Quote from: greatauror28 on February 19, 2018, 01:59:11 AM

Quote from: OhGodAGirl on February 19, 2018, 01:38:58 AM

Finally logged into this account after a while. Eep.

So much has changed!

- All cards should ship with a default 30.5-31.5 MH/s VBIOS in 2018.
- OhGodATool v 2.0 is incoming - voltage probing, on-the-fly editing in Linux, decode/encode, and a whole bunch more. For Polaris and Vega, I might add! And, a lovely GUI, provided by OhGodABird. Whoop!
- Guides for GDDR5 and GDDR5X in ELIG fashion are incoming. We're just waiting on the website to be finished. Now that I've grown some balls and started ignoring the threats from others, it's time to demistify all this shit.
- HBM2/GDDR6 will be the main focus for 2018. Get acquainted with it, folks.

Currently, I need to gather requests from the community: what do YOU want to see in an all-in-one GDDR5/GDDR5X editing/optimization tool? Besides it automatically training to get the optimal values.

I guess if the tool can suggest optimal values not only in timing straps but voltages as well, card dependent.
Also, is it possible to suggest a memory clock value based on ASIC quality of a card e.g. a card having higher asic quality can withstand higher clocks as well without tons of memory errors?

I hope Im making sense here and apologies if not.

So after speaking to nerdralph I think I might have to hire him for the work/development he's done on this front.

It's possible to write a training algorithm - indeed, GDDR5 already has this - to automate the best/optimal settings based on your card.

The big problem with this is it needs to be done everytime the card boots and it would need to be adapted for every memory vendor/card type. That's a lot of work.

I suppose we can have a database that the tool reads from and then writes to the card, and adjusts the values dynamically based upon miner feedback. Would need to hook into the miner somehow, pull the hashrate value, and contrast it with its referenced database.

This would be the fastest/easiest solution.

For voltages, this is tricky because so many factors determine a good offset: the straps you're running; the age of the card; its environment; dual mining; what miner, etc. Variables that often I have no way of pulling from a software environment.