Any word on amd vega hash rates?

[l1xx]

6x VEGA 64

Using https://github.com/halsafar/Vega64SoftPowerTableEditor
Modify registry:
- 1408mhz for P5,6,7.
- 1050mhz for mem
- 950 for core voltages

HBCC only works for 4/6 cards so I am short ~1000H.

Getting 10500H @ 1150W (total draw at the wall).

Total draw at the wall when idle is 250W.  

So that is ~150W per card.

Stable for 72 hours.  99% submission rate on nano pool.  My average on the pool is expected, 10500.

This is huge given if I try the same setup using stock settings the total draw is ~2000W.

Whats your average result time there? And is it the US or european server.
I have avg result time around 10 seconds, but not sure if thats good or bad tbh. Lot of stale shares lately, was around 3% before.

[rubine2323]

6x VEGA 64

Using https://github.com/halsafar/Vega64SoftPowerTableEditor
Modify registry:
- 1408mhz for P5,6,7.
- 1050mhz for mem
- 950 for core voltages

HBCC only works for 4/6 cards so I am short ~1000H.

Getting 10500H @ 1150W (total draw at the wall).

Total draw at the wall when idle is 250W.  

So that is ~150W per card.

Stable for 72 hours.  99% submission rate on nano pool.  My average on the pool is expected, 10500.

This is huge given if I try the same setup using stock settings the total draw is ~2000W.

Whats your average result time there? And is it the US or european server.
I have avg result time around 10 seconds, but not sure if thats good or bad tbh. Lot of stale shares lately, was around 3% before.

Are you guys using Blockchain Driver or 17.9.3?
I tried 17.9.3 everything was stable after i tired Blockchain driver my monitor was flashing arround. After uninstall and reinstall 17.9.3 it writes error installing driver. Now i cant install anymore a AMD Driver. Error 1603 appears during install. Tested AMD Clean uninstall tool and all other steps written on AMD Page. Any solutions? I really dont want to reinstall windows ...

[rednoW]

6x VEGA 64

Using https://github.com/halsafar/Vega64SoftPowerTableEditor
Modify registry:
- 1408mhz for P5,6,7.
- 1050mhz for mem
- 950 for core voltages

HBCC only works for 4/6 cards so I am short ~1000H.

Getting 10500H @ 1150W (total draw at the wall).

Total draw at the wall when idle is 250W.  

So that is ~150W per card.

Stable for 72 hours.  99% submission rate on nano pool.  My average on the pool is expected, 10500.

This is huge given if I try the same setup using stock settings the total draw is ~2000W.

950mV is to much for monero 1408/1050.
My cards are 1408/1100(some cards need 1080mem) with 905mV in registry (and OverdriveNTool) and 875mV in Hwinfo64 ... seems stable ...

[rubine2323]

6x VEGA 64

Using https://github.com/halsafar/Vega64SoftPowerTableEditor
Modify registry:
- 1408mhz for P5,6,7.
- 1050mhz for mem
- 950 for core voltages

HBCC only works for 4/6 cards so I am short ~1000H.

Getting 10500H @ 1150W (total draw at the wall).

Total draw at the wall when idle is 250W.  

So that is ~150W per card.

Stable for 72 hours.  99% submission rate on nano pool.  My average on the pool is expected, 10500.

This is huge given if I try the same setup using stock settings the total draw is ~2000W.

950mV is to much for monero 1408/1050.
My cards are 1408/1100(some cards need 1080mem) with 905mV in registry (and OverdriveNTool) and 875mV in Hwinfo64 ... seems stable ...

Do you use Blockchain Driver or the normal 17.9.3 Driver ?

[rednoW]

Do you use Blockchain Driver or the normal 17.9.3 Driver ?

Blockchain drivers only ... you'll not get 1900+h/s monero or 44+mhs eth with non-blockchain drivers

[miner1337]

6x VEGA 64

Using https://github.com/halsafar/Vega64SoftPowerTableEditor
Modify registry:
- 1408mhz for P5,6,7.
- 1050mhz for mem
- 950 for core voltages

HBCC only works for 4/6 cards so I am short ~1000H.

Getting 10500H @ 1150W (total draw at the wall).

Total draw at the wall when idle is 250W.  

So that is ~150W per card.

Stable for 72 hours.  99% submission rate on nano pool.  My average on the pool is expected, 10500.

This is huge given if I try the same setup using stock settings the total draw is ~2000W.

950mV is to much for monero 1408/1050.
My cards are 1408/1100(some cards need 1080mem) with 905mV in registry (and OverdriveNTool) and 875mV in Hwinfo64 ... seems stable ...

Fan settings are important btw.  With my settings the cards only hit ~56C while HBM2 mem is 70+.  Manual fan settings required.  I use 3100 RPM.



Yeah I am positive I can go a bit lower.  I wanted a stable base before I shot for the moon.  Next iteration on the rig now that I know these settings are stable I was thinking of 910mv.

I had my mem running at 1100 but for an extra like 10H per card I felt it wasnt necessary given the extra heat.   I'll be plotting results and optimizing this once I have more data.

[Shnikes101]

Fan settings are important btw.  With my settings the cards only hit ~56C while HBM2 mem is 70+.  Manual fan settings required.  I use 3100 RPM.

Yeah I am positive I can go a bit lower.  I wanted a stable base before I shot for the moon.  Next iteration on the rig now that I know these settings are stable I was thinking of 910mv.

I had my mem running at 1100 but for an extra like 10H per card I felt it wasnt necessary given the extra heat.   I'll be plotting results and optimizing this once I have more data.

A visual would be nice to see. I'm running at 900mv right now at 1100. Seems stable but if its only a 10 hash difference, I may back my memory off.

[STT]

Fan settings are important btw.  With my settings the cards only hit ~56C while HBM2 mem is 70+.  Manual fan settings required.  I use 3100 RPM.



Yeah I am positive I can go a bit lower.  I wanted a stable base before I shot for the moon.  Next iteration on the rig now that I know these settings are stable I was thinking of 910mv.

I had my mem running at 1100 but for an extra like 10H per card I felt it wasnt necessary given the extra heat.   I'll be plotting results and optimizing this once I have more data.

Memory isnt producing great heat on this card vs a normal GDDR5 setup, its just from being located so close to the gpu.  HBM2 is low volts and low heat apparently.    The second voltage shown for memory is actually like a floor to pair with the GPU voltage, so a bit lower then the left GPU volt would be about right

On another thread, I read 80c is where HBM2 starts to go wrong and cause problems.  I havent tested this exactly.   I found throttling if I dont set main target temperature to 69c or less.  
Results will vary by ambient temp also.  I read someone with a custom cooler had temp in the 40's for GPU core but hot spot showed over 100c which is a disaster.     Normally you should assume worse temps are 20c over the GPU temp given which is actually the best case for the card.   I guess the stock cooler just takes care of number 1 first which is the expensive gpu

[Hellae]

Now that I can easily build/rebuild the miner, decided to learn some basics of C++ today for my own profit:

Old Web Interface:
https://i.imgur.com/1F9pA6P.png

New Web Interface:
https://i.imgur.com/mh8IDsf.png

Now I can spend less time adding thread values.

I'll not share the miner because it's without devfee (and has all the devfee code removed aswell), but if you can build it and are interested on my code changes for this UI, those I can share.

---

I've finally decided to try the regedit thing.

The 3 vega 56 there are all at 1408@910mV (haven't tried 905 nor 900 yet. that's the next step) and 1100@1050mV (V64 OC with V56 voltage), with min fan speed at 3500 and 0% Power Limit.
System is stable, have been running 24h, and had to restart it to test the new miner changes I did.

The device 0 is always doing more hash that the others, many times at 1920H/s. I suspect it's because that's the card on the riser connected to the x16, since everything else is equal. Other than that, it can only be wattman's sketchy code...

[dk5664]

Anyone running with Win10 Creators update? I updated to Creators a few days ago and my hashrate is taking a big hit. It starts at 1.75k like before Creators, but drops to 1.4k after running just a couple of minutes. I've been trying many things but unable to get my V56 running at 1.75k stable. Anyone having this problem?

[A62662]

hi

i have 2x 56 cards.i flashed with 64 bios.

i put -20% /1100/-20%,hbcc ok--->1800/1850 h/s per 1 card,connect to my screen with hdmi.

when i put the second card,same tweaks, the card connected to screen lose 400h/s ?

is there a way to fix this?

thanks

edit : bios 64 drain more watts that the 56?

Yes, the higher frequencies will draw more power. Make sure your volt is stable or it might crash sonner or later.

[oomurashin]

Now that I can easily build/rebuild the miner, decided to learn some basics of C++ today for my own profit:

Old Web Interface:


New Web Interface:


Now I can spend less time adding thread values.

I'll not share the miner because it's without devfee (and has all the devfee code removed aswell), but if you can build it and are interested on my code changes for this UI, those I can share.

---

I've finally decided to try the regedit thing.

The 3 vega 56 there are all at 1408@910mV (haven't tried 905 nor 900 yet. that's the next step) and 1100@1050mV (V64 OC with V56 voltage), with min fan speed at 3500 and 0% Power Limit.
System is stable, have been running 24h, and had to restart it to test the new miner changes I did.

The device 0 is always doing more hash that the others, many times at 1920H/s. I suspect it's because that's the card on the riser connected to the x16, since everything else is equal. Other than that, it can only be wattman's sketchy code...

cool,its hashrate  only less  100H/s than my 64's same core clock.
Vega 56 is most profitable card at Monero mining!!

ill try mod miner later.

[rednoW]

I've finally decided to try the regedit thing.

The 3 vega 56 there are all at 1408@910mV (haven't tried 905 nor 900 yet. that's the next step) and 1100@1050mV (V64 OC with V56 voltage), with min fan speed at 3500 and 0% Power Limit.
System is stable, have been running 24h, and had to restart it to test the new miner changes I did.

The device 0 is always doing more hash that the others, many times at 1920H/s. I suspect it's because that's the card on the riser connected to the x16, since everything else is equal. Other than that, it can only be wattman's sketchy code...

Did you control the resulted gpu voltage in hwinfo64 and wattage from the wall?
I think you should set "memory" voltage (it is not memory voltage in fact) to the same value as your gpu core voltage.

[ccleouf]

Hello,

I'm new in the mining, I have 2 rx vega 56 for monero minig.

My setup ->
cpu g4400
cm asus z270-p
ram 4GO
PSU 1000W evga

My 2 cards are connected with riser, first riser on pcie 1x and seconde one on 16x.

with your setings (-30% gpu core clock, Mem@950mhz, -20 power limit and fan at 3200-3500) i have 3500 - 3600.

But i have an issue, my first card have better hashrate thant the second and i don't understand why...
And when i reboot i need to play with enabling or desabling hbcc, sometime work well without and sometime withe it enable :/

You think the pci e setting in bios (GEN 1, GEN 2 ...) and/or riser can have an impact ?

[mutual.consent]

hi

i have 2x 56 cards.i flashed with 64 bios.

i put -20% /1100/-20%,hbcc ok--->1800/1850 h/s per 1 card,connect to my screen with hdmi.

when i put the second card,same tweaks, the card connected to screen lose 400h/s ?

is there a way to fix this?

thanks

edit : bios 64 drain more watts that the 56?

The 64 bios has higher voltage limits and on the HBM memory its most useful.   HBM isnt using large amounts of power in any case, its like 10w difference
Theres a theory that the 56 actually overclocks and runs at high end better then the 64, maybe because it has less shaders.    That would be ironic, I guess mining is not using those extra shaders I dont know.  Its mostly about memory bandwidth I read.

With your setting of the power limit.  You want to lower the volts on left and right, lower the power limit yes but not long after you raise the power limit back up and this enables the best performance and you should retain the lowest voltage usage in any case.  It may wobble back and forth a bit, 120 to 140 watts I see but mostly its just showing a reading of 130w.   Some people no doubt have got lower power usage then that even, there is a correct sequence to get the best results though.  If driver crashes, log out and start over again as it wont run as well in my experience

How are you measuring power draw? Is this with a wall meter and is that total system draw or load draw that you've calculated after minusing idle power draw? I get about 200w total system draw with one Vega 64 @ 925mV 1408/1100. Haven't lowered the power limit but have assumed everyone has been quoting power draw after minusing idle or other system component draw. Would be surprised if you can get these figures for total system draw at the wall.

[l1xx]

My draw is stable around 1100W with 6x vega 64. Total system from the wall taht is, not some estimation.

[rednoW]

For 120-140watt mining xmr we need <840mV core voltage with low gpu clock. No way to get it with 1408clock and ~900mV voltage

[Hellae]

Did you control the resulted gpu voltage in hwinfo64 and wattage from the wall?
I think you should set "memory" voltage (it is not memory voltage in fact) to the same value as your gpu core voltage.

Nope. I didn't even knew Hwinfo64 has that feature. Last time I opened it it sent erroneus data to sensors (temperature at 500º) and the fans gone crazy. Had to reboot and didn't opened it again since. Maybe I need to update it.

Wattage on the wall, always. I'm at ~560W
https://i.imgur.com/u15eNWD.jpg

950mV is to much for monero 1408/1050.
My cards are 1408/1100(some cards need 1080mem) with 905mV in registry (and OverdriveNTool) and 875mV in Hwinfo64 ... seems stable ...

Reducing from 910 to 905 will be my next tweak. The 875 in Hwinfo64 I really need to google that. Always used it to read sensors only. Thanks.

wait... or are you refering to the 0.875 VDDC reading on Hwinfo64? I'm confused now.

[rednoW]

Did you control the resulted gpu voltage in hwinfo64 and wattage from the wall?
I think you should set "memory" voltage (it is not memory voltage in fact) to the same value as your gpu core voltage.

Nope. I didn't even knew Hwinfo64 has that feature. Last time I opened it it sent erroneus data to sensors (temperature at 500º) and the fans gone crazy. Had to reboot and didn't opened it again since. Maybe I need to update it.

Wattage on the wall, always. I'm at ~560W

950mV is to much for monero 1408/1050.
My cards are 1408/1100(some cards need 1080mem) with 905mV in registry (and OverdriveNTool) and 875mV in Hwinfo64 ... seems stable ...

Reducing from 910 to 905 will be my next tweak. The 875 in Hwinfo64 I really need to google that. Always used it to read sensors only. Thanks.

wait... or are you refering to the 0.875 VDDC reading on Hwinfo64? I'm confused now.

Yes, I'm talking about 0.875 reading in hwinfo (sensors only). You will get it if you set 900-905mV in patch.
I also had problems with hwinfo system hang but solved it with disabling I2C for gpu in settings as Mumak, hwinfo author, advised.

[Hellae]

Yes, I'm talking about 0.875 reading in hwinfo (sensors only). You will get it if you set 900-905mV in patch.
I also had problems with hwinfo system hang but solved it with disabling I2C for gpu in settings as Mumak, hwinfo author, advised.

I'm with 905mV on all cards now.
https://imgur.com/a/3WXyz

0.875 VDDC At the wall it dropped to 525W -35W

I'm comfortable with this power count so I'll not try to stress the cards even more with 900.

Last I want to try is new intensity values. Atm I'm with 2016/1700. 1800 on the second one gave lower hash.