Quote from: Comgold on July 11, 2018, 09:33:29 PM

Quote from: wacholek on July 11, 2018, 08:16:28 PM

Quote

Did you attentively read the reference? And that you will burn the card and all.
Try it

--setoverdrive % Set GPU OverDrive level (manual|high)

--setoverdrive #: This sets the percentage above maximum for the max Performance Level. For example, --setoverdrive 20 will increase the top sclk level by 20%. If the maximum sclk level is 1000MHz, then --setoverdrive 20 will increase the maximum sclk to 1200MHz

But personally I didn't try it.

May be --setoverdrive High

As a matter of fact I did read the reference thoroughly.

1. Setting --setoverdrive % is valid ONLY for the highest performance level which in case of Vega64 the 7'th. The 7'th performance level is @ 1711Mhz for the core. So increasing --setoverdrive % of 10% set (theoretically) the highest clock to 1882,1Mhz which is impossible to handle by the core at the stock voltage.

2. I don't know why Rocm set the 7'th clock level to 1711MHZ since the card have 7'th set to 1630Mhz under Windows.

3. Rocm can force Vega to work in 7'th clock level without the load( all diodes on the cards are on). Than it is 1630Mhz even though there is not such supported frequency.

First frequencies under load. Second idling.

Code:

==================== ROCm System Management Interface ====================
================================================================================
GPU Temp AvgPwr SCLK MCLK Fan Perf SCLK OD
  3 55.0c 218.0W 1536Mhz 945Mhz 82.75% high 0%
  1 58.0c 219.0W 1536Mhz 945Mhz 80.78% high 0%
  4 57.0c 219.0W 1536Mhz 945Mhz 82.75% high 0%
  2 59.0c 219.0W 1536Mhz 945Mhz 82.75% high 0%
  0 N/A N/A N/A N/A 0% N/A N/A
================================================================================
==================== End of ROCm SMI Log ====================

Code:

==================== ROCm System Management Interface ====================
================================================================================
GPU Temp AvgPwr SCLK MCLK Fan Perf SCLK OD
  3 49.0c 32.0W 1630Mhz 945Mhz 82.75% high 0%
  1 54.0c 39.0W 1630Mhz 945Mhz 80.78% high 0%
  4 53.0c 32.0W 1630Mhz 945Mhz 82.75% high 0%
  2 53.0c 32.0W 1630Mhz 945Mhz 82.75% high 0%
  0 N/A N/A N/A N/A 0% N/A N/A
================================================================================
==================== End of ROCm SMI Log ====================

4. There is no information about what kind of power is displayed in rocm-smi. If this is a hole card power we have a lot of room to growth. But if it is a Core power only. hmm.. I have never seen numbers above 211W. TDP VEGA 64 Air is 295W and LC 345W. Under Windows we can monitor at lest 4 different temperatur of the VEGA 64 (CPU thermal diode, HBM2 temp, Hot spot temp, vrm temp) and power for the core and for the GPU package. Rocm can provide only CPU thermal diode temp and some kind of udefined power consumption.

5. The most important temp in terms of Lyra2Z algo is the Hot spot temperature which we don't have under rocm. There is many speculations about what is a Hot spot temp in fact. I like to belive that it is actually a SoC temp. It has no direct contact with the heatsink so gets very hot. Especially during gaming or, as i presume, during Lyra2z mining. Using cryptonightv7 algo we can observe that the hotspot is up to 10*C higher than the core thermal diode since it is a memory intensive algo. But during gaming the difference can be up to 20*C and easy hit 100*C.

Conclusion:
Vega 64 on air can work up to 6'th freq level when mining (only 6 diodes on the cars in on) The limit is the power. At least the Vega 64 air bios not allowing to pass 220W.
Since the PowerPlay table can not overclock the 6'th or 5'th freq level to achieve higher hashing speed we have to undervolt the card to meet 220W limit. That should boost the core frequency by itself.
Since I don't know any way how to undervolt the the GPU we're back to square one.

At the end of the week I'am planning to install a water cooling on my 4 Vegas. I wonder how the frequency is scaling with the temperature. Than I will re-flash those cards using Vega LC bios.

I'll keep you posted.
Cheers:D

I generally try to receive the maximum productivity, at the smallest consumption of energy. What from us is gradually taken away by producers since it is necessary for them that the video card has failed through certain time what we would buy the new video card.

Indirectly it is confirmed by three monthly guarantee for the video cards mining edition.

Therefore with interest I will watch your experiments.

I wish you good luck.

Power consumption is not a problem for me since I have electricity for free. So only productivity is the key.
Using xmr-stak under Windows cryptonightv7 algo I get 2080h/s from a single cards. Core at stock and mem at 1100Mhz. That is 250W/ cards from the wall. 55*C core, 63*C hot spot, 63*HBM.
So far cards where air cooled, kept in 18*C air conditioned room. Now will be water. Today first tests.
Cheers:D