Topic: Custom RAM Timings for GPU's with GDDR5 - DOWNLOAD LINKS - UPDATED - page 64. (Read 155485 times)

I changed my mind, I am going to post the tRTW to prevent giving away to much info -  the tRTW in the following Stilt strap is 21

77 71 33 20 00 00 00 00 08 39 47 2A 50 55 0C 0B 24 20 45 04 00 46 C4 00 22 BB 1C 00 5C 0B 14 20 4A 89 00 A0 00 00 01 20 12 0C 21 1E 51 19 26 13

Elpida, hynix or samsung? core and mem settings?

 - that's funny.  I will do one- how about tWTR - it really doesn't matter to me

You guys are funny, I bet none of you have figured out all 20 of those timings.  It is tough as shit to figure out- here's a challenge for you 3-  if you kilo, jstefanop  and eliovp can each post one of the following 3 timings from The Stilts Hyinx 1250 strap- I will then be a believer, otherwise I say you're all full of shit/

1 - tWR
2 - tWTR
3 - tRCDR


Good luck

That figure sounds about right, it is tough to figure out without thinking outside the box.  Hex editors will do nothing to help decode the straps.

True, and Samsung seems to be the biggest b*tch in case of customizing straps..

In comparison with Hynix... big difference.
Simple example: Stock speed with just a copy paste of the straps on a 480 4G (XMR) around 700h, while with my custom timing it easily hits 800h..

So far that i know, only 3 people managed to create custom straps..  but i could be wrong here :p

Another important point I have realized (and suspected when The Stilt did not do it) is that Samsung Straps are arranged differently than the others in the bios and are not at all cross compatible.

Yes he has...I told him how to do it

Yeah I think so. I got them from
http://www.overclock.net/t/1554360/tahiti-memory-timings-patch-for-hynix-vram

Are those Stilt Timings?  I have read that thread but have not compared them to Stilts originals

If you guys are referring to the Tahiti memory straps I modded some with instructions here:

https://bitcointalksearch.org/topic/hacked-modified-memory-strap-bios-database-for-zec-mining-1676474


For ZEC it makes a difference, a R9 280X (stock ) is faster than a R9 290 stock with modded memory timings.

He may have, I do not know.  I know of a few people that have made custom straps by changing out various HEX values and have gotten a boost.  I think those results will be less than if someone is changing all 16-20 values.  I did a fair amount of RAM tuning when I was overclocking with DI/LN2 and changing one number can impact the other values significantly.

I can agree with you on that- it definitely took a shit ton of work, I have just started playing with the numbers. The Stilt has mentioned 2 important points:
1 - AMD et al have really screwed up some of the numbers.
2 - He said it took 40-50 hours to perfect a set of numbers.

I think he is right on both points. Some of the numbers that are programmed in are way off (which probably made it that much more difficult to decode) and tuning RAM properly is a tedious task.  

What would the hashing results of 5 to 10% tighter timings be? Are we talking significant increases to the point it would be worth it for 10 gpu? 20 gpu? 50 gpu? Etc?

and the current results are pretty interesting.

Wolf commented in a thread several weeks ago that he has made fully-custom straps.

Yes which is exactly what I posted in that thread...he just wanted to boast about something for whatever reason. I have decoded each timing value....took a shitton of work and trial and error. Its not the holy grail your looking for though...you might get an extra 5-10% tighter timings over just swapping lower straps.

 AtomWorks gives only the HEX code for a series of values and does not actually give individual values.  It is useful in helping to figure out the timings but it takes a lot of work to actually figure out what those values really are.

For example:
-Looking at the Linux stuff available online and you can find references to things like  MC_SEQ_CAS_TIMINGS.  Under that section there are a number of values/timings (6 values I think).  

-Within Atomworks it gives a Timing like 1250 and then gives the option to change  MC_SEQ_CAS_TIMINGS but does not give the actual 6 timings individually . It would give a string of HEX values that have the 6 values included but you would still need to know the 6 timings.  

-It gives something like this:
MC_SEQ_CAS_TIMINGS at 1250 MHz  Ox456534232424  (those aren't real values)

As you can see it can be helpful but not without knowing what the info means.  Atomworks is an AMD Program and it updates through servers at AMD - unfortunately those servers are not the same as they were and it does not update.  Also it is an old copy and does not include any new GPU's like 390,380,480 etc.  

Having said that it is useful for locating things like Timings and I2C info.  I have actually figured out the individual timings on the straps and I am currently working on custom timings.  I am sure others have the info but not many, unless they work for AMD or are The Stilt.  I may be interested in doing some custom timings for farms but haven't decided.

Just food for thought.....


DOWNLOADS:
THANKS TO OhGodAGirl and Wolf0 for these items to help EASILY decode and modify straps:

You can download OhGodATool here: https://github.com/OhGodACompany/OhGodATool/releases/
You can download OhGodADecode here: https://github.com/OhGodACompany/OhGodADecode/releases/
You can download OhGodACsumFixer here: https://github.com/OhGodACompany/OhGodACsumFixer/releases

THANKS TO nerdralph for the following:
https://github.com/nerdralph/strapread/blob/master/strapmod.py

THANKS TO doktor83 for compiling Windoze Versions:
The decoder tool for windows :
https://mega.nz/#!aJAkwAxK!74I_6-f628CKIVl2TZBNrjAoujgSrKOgdnsnlwlKS_w

Checksum fixer :
https://mega.nz/#!zIwVXL4B!IUQOXRqWDw6L15rvN5gBgBN0hxc6ncQu5YcnG41m90g