Need to remove PCIe Gen Limitation on P104-100 and P106-100

[aentaro]

Heyyyy!!!

Anyone here know how PCIe limitations were implemented by nVidia on Mining Cards such as the P106-100 or P104-100? Are the limitations at firmware level or at the hardware?
Is there a way to remove those limitations and make those cards able to run high bandwith tasks?
Hope anyone here knows more about this.. I couldn't really find anything on Google except that there's a bunch of tools such as NiBiTor, Pascal TDP Tweaker and nvidia settings, hence smi and other tools included in the toolbox. Has anyone already attempted to do something like that... modifying the BIOS so the card can communicate via a newer PCIe standard?

 

[Macro Device]

The lanes are physically destroyed. There's realistically zero things you can do about it. Can only suggest buying a proper x16 GPU. Normal 1070s go for a Franklin on 2nd hand, not a heavy hit on your wallet.

 

[RaptorLEO]

No,It is the limitations at the hardware .you only can use pcie 1.1x16 Because p106-100's GP106 only have pcie 1.1x16.p106 is Graphics Processor different to the GTX1060

 

[Assimilator]

It's always at the hardware level. There are no free lunches.

 

[eidairaman1]

Heyyyy!!!

Anyone here know how PCIe limitations were implemented by nVidia on Mining Cards such as the P106-100 or P104-100? Are the limitations at firmware level or at the hardware?
Is there a way to remove those limitations and make those cards able to run high bandwith tasks?
Hope anyone here knows more about this.. I couldn't really find anything on Google except that there's a bunch of tools such as NiBiTor, Pascal TDP Tweaker and nvidia settings, hence smi and other tools included in the toolbox. Has anyone already attempted to do something like that... modifying the BIOS so the card can communicate via a newer PCIe standard?

You might need to consider upgrading to to a rx 6700xt or equivalent

 

[aentaro]

Dang... thanks guys! I didn't expect anyone to respond me. Maybe I was to enthusiastic about giving the card a new purpose. But I think you didn't get my point. I have 16 lanes. The card runs @ 16x 1.1... the thing is the bus speed, or Gen or however that's called. Is there no way to fool the card into running at that mode, is that phisically destroyed on the PCB or chip or what? I get that there might be some of those cards that actually have the phisical line to the pci slot cut, however that doesn't seem to be the case with my Gigabyte P104-100s (Running at 4x 1.1 cause they only have 4 caps instead of 12) and my P106-100 (16x 1.1 has all caps installed)
Please ignore my ignorance, I'm not a GPU engineer but I'm curious about so many things..

 

[Transpanda]

I actually was on a fairly similar endeavour and in short flashing a custom bios that would boot and speed up pcie or enable lanes doesn’t seem possible and while all 16 lanes have the necessary resistors(or capacitors?) (on a P106), the lanes seem to be disabled on the processor itself.
In case you’re wondering this is the thread:

Thread 'I want to flash a GTX 1060 vbios onto a p106-90 6gb'
https://www.techpowerup.com/forums/...h-a-gtx-1060-vbios-onto-a-p106-90-6gb.327844/

In short the only things you can do to my knowledge are hardware modifications and use modified drivers, but nothing bios related aside from the TDP maybe.

 

[Wirko]

The P106-100 is PCIe 3.0 x16, according to TPU database and rare other sources (EVGA), while others say 1.1. Do two distinct variants exist?

 

[Transpanda]

The P106-100 is PCIe 3.0 x16, according to TPU database and rare other sources (EVGA), while others say 1.1. Do two distinct variants exist?

So to my knowledge the database is pretty lacklustre for cards like this. To the best of my knowledge they all run on pcie 1.1

I know more about the p106-90, so for that one I can confirm that it runs on 1.1 x4 and I haven’t heard of a model that has more lanes or a higher pcie generation. Also there are variants the database doesn’t have. Like the one I have that comes out of a mining prebuilt. It has 6gb instead of 3gb and only 640 shaders instead of 768.

So I wouldn‘t be surprised if the database is also missing info on the p104-100/p106-100

Considering the reason nvidia limited the pcie speed is to make it less useful for anything besides mining, I would be surprised if the p104-100 got differentiale treatment.

 

[masterdeejay]

Lanes are not destroyed. It is mostly a firmware limitation but it is not cracked yet. It is locked not in the bios but in the GPU internal firmware. (falcon security)
There are a few cards that can have the full pcie 16x lanes like the CMP 40HX just need to solder the pcie capacitors (still 1.1 limited).

I had a few mining cards:
P106-100 - 16x works at 1.1
P106-90 4x@1.1
P104 4x@1.1 4GB unlocked to 8GB (SMS not missing, full 16x but works at 4x)
P102 4x@1.1 5GB unlocked to 10GB (SMD missing 4x-16x)
CMP 100-210 1x@1.1 12GB vram but unlocked to 16GB (SMD missing but soldering it not unlock 16x)

So most of the nvidia mining cards are trash, and that is very bad.
The CMP cards also have limited computing power, tensor cores crippled (possible just driver limitation).

 

[aentaro]

Yes, I can confirm that some of those cards come with 4 SMD soldered and some others with 16. They all are 1.1 though. My guess is that Falcon is what doesn't allow users to just flash another BIOS onto it. It doesn't make sense to physically destroy lanes or communication busses, besides, how would it even be possible to just destroy busses ver. 2 and 3 and not 1.1..

Well in that case Ima hold on to my cards in case someone here has a breakthrough hacking Falcon or someone come up with a new approach to this 'problem'.

Thanks and may the force be with you!