Custom RAM Timings for GPU's with GDDR5 - DOWNLOAD LINKS - UPDATED

[ku4eto]

Frankly, i dont believe 39Mh/s is possible at all.

If i set the workunit size to 96 or 64 (something likethat), i get 50Mh/s on my 470. But how healthy the share is... no idea.

[ku4eto]

Guys,

OhGodATool is now available on Windows as well.

http://www.overclock.net/attachments/48014
http://www.overclock.net/t/1604567/polaris-bios-editing-rx5xx-rx4xx/3240

[doktor83]

Guys,

OhGodATool is now available on Windows as well.

http://www.overclock.net/attachments/48014
http://www.overclock.net/t/1604567/polaris-bios-editing-rx5xx-rx4xx/3240

Sorry to inform you but it's available on windouz ..well for about a month maybe more..

[ku4eto]

Thats not the Decode, thats another tool, named Tool. I dont see it anywhere in this topic. Or maybe i missed it?

[doktor83]

I compiled and shared it a while ago, but maybe Vento edited something, i don't know.

[ku4eto]

Doing over 500h/s on a RX460. Kek.

Before unlocking - 470h/s, after unlocking, 494h/s at 2000Mhz.

At 2200Mhz, getting 525H/s, maximum, 1V for IMC, getting some HW though.

[dwu]

I have trouble getting my Hynix H5GC4H24AJR 4GB stable.
777000000000000022339D00CE516A3D9055111230CB4409004AE600740114206A8900A00200312 0150F292F94273116
999000000000000022559D0010DE7B4480551312B78C450A004C0601750414206A8900A00200312 018112D34A42A3816
999000000000000022559D0031627C489055131439CDD50A004C06017D0514206A8900A00200312 019123037AD2C3A17
BBB000000000000022889D0073EE8D53805515143ECF560C004E26017E0514206A8900A00200312 01C143840C5303F17

I tried already zeroing FAW, 32AW on only 2000, or down to 1500, but it never was stable: often just freezes.
I also tried nerdralphs Hynix straps of this thread, but it was also not stable. I have them on a board with Samsung cards, that shouldn't make problems, right?

I got around 28Mh, but not stable. My Samsungs are at 31Mh.

I would be very glad for hints!

[bughatti]

To the master's of this thread, I would like to revisit a few questions if anyone is willing to answer.  When it comes to modding straps with all the options of settings that can be changed, is their a guide to what values can be adjusted.  I went back and looked at a few mod bios I have received over the last couple months, different cards and different mem types to better understand what was changed and why. 

My biggest issue with understanding is do you guys who have the knowledge to manually change mem straps use a guide tool or has it been purely trial and error over the last few years? 

Not asking anyone to give away secret sauce, just want to know if I am wasting my time looking for helpful guides or just start shooting in the dark!

[ku4eto]

Well, you would want first to know how exactly the algo is loading the memory - is it Read or Write?

From there, you know which timings you need to lower.

Also, timings have lots of formulas - 1 timing can depend on 2-3-4-5 more.

There is documentation - check JEDEC GDDR5 docs, DDR3 wikipedia, it can help with understanding a bit.

After that, its trial and error, until you hit good values.

[dallase]

 :'(Was testing MSI RX580 8GB w/ Samsung memory tonight.  

28.9mh/s with
 * amdgpu-pro 16.60
 * 1150/2100mhz
 * Uber Mix 3.1

28.1 mh/s with
  * amdgpu-pro 17.10 drivers
 * 1150/2100mhz
 * Uber Mix 3.1

And not really stable at all.

Not sure why 17.10 drivers cause such a big drop in hash rate..  Started looking at the timings from the 470's with Samsung vs the 580 with Samsung, and all of the timings on the 580 are slower.   trfc's and ras2ras are significant, and another clock on tcl.

Code:

------------------------------------------------------------------------------------------------------------------------
Strap (RAS)              trc     trcdr    trcdra     trcdw    trcdwa      trrd 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470          68        22        22        14        14         6 
Sammy-1625-rx580          71        24        24        14        14         6 
Sammy-1750-rx470          73        24        24        16        16         6 
Sammy-1750-rx580          77        26        26        16        16         6 
------------------------------------------------------------------------------------------------------------------------
Strap (CAS)            tccdl       tcl     tnopr     tnopw      tr2r      tr2w      tw2r 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470           3        21         0         0         5        29        15 
Sammy-1625-rx580           3        21         0         0         5        29        15 
Sammy-1750-rx470           3        22         0         0         5        29        16 
Sammy-1750-rx580           3        22         0         0         5        29        16 
------------------------------------------------------------------------------------------------------------------------
Strap (MISC)            trfc       trp    trprda    trpwra 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470         146         9        11        49 
Sammy-1625-rx580         178        11        12        52 
Sammy-1750-rx470         157        10        11        53 
Sammy-1750-rx580         192        12        13        57 
------------------------------------------------------------------------------------------------------------------------
Strap (MISC2)            faw  pa2rdata  pa2wdata     t32aw     tredc  twdatatr     twedc 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470          10         0         0         7         2         0         7 
Sammy-1625-rx580          10         0         0         7         3         0         7 
Sammy-1750-rx470          10         0         0         7         2         0         7 
Sammy-1750-rx580          10         0         0         7         3         0         7 
------------------------------------------------------------------------------------------------------------------------
Strap (ARB1)           actrd     actwr rasmactrd rasmactwr 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470          23        15        46        54 
Sammy-1625-rx580          25        15        47        57 
Sammy-1750-rx470          25        17        49        57 
Sammy-1750-rx580          27        17        51        61 
------------------------------------------------------------------------------------------------------------------------
Strap (ARB2)        bus_turn   ras2ras        rp  wrplusrp 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470          23       146        42        50 
Sammy-1625-rx580          23       178        45        53 
Sammy-1750-rx470          23       157        44        54 
Sammy-1750-rx580          23       192        48        58 
------------------------------------------------------------------------------------------------------------------------

Anyone have any bright ideas as to why?

[manotroll]

Those values above 30mh/s and possible at rx 480 with low comsumption?
100W?

[dallase]

:'(Was testing MSI RX580 8GB w/ Samsung memory tonight.  

28.9mh/s with
 * amdgpu-pro 16.60
 * 1150/2100mhz
 * Uber Mix 3.1

28.1 mh/s with
  * amdgpu-pro 17.10 drivers
 * 1150/2100mhz
 * Uber Mix 3.1

And not really stable at all.

Not sure why 17.10 drivers cause such a big drop in hash rate..  Started looking at the timings from the 470's with Samsung vs the 580 with Samsung, and all of the timings on the 580 are slower.   trfc's and ras2ras are significant

Code:

------------------------------------------------------------------------------------------------------------------------
Strap (RAS)              trc     trcdr    trcdra     trcdw    trcdwa      trrd 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470          68        22        22        14        14         6 
Sammy-1625-rx580          71        24        24        14        14         6 
Sammy-1750-rx470          73        24        24        16        16         6 
Sammy-1750-rx580          77        26        26        16        16         6 
------------------------------------------------------------------------------------------------------------------------
Strap (CAS)            tccdl       tcl     tnopr     tnopw      tr2r      tr2w      tw2r 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470           3        21         0         0         5        29        15 
Sammy-1625-rx580           3        21         0         0         5        29        15 
Sammy-1750-rx470           3        22         0         0         5        29        16 
Sammy-1750-rx580           3        22         0         0         5        29        16 
------------------------------------------------------------------------------------------------------------------------
Strap (MISC)            trfc       trp    trprda    trpwra 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470         146         9        11        49 
Sammy-1625-rx580         178        11        12        52 
Sammy-1750-rx470         157        10        11        53 
Sammy-1750-rx580         192        12        13        57 
------------------------------------------------------------------------------------------------------------------------
Strap (MISC2)            faw  pa2rdata  pa2wdata     t32aw     tredc  twdatatr     twedc 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470          10         0         0         7         2         0         7 
Sammy-1625-rx580          10         0         0         7         3         0         7 
Sammy-1750-rx470          10         0         0         7         2         0         7 
Sammy-1750-rx580          10         0         0         7         3         0         7 
------------------------------------------------------------------------------------------------------------------------
Strap (ARB1)           actrd     actwr rasmactrd rasmactwr 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470          23        15        46        54 
Sammy-1625-rx580          25        15        47        57 
Sammy-1750-rx470          25        17        49        57 
Sammy-1750-rx580          27        17        51        61 
------------------------------------------------------------------------------------------------------------------------
Strap (ARB2)        bus_turn   ras2ras        rp  wrplusrp 
------------------------------------------------------------------------------------------------------------------------
Sammy-1625-rx470          23       146        42        50 
Sammy-1625-rx580          23       178        45        53 
Sammy-1750-rx470          23       157        44        54 
Sammy-1750-rx580          23       192        48        58 
------------------------------------------------------------------------------------------------------------------------

Anyone have any bright ideas as to why?

Finally got it over 30mh.  From uber 3.1 strap, I had to bump up trc, trfc, ras2ras, and clocks now 1185/2125, running stable.   Seems like the 580's need a bit more core clock, but they do take a better undervolt, so power remains about the same.

[dallase]

Finally got it over 30mh.  From uber 3.1 strap, I had to bump up trc, trfc, ras2ras, and clocks now 1185/2125, running stable.   Seems like the 580's need a bit more core clock, but they do take a better undervolt, so power remains about the same.

Based on what I see here, the 16.60 drivers still outperform the 17.10s, although the gap has narrowed, probably because my timings are looser than uber 3.1

AMD 17.10.2 drivers

Screen #0
ETH - Total Speed: 29.985 Mh/s, Total Shares: 16, Rejected: 0, Time: 00:08
Screen #1
ETH - Total Speed: 30.160 Mh/s, Total Shares: 23, Rejected: 0, Time: 00:08
Screen #2
ETH - Total Speed: 30.031 Mh/s, Total Shares: 21, Rejected: 0, Time: 00:08
Screen #3
ETH - Total Speed: 30.057 Mh/s, Total Shares: 17, Rejected: 0, Time: 00:08
Screen #4
ETH - Total Speed: 30.100 Mh/s, Total Shares: 19, Rejected: 0, Time: 00:08

AMD 16.60 drivers

Screen #0
ETH - Total Speed: 30.148 Mh/s, Total Shares: 15, Rejected: 0, Time: 00:07
Screen #1
ETH - Total Speed: 30.368 Mh/s, Total Shares: 17, Rejected: 0, Time: 00:07
Screen #2
ETH - Total Speed: 30.223 Mh/s, Total Shares: 14, Rejected: 0, Time: 00:07
Screen #3
ETH - Total Speed: 30.224 Mh/s, Total Shares: 18, Rejected: 0, Time: 00:07
Screen #4
ETH - Total Speed: 30.219 Mh/s, Total Shares: 18, Rejected: 0, Time: 00:07

[RobertD]

What are max stable clocks for Hynix and Samsung ram?  My best Hynix max out at 2050MHz and Samsung at 2200MHz.

[dragonmike]

So I played around with nerdralphs tool for my msi 570 armor oc 4gb with hynix.  I copied 1500 up on both Micron-Elpida and hynix.  I then took the copied value and input in nerdralphs tool for both memory sets and pasted the output in 1625 1750 and 2000 on both memory sets.

Micron-Elpida
Old - 777000000000000022AA1C00315A6B3CA0550F15B68C1506006AE4007C041420CA8980A90200000 01712262B612B3715
New -777000000000000022AA1C00315A5B3CA0550F15B68C1506006004007C041420CA8980A9020000001012262B612B3715

Hynix
Old - 777000000000000022339D00CE516A3D9055111230CB4409004AE600740114206A8900A00200312 0150F292F94273116
New -777000000000000022339D00CE515A3D9055111230CB440900400600740114206A8900A002003120100F292F94273116

Can anyone tell me why I am getting 39 MH on this card now and should I be concerned?  I have watched it for about an hour and it is pretty steady at 39 with an occasional drop to 34.5





I am using smOS so unsure how to check for hardware errors

Are you still using that custom strap and are you still seeing these hashrates? With Elpida mem?

[lpedretti]

:'(Was testing MSI RX580 8GB w/ Samsung memory tonight.  

28.9mh/s with
 * amdgpu-pro 16.60
 * 1150/2100mhz
 * Uber Mix 3.1

28.1 mh/s with
  * amdgpu-pro 17.10 drivers
 * 1150/2100mhz
 * Uber Mix 3.1

And not really stable at all.

Not sure why 17.10 drivers cause such a big drop in hash rate..  Started looking at the timings from the 470's with Samsung vs the 580 with Samsung, and all of the timings on the 580 are slower.   trfc's and ras2ras are significant

Anyone have any bright ideas as to why?

Finally got it over 30mh.  From uber 3.1 strap, I had to bump up trc, trfc, ras2ras, and clocks now 1185/2125, running stable.   Seems like the 580's need a bit more core clock, but they do take a better undervolt, so power remains about the same.

What's Uber Mix ?

[lexele]

performance oriented memory straps for polaris cards with samsung memory

[Mattthev]

What's Uber Mix ?

http://www.overclock.net/t/1604567/polaris-bios-editing-rx5xx-rx4xx
->Performance timings

[ku4eto]

Guys, any ideas why when properly settings ACTRD,ACTWR,RASMACTRD,RASMACTWR and TRC, i get worse performance. They are following the formula, but i get less H/s in XMR compared to when they are not following the formula. Also it becomes too unstable.

[doktor83]

what formula?