RX580 8G ASUS: SAMSUNG MEMORY – K4G80325FB-115-D009PI2-101 – [~31.80 MH] – [TPD: 84W]
CORE: 1156 DPM: 3 VDD: 855 MEM: 2100 MDPM: 1 AU: ON amdmemtweak --REF 20
NOTE: I have two RX cards from ASUS. 570 8GB and 580 8GB both with Samsung memory. Whenever I apply OC settings for memory clock more than 2050Mhz I always get 'invalid shares' on these cards. Mining with MEMORY CLOCK settings: 2110Mhz/2100Mhz give me 3 invalid shares/734 all with a speed of ~32.15MH/s and a miner uptime of 2.5 h. If you want to minimize the errors to zero maybe try with lower values for memory clock and maybe a little higher for core clock. Miner I use is Pheonixminer with stock settings.
RX580 8G GIGABYTE SAMSUNG MEMORY – K4G80325FC – [~31.90MH/s] – [TPD: 63W – 67W]
CORE: 1155 DPM: 3 VDD: 855 MEM: 2120 / One of my card work best on 2085 with 31.80 MH MDPM: 1 AU: ON amdmemtweak --REF 20
NOTE: Out of the four identical Gigabyte cards, three are running over 31.90MH/s without errors and one is slightly below at 31.80MH/s because applying values for the memory clock above 2085MHz cuts the card off from the others and it stops mining - I guess it depends on that damn card lottery. With 2085MHz clock card runs stable and without errors.
RX570 8G ASUS: SAMSUNG MEMORY – K4G80325FC-115-C940PI0-100 – [~31.40] – [TPD: 77W]
CORE: 1156 DPM: 3 VDD: 850 MEM: 2050 MDPM: 1 AU: ON amdmemtweak --REF 20
NOTE: 570s by Asus. Memory errors in the form of invalid shares appeared in my case above the 2050MHz memory clock, although at that clock the card mine in at ~31.40MH/s. I was able to overclock this card to over 32MH/s but it was returning too many errors.
NVIDIA RTX3060TI SAMSUNG MEMORY – 7G-94.04.38.00.97 – [~62MH/s] – [TPD: 135W]
CORE: -500 MEM: 2600 PL: 135
NOTE: With the above OCs you should pull about ~62MH/s with this card without errors.
TEMPERATURES: Since the cards are undervolted, power consumption and temperature are also kept at a lower level. With the above settings, the rig consisting of 6xAMD and 1xNVIDIA is drawing ~940W on the wattmeter (not including the effectivity of the power supplies) and mine in at over 251MH/s.. With open windows in cool weather temperatures do not exceed 60 degrees. With the windows closed the temperature does not exceed 65 degrees at an average fan speed of 62%.
Polaris BIOS editor for Linux/MacOs/Windows:
Usage of Polaris Bios Editor [terminal based]:
MAKE SURE YOU HAVE BACKED UP YOUR CURRENT BIOS !!! YOU CAN DO THIS SEAMLESSLY WITH HIVEOS OR WITH OTHER AVAILABLE TOOLS.
- list all bios tables
./polaris-bios-editor-unix -i ORIGINAL.rom -s -r ALL
- list only table of VRAM_TIMINGS
./polaris-bios-editor-unix -i ORIGINAL.rom -s -r VRAM_TIMING
- apply UberMix-3.1 timings straps to array 10 and 11 (1750MHz and 2000MHz) on RX 580 8GB GIGABYTE GPU
./polaris-bios-editor-unix -i 32MH-RX580-8G-GIGABYTE-Samsung_K4G80325FC.rom -o RX580-8G-GIGABYTE-Samsung_K4G80325FC-modded-00.rom -n 777000000000000022CC1C00AD615C42F0590F15300D9708006007000B031420FA8900A00300000011112F3FBB354019 -p 10 11
PolarisBiosEditor141 for Windows is much easier and intuitive to use. On Linux it works fine with wine and mono. All you need to do is change the timings for the appropriate memory frequencies – usually higher values like 1750 or 2000. Be careful – do not accidentally paste a white space into the bios. If you do not like the modified bios, you can always flash the gpu with the original bios. Remember to disable all OC settings before flashing and preferably use maintenance mode if your system has support for it. And remember to avoid flashing in bulk mode – it is easy to make a mistake.
VRAM_TIMING used to mine Etherum [1750MHz-2000MHz] on both Gigabyte and Asus:
To increase hashrate remember that there is a tool like amdmemorytweak. This program has a ton of capabilities to manipulate GDDR5 memory on the fly. One recommended option for RX5xx cards is to add a memory refresh value to the OC settings: amdmemorytweak --REF 30. Although a value of 30 is recommended - for ethash - a value of 120 brings visible results (using Phoenixminer there is a jump by 0,1MH on each AMD card). In order to find out on what values the cards are working on, run the program with the command amdmemorytweak --current.
UPDATE: The above results were based on Phoenixminer and despite a slightly higher hashrate there were more invalid shares. The screen below shows similar settings but I use Lolminer and although there is a loss of about 2-3MH/s everything is run stable and without errors.
And the screen below shows my cards after I setup “0” for “sensitivity” in the bios. I had a similar view in front of me for about 2 hours when I try to fix it after flashing and tried to booting it. There was a problem with the driver as it could not read the value from the bios for “power play table”. Adding amdgpu.pp=0 to grub did not help. Reinstalling the drivers did not help either. New errors appeared after reinstalling the drivers. Flashing to the original bios also did not work. Finally the “hive-replace” function came to the fully repair this issue. After 15 minutes everything was back to normal.