pretty much even with p0 not enabled.

Remember as well I'm using a 1080ti, not 1070 for these benches. I'm not sure I can p0 it under full load with +400-500mem without fragging instantly or losing performance significantly.  On both my FTW3 air, SC2 Air and SC2 Hybrid, I can get a max of about +115-120core and +300, maybe stretching to +350 but for zero sum gains.
jr. member
Activity: 325
Merit: 2
I tried this and had an entire thesis of the tweaking and benching done, and my rig crashed the second I hit preview on the post.

Not once did it crash during testing, only trying to post.

So now, cliffs:

I tested this in place of the original hsrminer.exe supplied by miningpoolhub's also switching program, then rebenched all Neoscrypt algo's I have installed on my gaming rig (tested on here first vs taking my 2x1080ti 4x1070 rig down to bench for an a couple hours)

System - Ryzen 1800X @ 4.1 GHz (Corsair H115i Pro), Gigabyte AX370-Gaming 5 Mobo, 16GB Corsair LPX DDR-3200, Samsung 960 EVO NVME 120GB
GPU - EVGA 1080Ti SC2 Hybrid - Stock BIOS - Corsair SP120MM Fans Push/Pull Set @ 100% - NVidia Drivers 390.77 (set at prefer max performance):

All benched starting at 110% pwr, +110-115 core, +200 memory

Klaust/Nanashi/Tpruvot Algos - All max between 1,500 - 1,552 kH/s @ 2050 MHz
Just a miner HSRminer MOD - Max 1,507.47 kH/s @ 2050 MHz
Excavator (nicehash default) - Max 1,925.32 kH/s @ 2,037 MHz (Eventually peaking @ 1,985 MHz while hashing)

Nice job man, you've done a lot of work. Sadly, it seems that Hsrminer_neoscrypt_fork (and ccminer too) was running with Nvidia P2 state on, you can see that something went wrong - "GPU#0 ... 1080 Ti, flags: 1, 0, 0" , and it should be "0, 1, 0". P2 state means power level, P0 - highest, it allows maximum performance and overclocking, and P2 is limiting memory clocks especially for 1080 Ti. You can see user reporting same issue - memory clocks drops to low values under load.

From your screenshot I see that you have high core clock - 2050 mhz and rather low memory clock - 5200 mhz. You can see that user SCSI2 has 1080 Ti with ~1800 mhz Core and 5500 mhz Mem and his hashrate with my fork is about 1.9 mh/s. So your problem is that power level set to P2 and therefore you get low memory clocks under load. Now how to deal with it. User dragonmike post a solution.

I will explain the situation. It's not some new thing, it was happening before, for example you can read about it here and here.

So what you should try to do:

1) Run any miner to create load on GPU, go to "C:\Program Files\NVIDIA Corporation\NVSMI" and run next command: nvidia-smi -q -d PERFORMANCE

You will probably see that your GPU is working at P2 state.

You can also check P state by downloading Nvidia Inspector 1.9.7.6 and running it while GPU is under load, it will show P state level in GUI.

2) Download Nvidia PROFILE Inspector 2.1.3.20. This tool will let you to disable P2 state, so your GPU will work at P0 under load.

3) Reset your overclocking setting to default values, i.e. 100 PL/ +0 core/ +0 mem.

4) Run Nvidia PROFILE Inspector 2.1.3.20 and set "CUDA - Force P2 State" to OFF and press Apply.

5) Now apply your overclocking setting, but you should watch memory values, as with P0 state enabled default memory clocks are higher (+200 for my 1070),
so if you apply your old +200 you will get in result +400 and it can be too much.

6) run miner and check current P state level as described in 1), it now should be P0. Check hashrate, hopefully with 2050 mhz core and 5500 mem you can reach 2 mh/s in hsrminer_neoscrypt_fork_hp.exe


newbie
Activity: 5
Merit: 0
I tried this and had an entire thesis of the tweaking and benching done, and my rig crashed the second I hit preview on the post.

Not once did it crash during testing, only trying to post.

So now, cliffs:

I tested this in place of the original hsrminer.exe supplied by miningpoolhub's also switching program, then rebenched all Neoscrypt algo's I have installed on my gaming rig (tested on here first vs taking my 2x1080ti 4x1070 rig down to bench for an a couple hours)

System - Ryzen 1800X @ 4.1 GHz (Corsair H115i Pro), Gigabyte AX370-Gaming 5 Mobo, 16GB Corsair LPX DDR-3200, Samsung 960 EVO NVME 120GB
GPU - EVGA 1080Ti SC2 Hybrid - Stock BIOS - Corsair SP120MM Fans Push/Pull Set @ 100% - NVidia Drivers 390.77 (set at prefer max performance):

All benched starting at 110% pwr, +110-115 core, +200 memory

Klaust/Nanashi/Tpruvot Algos - All max between 1,500 - 1,552 kH/s @ 2050 MHz
Just a miner HSRminer MOD - Max 1,507.47 kH/s @ 2050 MHz
Excavator (nicehash default) - Max 1,925.32 kH/s @ 2,037 MHz (Eventually peaking @ 1,985 MHz while hashing)


Benchresult #1 of 5- Klaust/Nanashi max @ 1,552 kH/s
https://i.imgur.com/YZwDU6t.png

Benchresult #2 of 5- Klaust/Nanashi/Tpruvot max @ 1,552 kH/s, Excavator max @ 1,899 kH/s
https://i.imgur.com/jc7vB7V.png

Benchresult #3 of 5- Klaust/Nanashi/Tpruvot max @ 1,552 kH/s, Excavator max @ 1,899 kH/s
Just A Miner (HSRminer.exe) MOD - max 1,507.47 kH/s @ 2050 MHz
https://i.imgur.com/RwCJO0N.png

Benchresult #4 of 5- Klaust/Nanashi/Tpruvot max @ 1,552 kH/s, Excavator max @ 1,899 kH/s
Just A Miner (HSRminer.exe) MOD - max 1,507.47 kH/s & 1,495.03 kH/s
https://i.imgur.com/xpbAOjK.png

Benchresult #5 of 5- Excavator (nicehash default) Finalbench - 1,925.32 kH/s @ 2,037 MHz
https://i.imgur.com/lwqweAa.png
jr. member
Activity: 325
Merit: 2
Gotta agree. Moved from KlausT to this miner and I had an nice increase. I went from a total of 27.9 average to 29.1 MH/s
not bad
Try to push it a bit moar! Cheesy

You want to reach nice round number of 30 MH/s ?  Cheesy

jr. member
Activity: 325
Merit: 2
@JustAMiner: what was the idea behind making a special version with high priority?

I understand why someone might want to use hi-pri to boost the hashrate and on some rigs it does help, but even the original version had the -c parameter (--cpu-priority) which did the trick just fine and that's how I have been using it for a few weeks now.

Also, looks like the _hp version does nothing compared to the regular one. At least not on my 1080Ti rig I'm currently testing it on. Just starting the process with high priority is not enough and perhaps the original ccminer does some more magic to improve the efficiency. Just like it does not help to go into Task Manager and bump up the process priority to High. Might need to look into the original ccminer code to figure out what -c parameter does.

Otherwise, another solid release yet again. I love it! If it reports all the vitals in Awesome Miner via API, I will pay you for the effort.

-scsi

Idea was that not everybody know about -c option. Other user was asking how to pass -c under Nicehashminerlegacy, and so on. That's why I decide to set priority to high by default inside program and make it separate version.

As for ccminer, in my benchmark against it I was using highprio version of my fork but with default intensity, and ccminer was running with highpriority AND intensity set to 17 instead of default 15. So no magic.

Confirming API now reporting vitals as expected in Awesome Miner. See screenshot below.

@Just_a_miner - PM me with your mailing address and I will send you one Delta 2400W server PSU for your effort.

Thanks man, but it's too much. If you want to support me, you can send some 0.xx1 of btc or ltc to donation addresses.
jr. member
Activity: 325
Merit: 2
testing on my three rigs with high priority version.
Will check average hashrate on pool in a few hours to compare
Shown hashrate is definitively higher on rig but needs to wait report on the pool to confirm

edit : i can confirm that the average hashrate on the pool is higher by a factor of 1500M/hs going from 23400 -> 25000. This is real

Thanks for feedback.

Confirmed the -d parameter works correctly now above 10 GPU's. I am dual mining with 12 now.

Justaminer, before I start playing with the new higher intensity program- can I ask what you are OC'ing your 1070's at with the new program, and getting stable results to get a starting benchmark?

Intensity is the same, just miner process priority set to high. My benchmark was done with MSI GTX 1070 Gaming X with PL 90%/+100 core/+400 mem. It works stable. Actually it works even with 80-85pl%/+150 core/+400 mem , I just don't want stress card too much.
jr. member
Activity: 557
Merit: 5
Does this fix the deteriorating hashrate issue found with the regular HSR miner?

Detetriorating hashrate is related to nvidia driver crashing and msi afterburner losing access to your card and resetting the o/c settings. Or your cards are temp throttling.
Check your windows log, they will have a nvidia driver error in it i guarantee it if it's not temp throttling
full member
Activity: 206
Merit: 100
Does this fix the deteriorating hashrate issue found with the regular HSR miner?

In my experience the hashrate does not deteriorate at all even after being running for days non-stop. You sure it was not caused by high temperatures and GPU clock rate throttling in hsr miner?
jr. member
Activity: 98
Merit: 3
Does this fix the deteriorating hashrate issue found with the regular HSR miner?
full member
Activity: 206
Merit: 100
Confirming API now reporting vitals as expected in Awesome Miner. See screenshot below.

@Just_a_miner - PM me with your mailing address and I will send you one Delta 2400W server PSU for your effort.

full member
Activity: 206
Merit: 100
@JustAMiner: what was the idea behind making a special version with high priority?

I understand why someone might want to use hi-pri to boost the hashrate and on some rigs it does help, but even the original version had the -c parameter (--cpu-priority) which did the trick just fine and that's how I have been using it for a few weeks now.

Also, looks like the _hp version does nothing compared to the regular one. At least not on my 1080Ti rig I'm currently testing it on. Just starting the process with high priority is not enough and perhaps the original ccminer does some more magic to improve the efficiency. Just like it does not help to go into Task Manager and bump up the process priority to High. Might need to look into the original ccminer code to figure out what -c parameter does.

Otherwise, another solid release yet again. I love it! If it reports all the vitals in Awesome Miner via API, I will pay you for the effort.

-scsi
member
Activity: 531
Merit: 29
@just_a_miner

Are you able to make the "overall speed" messaging appear more frequently?


Probably yes, I will see what can be done.

Maybe its card specific, as your tests are with 1070 and mine with 1080ti.

It's very strange because users report ~1.9 mh/s hashrate for one 1080ti (and 17.05 mh/s for a rig with 9 cards), so with 2 cards you should get about 3.8 mh/s and not 3.55 or even 3.4.

Those seem to be at 100% power level. My test were at 70%. Maybe that's why.
newbie
Activity: 72
Merit: 0
Confirmed the -d parameter works correctly now above 10 GPU's. I am dual mining with 12 now.

Justaminer, before I start playing with the new higher intensity program- can I ask what you are OC'ing your 1070's at with the new program, and getting stable results to get a starting benchmark?
jr. member
Activity: 557
Merit: 5
Does other people pool speeds vary?  I see it vary greatly, from like 7MH/s to 25MH/sec

How to get average speed on nice hash?  anyone?

check the average, not the instantaneous figure which is far from accurate
newbie
Activity: 17
Merit: 0
Does other people pool speeds vary?  I see it vary greatly, from like 7MH/s to 25MH/sec

How to get average speed on nice hash?  anyone?
jr. member
Activity: 557
Merit: 5
testing on my three rigs with high priority version.
Will check average hashrate on pool in a few hours to compare
Shown hashrate is definitively higher on rig but needs to wait report on the pool to confirm

edit : i can confirm that the average hashrate on the pool is higher by a factor of 1500M/hs going from 23400 -> 25000. This is real
Pages:
Jump to: