Hello, i have Power Color 5700XT, micron memory, downloaded 5700 non-XT bios for the same card, applied every single timining the way its explained, first copying the 1500 to the bottom then modifying the tREF for each value from your table, then uploaded the bios to the card, what else i missed ? I applied different clocks and etc. Only difference is i’m on kernel 5.0.21 which is the oldest one. If i put the new one 5.6.0beta and add 4GB more ram so it start to load the dag it makes no difference, even sometimes its more unstable release for me. Not to mention that on a card (Sapphire) where memory is 100% samsung this image see it as Micron. So i have some doubts about what you see and if its real. Have you measured really consumption from wall ?
Hello,
Same for me : it seems not being same effective for Power Color red Devil RX5700XT. Anyway, tREF seems to have no effect, but copying the memory straps is improving a bit . Measured the consumption it show me to have almost same value (~range135-145W, pretty inaccurate ), despite of what HIVE OS reporting. Hash rate increased from 55 to 56, same OC. Can achieve 59, but consumption and heat will go to the sky … . My best personal setup is:
However, if we want to lower the value displayed by HiveOS, it simply modding a Liquid Devil BIOS by loading the MPT from Red Dragon/Devil, limiting consumption to 140W, raising some frequency (according to topics from IGOR’s lab) and copying the right memory straps, for both types. After that will show 9# watts, but real consumption will be still 14# watts
I found my sweetspot with my non-XT at 55 mhash and 72 W. but thats still 120 W out of the wall with a high efficiency power supply. you cant trust the values shown in hiveos. Soc clock e.g. reduces power consumption measured but doesnt show any difference in hiveos
Hello, i have Power Color 5700XT, micron memory, downloaded 5700 non-XT bios for the same card, applied every single timining the way its explained, first copying the 1500 to the bottom then modifying the tREF for each value from your table, then uploaded the bios to the card, what else i missed ? I applied different clocks and etc. Only difference is i’m on kernel 5.0.21 which is the oldest one. If i put the new one 5.6.0beta and add 4GB more ram so it start to load the dag it makes no difference, even sometimes its more unstable release for me. Not to mention that on a card (Sapphire) where memory is 100% samsung this image see it as Micron. So i have some doubts about what you see and if its real. Have you measured really consumption from wall ?
I’m talking about software numbers. If 130 is at the wall then it’s great. Software is real value but just only get the gpu w not the whole card…
Hello,
Same for me : it seems not being same effective for Power Color red Devil RX5700XT. Anyway, tREF seems to have no effect, but copying the memory straps is improving a bit . Measured the consumption it show me to have almost same value (~range135-145W, pretty inaccurate ), despite of what HIVE OS reporting. Hash rate increased from 55 to 56, same OC. Can achieve 59, but consumption and heat will go to the sky … . My best personal setup is:
tref definitly change something if you do it in bios mod. In amdmemtweak or hiveos it doesn’t change anything.
I found my sweetspot with my non-XT at 55 mhash and 72 W. but thats still 120 W out of the wall with a high efficiency power supply. you cant trust the values shown in hiveos. Soc clock e.g. reduces power consumption measured but doesnt show any difference in hiveos
any more details ? i would want to see if i could achieve the same with my XT flashed non XT
What else can I say? I changed memory straps as described (1500 copied to higher ones) and that´s it…clocks and voltages are made via hiveos. If you dont have high power costs, you can raise core clock and mem clock a little, but for my rig it´s no more cash at the end with actual calculations (price, difficulty, power-costs).
If your cards can manage it, you can lower core-voltage down to 750 mV…then you can reach 70w
Here is what I’m getting with MSI 5700XT Mech OC.
Looks like these cards aren’t great. I’ve read that the thermal pad location MSI uses isn’t effective. I got mine for free so it was worth a look. I’ll be switching to a different vendor, unless anyone has a recommendation for ways to lower the power usage on these.
MSI 5700 Gaming X is awesome for mining. I get 55.5 Mhash (TRM) for 97W (132W from the wall per GPU)
They are not that difficult to open and relocate the thermal pads to sit properly.
Should lower your Gddr temps somewhat.
This is stable for me. From wall it is 1670W.
Vhen i try change MVDD Rig was unstable.
My rig is 10x SAPPHIRE PULSE RX 5700 XT
2X NOCTUA NF-A14 industrialPPC-24V-3000 PWM IP67 Q100
got this stable settings on my 6x RX 5700 XT Rig from the Wall its 920 Watt …
i only have made the 1500/1550 mem straps kopie/past … the tREF settings only works for rx 4xx/5xx cards with gddr5 mem i think … i did not tested this tREF change on my bios of the rx 5700 xt cards
I tried to find out what tref change is good for, but I didnt get an answer. still dont know if its for more mhash, less watt, less temp, more stability or what else it affects…lol…
-
tREF Timing: The amount of time it takes before a charge is refreshed so it does not lose its charge and corrupt. Measured in micro-seconds (µsec).
https://www.techpowerup.com/articles/overclocking/AMD/memory/131/2
… this is what i found
I think if you give it an higher value it leaves longer the data inside the memory …
if this is the point how long is the right time for get an new refresh…
yeah…correct…but what does it affect? still not answered 
more mhash? less watt consumption? more stability? still no answer about this change…we all change this value and no one knows why…
Ok so than its about us to Check what it dose… i will pick one rx5700xt out of my rig and will Check this
First i will search for the specs from the memory direktly from the manufaktur
Than i will Check what dose ethash need
Technical Note:
GDDR6: The Next-Generation Graphics DRAM
Some Infos for the Micron-Memory…
And i found a Formular for the refresh time for memory…
!! This is from my PowerColor Red Dragon RX 5700 XT !!
!! 1500/1550 Straps Copie/Paste already done !!
BIOS Memory Info:
Settings for tREF - Micron Module on my
PowerColor Red Dragon RX 5700 XT / 1500/1550 Straps Copie/Paste
Standard Settings for tREF - Micron Module on my
PowerColor Red Dragon RX 5700 XT
Note that among ram timing items only tRFC and tREF are indexed by ‘time unit’ (second) while others by clock count (how many clock cycles (= inverse time unit) are consumed for a ram timing’s operation). It’s because they (tRFC and tREF) directly indicate characteristics of electrical working mechanism of a memory chip. Both are related to a memory’s refresh rate: tREF determines how frequently a chip shall be refreshed and tRFC determines the time elapsed between the moment of ‘refresh’ (memory access is unavailable) and the moment when a memory access is back to available. So in case of tREF, the bigger value implies the better performance.
Some editional Tools and Infos:
OK…my english isnt that bad, is it? I dont know how to explain that more clearly…pls choose one of the different answers for the following question:
when I raise the tREF value…my mining rig goes:
a) it gets more hashrate
b) it consumes less watts
c) it runs with less memory temps
d) it runs more stable with higher memory clocks
… witch language is your fav ?
@ SollteGehen
Can you try these straps on your system, and compare the hash are with the same configuration?
I try a few things now … i want to find out the best settings for tREF on my PowerColor Red Dragon RX 5700 XT and change the values with my own calculations ( must not be right )… the work is still in progress but for now i got 57 Mh/s at 119Watt shown in HiveOS and an Tempreture rise about 2°C
Miner is Phönixminer i think the shown watts is an calcuation thing and didnt reflect the true wattage
… Work is still in Progress … need a bit time to check about invalit shares …