Topic: Mining Bitcoin Interest - new progpow algorithm (Read 2054 times)

Thanks pbfarmer, your explanations are much appreciated.
I will try testing more accordingly. Setting GPU_P7 and Mem_P3 voltages to an equal value should likely remove some headaches then.

I heard about the "applying under load" thing, I'll have to give it a try. It can however cause the miner to crash iirc. I'll have to keep monitoring with GPU-Z in parallel.

It's due to the AVFS and/or ACG feature on vegas - they throttle core clock based on load and available power.  If you keep dialing up your voltage, or lower your clock, you'll see a smaller and smaller gap between the setting and the actual.  That's why i always say to target 1400 (or whatever value) 'effective clock' - that's what really matters.

This feature is supposedly only enabled in P5-7, so technically you could write your top clock to P4 (in PPT) and disable P5-P7, and you shouldn't see this throttling - though you may have to adjust your clocks and/or voltages as a result.  That being said, it didn't work last time i tried - though when ETH mining, I only use P0 (disable P1-7) and have definitely seen that the clock droop was minimal in that case.  I've also tried disabling AVFS in the PPT - didn't do anything.

This is likely the answer to your second post as well.  The problem is that we're setting voltages that are fine for the given clocks at idle, but then AVFS has to adjust the clock based on the applied load (during mining) and likely can't do so fast enough, leading to a small window of power starvation and then a crash.  I have a feeling (unconfirmed) that the 'mem' voltage is also acting as a floor to AVFS, so may be more stabilizing in your Profile_25.  Try also setting mem_p3 to 1000 in your profile_30 and see if that makes a difference.

One other thing to keep in mind - I've noticed odd behavior where profile settings applied prior to load do not always fully 'take', while when applied under load, they do.  Setting the same profile in PPT behaves the same as the 'under load' situation, so i always tune while the miner is actually running (except mem clock, which can't be changed under load), then write my final settings to PPT.

pls take a look here first and going through all steps..maybe you have forget to set some stuff at windows or you have one faulty card (it´s more easy to analyze one card per card)
http://vega.miningguides.com/

MAYBE its also easier for you, not to use PPTable...just deactivate all P-States you dont need for mining:
GPU_P0=852;800;0
GPU_P1=991;850;0
GPU_P2=1084;850;0
....
same with Mem


after this...create a overdriventool profil for each card with MEM_P3=850/851/852/853... (then u can identify the cards better cause many miners dont show GPUZ first card at first place).

after you did this steps you should start with this P7 in all profiles
GPU_P7=1350;850
Mem_P3=850;850       (851;850/852;850....)

start miner...wait some minutes. no crash = Mem_P3 +10mhz (maybe at first you can also go directly to 950 if u are sure that its all samsung memory on the cards)
after you reach stable Mem_P3 you can go further with increasing GPU_P7 (+10mhz steps!)

Similarly, this will work...

[Profile_25]
Name=Vega64x16rTEST1
GPU_P0=852;800
GPU_P1=991;850
GPU_P2=1084;850
GPU_P3=1138;850
GPU_P4=1150;850
GPU_P5=1202;850
GPU_P6=1399;915
GPU_P7=1408;950
Mem_P0=167;800
Mem_P1=500;800
Mem_P2=800;850
Mem_P3=900;1000
Fan_Min=1000
Fan_Max=4900
Fan_Target=70
Fan_Acoustic=2400
Power_Temp=85
Power_Target=-10

...but this will crash almost immediately:

[Profile_30]
Name=Vega64x16rTEST2
GPU_P0=852;800
GPU_P1=991;850
GPU_P2=1084;850
GPU_P3=1138;850
GPU_P4=1150;850
GPU_P5=1202;850
GPU_P6=1250;870
GPU_P7=1350;1000
Mem_P0=167;800
Mem_P1=500;800
Mem_P2=800;850
Mem_P3=900;875
Fan_Min=2500
Fan_Max=4900
Fan_Target=70
Fan_Acoustic=0
Power_Temp=85
Power_Target=0

Flabbergasting, huh?

Something I don't understand is this.
I've got the following:


Why doesn't the GPU run at 1375 MHz at 1V like I'm telling it to?

Also funnily enough, these settings will only work a few minutes and then the cards will crash. So although I'm giving them more volt over GPU_P7 than I used to over MEM_P3, they still go boom. It's like a massive headfukc! 

Yeah. Needs way too much power on Vegas, no matter what I try it seems.
Was able to achieve better numbers on Grin thanks to your tips... Now all we need is better mining software... -.-

It's entirely possible progpow just needs a whole lot more power - based on https://medium.com/@ifdefelse/understanding-progpow-performance-and-tuning-d72713898db3 it looks like the algo was basically designed to fully utilize a GPU, meaning much more power would be required (try dialing your core down to 1250 and see what happens.)  That being said, if these settings don't work for cnv2, you def have other problems.  Also, these should absolutely work for GRIN   Have that running at much lower mv than this on all my 64s.

Fan_acoustic somehow modulates fan speed and/or throttles GPU to keep fan under a certain speed for noise purposes - I always just set it to 0 so it has no effect.  Power_temp is your max temp at which throttle/shutdown happen.  85 is reasonable, and should be plenty high enough at these voltages.

EDIT: just noticed your latest test was w/ x16.  That's a pretty tough algo to go low voltage on - it's basket of different algos all rolled into one, so it's power needs change frequently.

Right, so I tried this and started Wildrig (x16rt to mine Veil):

[Profile_29]
Name=VegaGRINexp
GPU_P0=852;800
GPU_P1=991;850
GPU_P2=1084;850
GPU_P3=1138;850
GPU_P4=1150;850
GPU_P5=1202;850
GPU_P6=1350;870
GPU_P7=1378;875
Mem_P0=167;800
Mem_P1=500;800
Mem_P2=800;850
Mem_P3=1100;875
Fan_Min=2500
Fan_Max=4900
Fan_Target=55
Fan_Acoustic=2400
Power_Temp=85
Power_Target=0

...and the miner crashed within two seconds.

Thanks a lot for this, buddy. I wish I had known some of these wisdoms a year ago, would probably have saved me many headaches!

The -10 power target I only set to get the cores a tad cooler. Early reviews were obtaining better results in game benchmarks when lowering power targets on ref cards for that reason. I realise that if you set voltages manually this should negate the need to play with power target.

Just saw about the "unordered" core clocks. That was something that happened as I progressively lowered GPU_P7 (as the card kept crashing). It ended up being lower than P6. I just hadn't actually noticed, thanks for pointing it out.

Obviously I knew I was steering voltages with Mem_P3, I didn't however realise using GPU_P7 would work if the the Mem_P3 value was lower. Painfully enough, I had to knowingly increase the Mem_P3 value to the levels you see above or else the cards would crash when mining.

I'll start by decreasing fan_target in any case.

The only instances I run higher mem clocks (up to 1140) is when mining CN variants.
As far as I remember, I flashed the Sapphire reference 64 bios (they are ref Sapphire 56's - so the bios should work... in theory).

I've got quite a few things to try in any case.

Finally, what exactly are fan_acoustic and power_temp used for?

EDIT: also, it shall be noted that these settings are what I use for the x16-family of algos, which are arguably more core intensive than CNv* and ETHash. Maybe they do need more juice no matter what?

Not sure this is going to solve your problem, but may be a start.  These profiles are pretty messed up.

• I'm not sure what the need is for the reduced power_target - I've never touched this, as the voltage settings should be enough (I don't think anything less than -20 would do much anyway, but still...)
• Your fan_target of 75 is incredibly high.  That would allow your core to get to 75, while your hot spot and hbm would likely be 10+ degrees higher than that.  Mem that hot certainly causes throttling, if not instability.  My fan_target for AIB cards is ~ 40-45.  For reference cards (blowers) its 50-55.
• Your core clocks on your 'flaky' profile are unordered.  Both clocks and voltages need to be in ascending order, otherwise you will end up in a weird situation, like maybe running in a lower state.  It's best to set P6 clock and voltage just above P5 - it's never used other than possibly as a brief transitional state.  For mining, you will always be in P7 if properly configured (unless you purposely lock it out.)
• Because it is the higher of the two, your mem voltage is overriding your core voltage.  So actually you're running at 1000 & 1020mv which is extremely high.  That, coupled w/ your temp targets is likely the main cause of your problems.  Your mem voltages should always be kept below your core voltage, so you can do all voltage tuning w/ just the core mv value.  Also, your mem P2 voltage is unnecessarily high - by defualt it is tied to GPU_P2 (iirc) by the PPT, and there's no need to change it from that.

Also, just on a side note, as this would affect your performance, but not cause instability:  You can go up to 1107Mhz on your mem w/o any real change in power use because the SOC only has 2 tiers in mem state P3 as far as I've seen - 1107MHz for any mem clock <= 1107, and 1200 for any mem clock > 1107.  The actual mem clock is drawing on a separate power line, so while raising the clock will increase total power use (slightly), it will not affect the core stability.

EDIT:  Another thought...  It's also possible (unless you pulled the bios from a 64 of your own,) that you have an incompatible bios flashed - say from a different mfg - that's causing problems.  If you're not going to run your mem above 1107, there's really no benefit to the 64 flash anyway - you're just using more power for the mem.  Maybe try flashing back to stock, then making the changes above.  Keep in mind that your top mem clock will be more like 950 vs 1107 in that case.

As an update... I have taken the Vegas off the Onda board and plugged them - through risers - into a Gigabyte GA-H110-D3A.
I have added a second PSU (the 1600W EVGA G2 now handles 4 Vegas, the mobo and SSD, and there's a 750W PSU taking care of the last 2 Vegas exclusively) and checked that I am only using power cables that came with the respective units. So power is taken care of.

Risers have all been swapped.

Using the PP table above, I cannot run the Vegas with anything less than the below:

[Profile_25]
Name=Vega64x16rTEST1 <---this config works on the "better" cards
GPU_P0=852;800
GPU_P1=991;850
GPU_P2=1084;850
GPU_P3=1138;850
GPU_P4=1150;850
GPU_P5=1202;850
GPU_P6=1399;915
GPU_P7=1408;950
Mem_P0=167;800
Mem_P1=500;800
Mem_P2=800;915
Mem_P3=900;1000
Fan_Min=1000
Fan_Max=4900
Fan_Target=75
Fan_Acoustic=2400
Power_Temp=85
Power_Target=-10

[Profile_26]
Name=Vega64x16rFLAKY <--- this works on the weaker cards
GPU_P0=852;800
GPU_P1=991;850
GPU_P2=1084;850
GPU_P3=1138;850
GPU_P4=1150;850
GPU_P5=1202;850
GPU_P6=1399;915
GPU_P7=1388;950
Mem_P0=167;800
Mem_P1=500;800
Mem_P2=800;915
Mem_P3=900;1020
Fan_Min=1000
Fan_Max=4900
Fan_Target=75
Fan_Acoustic=2400
Power_Temp=85
Power_Target=-10

As you can see, we are far, FAR away from  sub-900mV.
Mind you this is no CNv2, but still. I'm "only" around 1400MHz on the core. Not giving it anywhere near stock frequencies.

Hello,

I believe you can tell most miners how to use certain devices by adding -d parameter in your batch file or shell script. Example: -d 1,2,3,4 to skip device 0 if that is your internal GPU.

If you are struggling to get windows drivers and miners configured properly, please give PiMP OS a look. It's a turnkey package that runs out of the box and will save you hours of frustration. You can also avoid the dreaded "Windows Updates" problems as well as gain better stability and performance, and a suite of troubleshooting tools for when things go wrong.

Thank you!

~ melt
getpimp.org | miner.farm

 try only 3 cards  and alternate the slots.

I used to do that with that board and it allowed me to push cards pretty hard.

Thanks for the good insight, it's given me food for thought. I'll have to test with mem clocks under 1107 and check if I can lower core voltage accordingly.

They are ref Sapphires and do have Samsung memory indeed.
The power delivery, huh. Who knows. I've got an EVGA 1600 G2 to power the six of them which leaves enough headroom, at least for CNv1/2. They are all plugged on an Onda D1800 board, which might actually be the culprit. I'm not sure how well built that board is. There might just be power leakage/loss from there.

Your top SOC is 1200, but you only run that w/ a mem clock over 1107.  For <= 1107 clocks in mem state P3, you'll be running an 1107 SOC, which is the major factor (possibly more so than core clock) in what voltage you'll need.  So a mem clock range from 1100-1140 can cause a massive power swing - right at 1108Mhz.  1200 SOC requires around 900mv, 1107 requires ~835mv (+/- 10mv) from my experience - at least w/ CNv2.  SOC and core seem to share voltage lines, so depending on core clock, those numbers can go up/down.  For instance, w/ ethash, core frequency can go all the way down to 850MHz (if not lower, but that's the driver floor) w/o affecting performance, which leaves a lot of power headroom for SOC - i can get my 64s all the way down to 805mv still running 1107 mem.

950-975mv sounds really high to me, even for a 1200 SOC.  At least for CNv2, that would be crazy high.  It's possible PP is super taxing on the core, pulling more power that way, and thus starving the SOC causing instability.  But if you keep your mem under 1107, it's hard to believe you would need even 900mv.  Sounds like something else - I run ref Sapphires as well as Nitros, and they all behave roughly how I've detailed here.  Btw, I'm assuming samsung mem?  maybe diff if hynix - possible that you're just running too high clocks for it.  Otherwise, if it's not your settings, maybe you have a power delivery problem?

The PPT has set SoC at 1200. I run mem clocks between 1100 and 1140 on CNv2 depending on the GPU. I need to give the core at least 950mV via mem_P3 in overdriventool for 1408 MHz or else the cards will crash (975 for the weaker cards). So I'm really not sure how people get theirs to mine at 800-850mV, it's very far beyond reach for my 6 cards for sure.

...unless I am doing something wrong.

In my experience w/ a number of other algos (haven't tested PP yet,) AIB and ref aren't all that different performance or efficiency wise - mainly it's just about cooling.  As for your PPT, i can say for certain that at least on CNv2, those voltages would be completely unnecessary for those clocks - even using TRM which is the most taxing CNv2 miner out there.  Just as @Marvell2 stated - 800-850mv (avg about 840) is what seems to be necessary for an 1107 SOC clock, while also running a core clock in the range you're using.  Furthermore, given you're flashed to a 64 bios, you should be able to go up to an 1107 mem clock w/ no change to voltage - whether that provides any benefit to this algo vs your 1000Mhz setting, I'm not sure.

Holy thread resurrection batman!
Actually I had never seen that answer so happy days.

I am indeed running a modded soft PP table on the Vegas. But I can't run such low voltages or the cards will crash. I've got reference Sapphire RX Vega 56's flashed to 64's btw. Have you got AIB Vegas?

This is my reg file, in case someone can spot anything terrible in it:

Windows Registry Editor Version 5.00
Get in touch via bitcointalk on updates or further development wishes. User brnsted
ETH A13015081E1f9CF27A2843e2A623F362D0d08666
XMR 47NgZN7Mp5CcfKzUKhFHXiMuXghudzFegUPWaTUJU71QVze3P1o2gd9MuhroHrYGAH1spqWEWDrnvUp eqFCic8B772ySpCM
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\0005]
"PP_PhmSoftPowerPlayTable"=hex:B6,02,08,01,00,5C,00,E1,06,00,00,EE,2B,00,00,1B,\
  00,48,00,00,00,80,A9,03,00,F0,49,02,00,8E,00,08,00,00,00,00,00,00,00,00,00,\
  00,00,00,00,00,02,01,5C,00,4F,02,46,02,94,00,9E,01,BE,00,28,01,7A,00,8C,00,\
  BC,01,00,00,00,00,72,02,00,00,90,00,A8,02,6D,01,43,01,97,01,F0,49,02,00,71,\
  02,02,02,00,00,00,00,00,00,08,00,00,00,00,00,00,00,05,00,07,00,03,00,05,00,\
  00,00,00,00,00,00,01,08,20,03,52,03,52,03,52,03,52,03,52,03,6B,03,89,03,01,\
  01,89,03,01,01,84,03,00,08,60,EA,00,00,00,40,19,01,00,01,80,38,01,00,02,DC,\
  4A,01,00,03,90,5F,01,00,04,00,77,01,00,05,90,91,01,00,06,C0,D4,01,00,07,01,\
  08,D0,4C,01,00,00,00,80,00,00,00,00,00,00,1C,83,01,00,01,00,00,00,00,00,00,\
  00,00,70,A7,01,00,02,00,00,00,00,00,00,00,00,88,BC,01,00,03,00,00,00,00,00,\
  00,00,00,38,C1,01,00,04,00,00,00,00,00,00,00,00,88,D5,01,00,05,00,00,00,00,\
  01,00,00,00,70,D9,01,00,06,00,00,00,00,01,00,00,00,00,26,02,00,07,00,00,00,\
  00,01,00,00,00,00,05,60,EA,00,00,00,40,19,01,00,00,80,38,01,00,00,DC,4A,01,\
  00,00,90,5F,01,00,00,00,08,28,6E,00,00,00,2C,C9,00,00,01,F8,0B,01,00,02,80,\
  38,01,00,03,90,5F,01,00,04,F4,91,01,00,05,D0,B0,01,00,06,C0,D4,01,00,07,00,\
  08,6C,39,00,00,00,24,5E,00,00,01,FC,85,00,00,02,AC,BC,00,00,03,34,D0,00,00,\
  04,68,6E,01,00,05,08,97,01,00,06,EC,A3,01,00,07,00,01,68,3C,01,00,00,01,04,\
  3C,41,00,00,00,00,00,50,C3,00,00,00,00,00,80,38,01,00,02,00,00,A0,86,01,00,\
  04,00,00,01,08,00,98,85,00,00,40,B5,00,00,60,EA,00,00,50,C3,00,00,01,80,BB,\
  00,00,60,EA,00,00,94,0B,01,00,50,C3,00,00,02,00,E1,00,00,94,0B,01,00,40,19,\
  01,00,50,C3,00,00,03,78,FF,00,00,40,19,01,00,88,26,01,00,50,C3,00,00,04,40,\
  19,01,00,80,38,01,00,80,38,01,00,50,C3,00,00,05,80,38,01,00,DC,4A,01,00,DC,\
  4A,01,00,50,C3,00,00,06,00,77,01,00,00,77,01,00,90,5F,01,00,50,C3,00,00,07,\
  90,91,01,00,90,91,01,00,00,77,01,00,50,C3,00,00,01,18,00,00,00,00,00,00,00,\
  0B,E4,12,60,09,60,09,4B,00,0A,00,54,03,90,01,90,01,90,01,90,01,90,01,90,01,\
  90,01,00,00,00,00,00,02,04,31,07,DC,00,DC,00,DC,00,90,01,00,00,59,00,69,00,\
  4A,00,4A,00,5F,00,73,00,73,00,64,00,40,00,90,92,97,60,96,00,90,55,00,00,00,\
  00,00,00,00,00,00,00,00,00,00,00,00,00,00,02,02,D4,30,00,00,02,10,60,EA,00,\
  00,02,10

r9 280x - 2.8 mh 

Yes, power value in watts and no longer dpm state since 1.3.x, similar to nvidia pwr settings.

if i recall, ethos power setting is a straight power value in watts, not a power limit in percent. right?

Yes seems so, powerlimit has also an impact.   Just didn't flash my GPUs to default VBios cause i'm using ethOS in multirig config, so i can switch miners any time.

do the modded memory straps do anything for progpow? has anyone tested it side by side with a card before and after memory straps bios mod? or is everyone just assuming it helps?

from what i'm reading, it sounds like progpow is totally core speed dependent and doesnt really care about the gpu memory. is that correct?

0,9V = 900mV

My settings for progpow

RX580 @ 10,3 mh/s
Core: 1240
Mem: 1750
Vcore: 890mV
vddci: 900mV
dpm: 5
pwr: 160W (@144W)

RX570 @ 8,5 mh/s
Core: 1150
Mem: 1750
VCore: 890mV
vddci : 900mV
dpm: 5
pwr: 150W(@133W)

Modded straps for all cards.

what do you mean by 0,9?

most of my rx are moddes to like 1130 core ,850mv 1875 mem , 925mv

they are great at cn 7 and eth, but horrible at progpow.  I max at like 8mhs and unstable as hell

I have mem straps modded too

@dragonmike ? are your vegas modded with power play tables etc ?

I have 8 vegas and they are all modded voltages are like 850 to 800 for thr core and they pull no more than 160 to 180 watts even on x16 , progpow and like 140 watts on cn forks .

vegas must be power play modded with that spreadsheet unless you have refernce card which has a power play mod floating around.

I never even touch the power limit on any of my amd cards, it’s overdriven tool for my rx cards and powerplay plus overdriven for the vegas.

My vegas are my best mining cards hash/watt.

btw why is there no power play mod for rx cards?

Why you reduce the PL Limit on RX570? This is stupid. Lower Voltage and not the PL 
Every RX570 can run 1250 on gpu core with 0,9 vgpu.

I think you missed the point.
Of course you can lower the power limit on ProgPow. I'm doing just that with my Vega cards for example. I need to drop it by 30% to keep a rig of six Vega64's under 1400W... Which unfortunately it has a large impact on the hashrate. So it might improve your efficiency (slightly) but not enough to make it remarkably more profitable. On RX 570's dropping PL by only 10% radically lowers hashrate and it worsens as you go lower.

You know that the power depends on the voltage, right? Lower the power limit to 50% on nvidia or lower the voltage/core clock on amd and it'll draw less than ethereum. Nobody is forcing you to run at max power level, especially when the efficiency of hash/watt decreases with the power draw.
Lyra2z can still be mined at the original power level. You can simply lift the limit up and it'll hash more.

ProgPow and most other core-based algorithms scale perfectly well with core count and doesn't have these weird discrepancies where a weak card (RX 570) hashes as much as a twice more expensive one (GTX 1070). That way you can get 2x more performance from 2x more expensive card, so it just balances it out. RX 570 and 580 will be as powerful as GTX 1060 3 and 6GB. Vega 56/64 will be as powerful as GTX 1070/1070Ti.

Yes, there are algorithms where AMD or NVIDIA are still better, but right now we're focusing on ProgPow.

Everyone seems to rave about ProgPow as being great against ASIC proliferation and all... All hail ProgPow! Woohoo!...

...ignoring the fact it's a very power hungry algo. So in these days of low profitability hardly anyone will reap any benefits from BCI mining. That, unfortunately, is a double-edged sword which promotes efficiency of ASICs vs GPUs.

We used to have lyra2z as "power saving" algo. Unfortunately devs chose the route of optimising it by allowing for more power to be thrown at it to gain hash. Now it's just as hungry as everything else out there.

The challenge would be to design an efficient algo, that embeds decent ASIC resistance and doesn't scale much by just adding more power.


Maybe I just described CryptonightV2...

I don´t think so. I test Progpow on 2x 1050ti. ETH and BCI is nearly the same.

Yep, i saw you re mining BCI on another topic.
Acording to calculators, it s nonsense to mine BCI.
I prefer to mine BBR on wildkeccak. I can buy BCI after selling them. Then stake
I know you have solar energy and dont pay the bill. But you must calculate the profit.

you could be right.

I mine them but I will push on bci for a few months.  I would like to get 1000 coins and stake them.

is it profitable ?

I think ETH or cryptonight is more profitable than progpow

My own settings/quick tests:

RX580 4GB
Core: 1240 MHz
Mem: 1750 MHz
Vcore: 900mV
MCore 900mV
pwr: 150W

10,7 mh/s

Core: 1150 MHz
Mem: 1750 MHz
Vcore: 880 mV
MCore: 900 mV
pwr: 130W

9,4 mh/s

RX570 4GB
Core: 1150 MHz
Mem: 1750 MHz
Vcore: 880 mV
MCore: 900 mV
pwr: 150W

8,7 mh/s

RX570 4GB
Core: 1150 MHz
Mem: 1750 MHz
Vcore: 880 mV
MCore: 900 mV
pwr: 130W

7,4 mh/s

From my own experience:

RX 570 4GB: 7.5 MH/s (-10% PL)
Vega 56 (flashed to 64): 15 MH/s (at -30% PL - higher clocks can be achieved if you've got money to burn)

Anywhere to find reliable benchmark list for different AMD/Nvidia GPUs on ProgPOW algo?

And your AMD GPUs are?

Anyway remove any overclocking and raise or remove powerlimits, use undervolting only.  If you're running a VBios mod with heavy modifications flash default VBios.


 

add:
--opencl-platform 1
to your command

like sgminer

I m trying to mine BCI but the miner only use onboard graphic card and give 0.52 MH/s
How could I use my AMD GPUs ?

AMD Miner: https://www.bitcoininterest.io/downloads/progpowminer-amd-windows.zip

My BAT file: