Issue with AMD RX 6800 in latest HiveOS Version 0.6-206@210723

[Gasico]

With the previous version of HiveOS i was getting pretty decent results with my AMD RX 6800 Reference cards, even without reaching the full potential hashrate (due to SOC TDC Limit capped at 30A) it was still pretty good cause to the power consumption being very low at 100w for each card

Those are the results with the previous version of HiveOS v 0.6-205@210723

and with the latest v 0.6-206@210723

as you can see the OC settings are the same, i didn’t changed anything but the power consumption increased by 10w for each card and i can confirm it by watching the powermeter

in the changelog of the latest version there is " Fixed some confirmed amd-oc bugs on Vega20/Navi10/Navi20 (fixed setting core voltage on some RX 6800" so i was thinking that with this update they will probably fix the TDC SoC Limit but that’s still an unfixed problem

i went back to the 0.6-205 and now the power consumption is again at around 100w.

The problem tho is still the TDC SoC Limit, even if the amd-oc log says that there is a change in the PowerPlayTable to the TDC Limit it is actually doing nothing i think, i tested the GPU0 on windows and with 33A it was reaching 62 MH/s while on HiveOS it is stuck at 58-59 MH/s

hope to help devs with my log to fix those issues

[vanadium]

I’m seeing the same issue. +10w on my rx 6800, same configuration, same hashrate.

I don’t think it’s linked to the version as i’m able to hit the same power consumption (around 100w) on version 0.6-206@210723.

See my screenshot here:

I confirm with my power meter that I see +10W on my card too. I can go down to 97w / 56mhs only on first boot. Afterwards, it’s 107w.

[HaloGenius]

@vanadium

I’m seeing the same issue. +10w on my rx 6800

It’s not Nvidia GPU. AMD display only GPU core consumption and it’s doesn’t matter because inaccurate, what you see on AMD GPUs as consumption. For checking power usage buy wattmeter or WiFi smart-plug with builtin wattmeter.

[vanadium]

“I confirm with my power meter”. Re-read my previous post. I’m using a wi-fi smart plug with a power meter to confirm increased power usage.

+10w both seen in HiveOS AND on my smart plug.

[HaloGenius]

@vanadium

I see +10W on my card too. I can go down to 97w / 56mhs only on first boot. Afterwards, it’s 107w.

show amd-info with 107W

[vanadium]

97w:

=== GPU 0, 07:00.0 Radeon RX 6800 16368 MB ===
Bios: 113-D4120900-101
Core: 1150 MHz 618mV, Mem: 1000 MHz
PerfCtrl: manual, Load: 99%, MemLoad: 100%, Power: 97.0 W, Cap: 203 W
Core: 58°C, HotSpot: 62°C, Mem: 68°C, Fan: 44%, RPM: 1485
Core state: 1, clocks: 500 1150*
Mem state: 3, clocks: 96 456 673 1000*
SOC state: 1, clocks: 480 640* 1200
DCEF state: 1, clocks: 417 800* 1200
F state: 1, clocks: 500 1041* 1551
PCIE Link speed:n/a, PCIE Link width:n/a
Memory total: 16368.00 MB, used: 4492.09 MB, free: 11875.91 MB, type: Samsung GDDR6

107w:

=== GPU 0, 07:00.0 Radeon RX 6800 16368 MB ===
Bios: 113-D4120900-101
Core: 1150 MHz 618mV, Mem: 1000 MHz
PerfCtrl: manual, Load: 99%, MemLoad: 100%, Power: 107.0 W, Cap: 203 W
Core: 58°C, HotSpot: 63°C, Mem: 72°C, Fan: 44%, RPM: 1485
Core state: 1, clocks: 500 1150*
Mem state: 3, clocks: 96 456 673 1000*
SOC state: 1, clocks: 480 685* 1200
DCEF state: 1, clocks: 417 872* 1200
F state: 1, clocks: 500 1140* 1551
PCIE Link speed:n/a, PCIE Link width:n/a
Memory total: 16368.00 MB, used: 4492.09 MB, free: 11875.91 MB, type: Samsung GDDR6

It seems like the difference are on Soc, DCEF and F frequencies. Temperatures are higher too, obviously.

[HaloGenius]

@vanadium

It seems like the difference are on Soc, DCEF and F frequencies. Temperatures are higher too, obviously.

I should say that SoC Clock not optimal on both

Try set Core Clock as 960 MHz … and then if it’s work for you SOC VDD as 950 mV

[vanadium]

It’s done, but no change, still 106w in hiveos / 175 - 180w at the wall:

amd-info with those settings:

=== GPU 0, 07:00.0 Radeon RX 6800 16368 MB ===
Bios: 113-D4120900-101
Core: 1150 MHz 618mV, Mem: 1000 MHz
PerfCtrl: manual, Load: 99%, MemLoad: 100%, Power: 107.0 W, Cap: 203 W
Core: 58°C, HotSpot: 63°C, Mem: 72°C, Fan: 44%, RPM: 1485
Core state: 1, clocks: 500 1150*
Mem state: 3, clocks: 96 456 673 1000*
SOC state: 1, clocks: 480 685* 1200
DCEF state: 1, clocks: 417 872* 1200
F state: 1, clocks: 500 1156* 1551
PCIE Link speed:n/a, PCIE Link width:n/a
Memory total: 16368.00 MB, used: 4492.09 MB, free: 11875.91 MB, type: Samsung GDDR6

[HaloGenius]

@vanadium

SOC state: 1, clocks: 480 685* 1200

Execute from Run menu on Dashboard
cat /sys/class/drm/card0/device/pp_dpm_socclk

[Gasico]

Let’s put aside for a moment the increased power consumption in the latest HiveOS version, when can we expect the TDC SoC Limit to be fixed?

[vanadium]

@HaloGenius

cat /sys/class/drm/card0/device/pp_dpm_socclk

root@hiveos:~# cat /sys/class/drm/card0/device/pp_dpm_socclk
0: 480Mhz
1: 640Mhz *
2: 1200Mhz

[vanadium]

I confirm that reverting to 0.6-205@210723 solves the power consumption issue too. I’m back to 97w.

[Verbl-Kint]

Are you using the stable or the beta?

[ChristopheAbher]

Personnally, the best setup for my RX 6800 Rig is :

I have forced with --eth_config=A848 to have a good intensity on each card.I removed “aggressive mode” which created a problem once or twice a day (GPU dead).

It is important to work with SSH Console :

For now, RX 6800 is better with Windows. We can up to 63Mh. I hope the new AMD drivers for HiveOS will rival Windows.

[deusofpower]

I have applied the F-State to both cards. One has DCEF State 0 (which is the one that reaches at least 60Mh/s… but unfortunately not more), the other GPU has DCEF State 1 and it is not hashing enough… only 47Mh/s.
Can UPP modify the PowerPlay table also in regards to DCEF State? I think this could be one reason why it’s not hashing enough… (but I can be wrong on that).

Any suggestions?

[Grea]

Dcef scripting if you choose that direction:

Advice for 6800, 6800xt and 6900 xt OC settings

Hello, after I discovered the dependence of F-state and hashrate. I saw that the other important thing is to lock all the frequencies in one state 0 (or 1) to do it we need to change the maximum values ​​close to the minimum. As an example: On RX 6800 (non XT) min (state 0) Dcef is 417 when you set maximum (state 1) closer to minimum it always be in state 0 upp -p /sys/class/drm/card1/device/pp_table set smc_pptable/FreqTableDcefclk/1 = 420 --write or when you set min F-state to 1551 also t…

@deusofpower

the other GPU has DCEF State 1 and it is not hashing enough… only 47Mh/s.

fwiw: My sample size of (1) Sapphire will absolutely not run well at the 1075 memory maximum and needs more voltage than any of my other 6800’s. This is on latest stable kernel, no scripts:

[deusofpower]

Thank you for your reply.

I’ve done some more testing (actually lots of hours…).
The F-State is locked and it stays locked but what I’ve seen from continuous testing is that both SOC State and DCEF state are continuously changing between 1 and 0 everytime the wattage of cards drop from 130 - 140 watts to 89 - 90 watts and viceversa. (i got some results… poor, but still better than before playing around with the SoC Frequency and SoC VDDmax in HiveOs overclock settings on each card…)

This is inherently locked to the hashrate fluctuation between 48 Mh/s and 56 Mh/s. Both cards in different settings can and reach 60 - 63 Mh/s but never ever together.

I just don’t get it why one card is locked with it’s perfomance to the other card.

It’s starting to get really frustrating. (but i’m not giving up on this…)

[Grea]

@deusofpower

I just don’t get it why one card is locked with it’s perfomance to the other card.

It’s starting to get really frustrating. (but i’m not giving up on this…)

fwiw: we have been seeing such sympathies between the 6800 series for a long time. I have never seen a good explanation or confirmation of it in detail.

I moved the “problem” 6800’s to other non-6xxx rigs to isolate them. They are all 61MH/s+ now.

Time not mining is ETH lost and I am not willing to fight it when I can solve it.

[deusofpower]

Agreed.

It’s just that I have no more space on other rigs… ^^ So i was too lazy to change cards between rigs and additionaly re set all the overclocks.
But if i’m not able to sort this out shortly i may choose your path.

[deusofpower]

At the end it was the only way that worked for me too. Isolating those cards in other rigs is the way to go. All 6800 are now 62 Mh/s.