Running EVGA Precision X1 on rig with mixed RTX 3000 on it to update Firmware/BIOS

[pclausen]

I have a couple of rigs with EVGA GPUs in the and they are not all performing the same despite being identical models (most are FTW3s) and using the same clock profiles.

I got this rig with all EVGA except for one ASUS (no cards in this rig are LHR)

And then I got this rig that has all LHR EVGA cards in it:

Note how the 3080Ti at the bottom is only doing 48 MH/s where the other 2 are in the 50's mining Raven?

Anyone tried letting Precision X1 loose on a rig like this to update the firmware on all the GPUs?

The safe thing to do would be to take out one card at a time and stick in a rig by itself and let X1 do its thing, but that would be a real pain.

 

[pclausen]

So I added this to my T-Rex call:

--lock-cclock 0,1200,0,1200,0,1200

So locking just the 3080Ti's to 1200 for now. And I get this:

So a nice bump on all 3 3080Ti's. However, that one on the bottom (brand new that just arrived today), is still lacking behind the others.

Btw, I didn't mess with my GPU clocking profile for the 3080Ti's on Kawpow, it still looks like this:

I wonder if I should change that now that I'm playing with core lock?

The reason I brought up updating the Firmware/BIOS in the first place, is that I also have another rig with a 3080Ti Hybrid in it, and it benefitted greatly from letting Precision X1 update the firmware as there was a know issue with limiting max power with the initial firmware/bios that shipped on the card (EVGA 3080Ti XC3 Hybrid).

 

[pclausen]

Yep, that was it. Set PL to 85% and they are now all hashing at 57.5 MH/s.

But that last one sure is power hungry at 1200 core! I'll play around locked core clock to see where the sweet spot is for efficiency. Patrik is adding a 3rd row to the Performance column to show efficiency in the next release. So that will make tuning easier on a per GPU basis unless you look at the miner console.

 

[motqalden]

I loaded precision on a rig with a xc3 3070 and an asus. I was only doing it to disable the rgb but it immedietly updated firmware. No issues. Actually i think i did that on two similar rigs

 

[pclausen]

I went ahead and installed Precision X1 on my rig with all the EVGA LHR cards. Sure enough, as soon as I launched for the first time, I got this popup:

And after it was done with the first card, I got this:

That repeated a total of 3 times (so 4 cards were updated). And then Precision X closed down. I relaunched, and it again told my firmware needed to be updates, so I let it do the 5th GPU. It never asked me for the 6th GPU, so it mush have already been on the latest version.

The BIOS versions never changed after the update on any of the GPUs, which I guess was expected. I couldn't find a place in GPU-Z where it shows the firmware version, so I couldn't verify the before and after firmware versions.

Anyway, all hash rates and power levels remained the same after the update.

xenium I use Awesome Miner to control clocks and power levels. I don't even have Afterburner installed on that rig. Anyway, I did set the PL to 300 watts (75%) on that last card and played with the core clock, but could not get it past 50.5 MH/s.

 

[pclausen]

I hear you on running Windows as the OS for mining. It can be a real pain. I do have some rigs running Ubuntu (mostly my AMD rigs) since there's a Awesome Miner client for both Windows and Linux. I'll probably switch back to all Ubuntu at some point, but updating drivers and dealing with xorg.conf can be a real pain too. Been thinking of going back to HiveOS as well, which I use to run on everything. On my AMD rigs on Ubuntu, I have Awesome Miner launch a script on startup that sets the core clock and memory. Like with the new T-Rex option that locks core, that kind of dictates the voltage as well.

Sample script that I call:

# Clocks and Power sudo upp -p /sys/class/drm/card1/device/pp_table set --write smc_pptable/FreqTableGfx/1=1600 smc_pptable/FreqTableUclk/3=885 #Nitro+ 1600 / 710 stable sudo upp -p /sys/class/drm/card2/device/pp_table set --write smc_pptable/FreqTableGfx/1=1500 smc_pptable/FreqTableUclk/3=885 #Nitro+ sudo upp -p /sys/class/drm/card3/device/pp_table set --write smc_pptable/FreqTableGfx/1=1500 smc_pptable/FreqTableUclk/3=900 #Red Devil 1500 / 905 stable sudo upp -p /sys/class/drm/card4/device/pp_table set --write smc_pptable/FreqTableGfx/1=1420 smc_pptable/FreqTableUclk/3=900 #Nitro+ # Fan Speed reporting /home/m1/FanFixer.sh

 

[pclausen]

Yeah, I have a base Ubuntu 20.04 image that I drop on a small SSD (like 60Gb) and after the first boot, I just change the host name, reboot and tell add that host to Awesome Miner, and away I mine. With the newer 3000 series Nvidia cards, I get better performance with the latest drivers on Windows than the drivers that are part of base Ubuntu image. I might look at switching those rigs (with the 3000 GPUs) back to Ubuntu at some point. Also, Ubuntu does not show the memory temp on Nvidia cards, which is one of the main reasons I switched to Windows with the 3000 series. It was quite an eye opener to see the memory temp on my 3090's and even my 3080's. So I added heatsinks to the back plates on the 3090's and set the fans higher on the 3080's, which got them in check.

All my 5700XT's have been BIOS modded and strapped. I'm close to 60 MH/s on some of them. Here's one rig:

Btw, the power readings above is after adding 35% to account for the power used for the rest of the card (memory, fans, etc).

Here's the raw console output from the miner: