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

[ku4eto]

Heh, got my Elpida strap stable.

At 1200/2000Mhz, i am getting 975h/s with 896/8/2 and Compute Mode.

At 1280/2000Mhz - 1020h/s.

With the other card, i am getting 1068h/s at 1380/2000, or 1038h/s at 1280/2000.

[1715084326]

1715084326

[1715084326]

1715084326

[1715084326]

1715084326

[1715084326]

1715084326

[1715084326]

1715084326

[1715084326]

1715084326

[RaidoP2]

Hi

Currently I have HIS RX570 4GB with Elpida EDW4032BABG. Default straps is below:
1500: 777000000000000022AA1C00315A6B3CA0550F15B68C1506006AE4007C041420CA8980A90200000 01712262B612B3715
1625: 777000000000000022AA1C0073627C41B0551016BA0D9606006C060104061420EA8940AA0300000 01914292E692E3B16
1750: 777000000000000022AA1C00B56A7D46C0551017BE8E1607006C07010C081420FA8900AB0300000 01B162C3171313F17
2000: 999000000000000022AA1C0018F77E4FD055121946501708006C07011D0C1420FA8980AC0300000 01E19323781364718

I am copying 1500 to 1625, 1750 and 2000. By setting memory clock to 1950Mhz (1040mhz core), I am able to get 750hash/sec cryptonight or 28 Mh/sec Ethash. Any idea if this can be improved further?

Thanks

What mining software do you use ?

Try this strap : 777000000000000022AA1C00315A6B3CA0550F15B68C1506006AE4007C041420CA8980A90200000 01914292E692E3B16
               or : 777000000000000022AA1C0073627C41B0551016BA0D9606006C060104061420EA8940AA0300000 01B162C3171313F17
         maybe : 777000000000000022AA1C00315A5436A0550F15B68C1506004082007C041420CA8980A9020004C 01712262B612B3715    ( 990h/s on my card @1250,2075 )

Please report the results.
 Good luck  

Hi, do you think I could also try it out on my R9 380 with Elpida memory after I convert it around?

Do you think I could use the same timings for Hynix AJR or could you suggest one for AJR?

My current R9 380 hynix 1500 strap looks like this: 7771332000000000CE516A3D9055111230964909004AE60022339D08740114206A8900A00200312 0150F292F94273116.

Many thanks!

[grobuzga]

Hello guys!

I've stumbled upon some post on timings editing and decided to tinker a little with my cards. I have Gigabyte RX480 4Gb G1, Sapphire RX470 Nitro 4Gb and MSI RX460 OC 2Gb. All my cards have Hynix memory, H5GC4H24AJR.

I took the timings from PBE (UberMix 3.x AFAIK), which PBE offered for my cards and done some improvements onto it. Gigabyte cards wouldn't work good with it, HWinfo reported thousands of errors even at 2000MHz. But Sapphire and MSI received noticeable boost in hashrate: from 30.1 to 30.5MH for RX470s, and from 13.2 to 14MH for RX460s, with same clocks as before timings modification.

Here's the strap from one of my RX470s (>1500): 777000000000000022339D00CE515A3D8055111230CB440900400600740114206A8900A00200312 0100F292F94273116

This one is from RX460 2Gb (>1425): 777000000000000022339D00CE515A3B805511112FCBD408004006006C0014206A8900A00200312 0100F272D8D263015

If those straps work good for you, please let me know. Also won't mind to receive some ETH as thanks : 0x1F8FeA63Fd7F7465f8aB788277b9afE4B9AFf5E8

[vladutz1186]

Hello ! I have some problems with one rig who have 6 x AMD R9 380 (Sapphire, Asus Strix and MSI) GPUs on ETH . I saw people who get 23-25 MHs with this R9 380, but i can get only 20 MHs.

I try to mod bioses for all GPUs with hex editor, few versions of moded bios ... ( strings from 1500 Mhz to put up and strings from 1425 Mhz to put up ), but no more hashrate !

Hope to somene can help me ! Thanks in advance !

OS : ethOS
Miner : sgminer
Oc settings : Core - 1130 / Mem - 1660 , Core - 1050 / Mem - 1580 , and others more OC settings !

Edit :

Sapphire Nitro : Strings from 1500 Mhz to 1625 and 1750

(1500 Mhz) F0 49 02 02 77 71 33 20 00 00 00 00 CE 51 6A 3D 90 55 11 12 30 96 49 09 00 4A E6 00 22 33 9D 08 74 01 14 20 6A 89 00 A0 02 00 31 20 15 0F 29 2F 94 27 31 16

[DJMorgan]

My best Samsung timings still do 33MH/s+ on my 580, so I would think so.

This is a delightful result!
Share your knowledge of how to select?

ASUS RX580 O8G Samsung (K4G80325FB) - maximum that made 30 ETH

originally time:
1750 and 2000

Code:

777000000000000022CC1C00106A6D4DD0571016B90D060C006AE70014051420FA8900A0030000001B11333DC0303A17
777000000000000022CC1C0031F67E57F05711183FCFB60D006C070124081420FA8900A0030000001E123A46DB354019

1750 and 2000 <--

Code:

777000000000000022CC1C00CEE55C46C0590E1532CD66090060070014051420FA8900A00300000012123442C3353C19

core 1170 mem 2140 (or 2120)

[R0land]

I had found a new better memory strap for Elpida RAM. More than 1kh/s cryptonight xmr !

RX 470 : 1020 h/s  , sgminer @1250 core, @2100 mem, 66W (gpu-z)

                   942 h/s , @1150 core, @2050 mem, 52W (gpu-z)

777000000000000022AA1C00AC615B3CA0550F142C8C1506006004007C041420CA8980A9020004C 01712262B612B3715

I don´t know the eth hash speed.   Maybe somebody can check it ?

[OhGodAGirl]

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.

[greatauror28]

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 I’m making sense here and apologies if not.

[OhGodAGirl]

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 I’m 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.

[nerdralph]

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

Perhaps it's a bit pedantic, but the training is to adjust the electrical interface for the gddr5 8-bit data bus.  It finds the optimal drive & termination characteristics in order to avoid errors, but it doesn't adjust timing.  So for example when your card heats up from 20C to 80C, the resistance of the traces increase, which means more attenuation of the electrical signals.  That may lead to more CRC errors, and when they reach a threshold the memory controller will initiate a retrain.  The training sequence itself will then read and write fixed data patterns on the bus, trying things like increasing the output driver strength to overcome attenuation issues.  Increased driver strength may lead to data reflections causing more errors, which could be corrected with lowering the termination resistance on the receiving end.

Another tweak I've wondered about is monitoring memory die temps.  GDDR5 chips have internal thermal diodes similar to the GPU, and they can be enabled by setting a bit in mode register 7.  Knowing the temperature of each memory die could provide useful diagnostic info for problems such as a heatsink not having good contact with the chip.

[nerdralph]

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.

For ethash it can likely be done without any feedback from the miner on hashrate, as it's all about 128-byte random burst read performance.
For ZEC it is a bit trickier, since different kernels can have slightly different memory access characteristics, but in general short (32-byte) random memory write performance is the limiting factor.
For ZEC and possibly cryptonight, it might also be helpful if auto-precharge can be controlled.

[Branko]

Anyone deciphered yet whats in those three seq_misc parts of strap?

[nerdralph]

Anyone deciphered yet whats in those three seq_misc parts of strap?

I think it was already discussed.  They are mode registers.
MC_SEQ_MISC1      (MR1/MR0)
MC_SEQ_MISC3      (MR5/MR4)

I'd have to go back through my notes and code to confirm if MISC8 is mode registers as well.

[Branko]

Anyone deciphered yet whats in those three seq_misc parts of strap?

I think it was already discussed.  They are mode registers.
MC_SEQ_MISC1      (MR1/MR0)
MC_SEQ_MISC3      (MR5/MR4)

I'd have to go back through my notes and code to confirm if MISC8 is mode registers as well.

I was just curious, as if thats the case, those would hold for example WRITE latency
and CAS latency, but it seems people never modify them, just use default value
(at least that seems to be the case for all the straps I saw circulating around)

[nerdralph]

Anyone deciphered yet whats in those three seq_misc parts of strap?

I think it was already discussed.  They are mode registers.
MC_SEQ_MISC1      (MR1/MR0)
MC_SEQ_MISC3      (MR5/MR4)

I'd have to go back through my notes and code to confirm if MISC8 is mode registers as well.

I was just curious, as if thats the case, those would hold for example WRITE latency
and CAS latency, but it seems people never modify them, just use default value
(at least that seems to be the case for all the straps I saw circulating around)

That's because the mode register CAS latency must match the SEQ_CAS value.

[ku4eto]

I had found a new better memory strap for Elpida RAM. More than 1kh/s cryptonight xmr !

RX 470 : 1020 h/s  , sgminer @1250 core, @2100 mem, 66W (gpu-z)

                   942 h/s , @1150 core, @2050 mem, 52W (gpu-z)

777000000000000022AA1C00AC615B3CA0550F142C8C1506006004007C041420CA8980A9020004C 01712262B612B3715

I don´t know the eth hash speed.   Maybe somebody can check it ?

This strap is bad. Both of my cards bsod hard with it. Also its slower and runs with more power. Trfx and ras2ras are too tight.

[Branko]

I had found a new better memory strap for Elpida RAM. More than 1kh/s cryptonight xmr !

RX 470 : 1020 h/s  , sgminer @1250 core, @2100 mem, 66W (gpu-z)

                   942 h/s , @1150 core, @2050 mem, 52W (gpu-z)

777000000000000022AA1C00AC615B3CA0550F142C8C1506006004007C041420CA8980A9020004C 01712262B612B3715

I don´t know the eth hash speed.   Maybe somebody can check it ?

This strap is bad. Both of my cards bsod hard with it. Also its slower and runs with more power. Trfx and ras2ras are too tight.

I often see TRFC and ras2ras for elpida straps at that value

Its actually default value for 1500MHz strap

[ku4eto]

I had found a new better memory strap for Elpida RAM. More than 1kh/s cryptonight xmr !

RX 470 : 1020 h/s  , sgminer @1250 core, @2100 mem, 66W (gpu-z)

                   942 h/s , @1150 core, @2050 mem, 52W (gpu-z)

777000000000000022AA1C00AC615B3CA0550F142C8C1506006004007C041420CA8980A9020004C 01712262B612B3715

I don´t know the eth hash speed.   Maybe somebody can check it ?

This strap is bad. Both of my cards bsod hard with it. Also its slower and runs with more power. Trfx and ras2ras are too tight.

I often see TRFC and ras2ras for elpida straps at that value

Its actually default value for 1500MHz strap

My bad, i didnt say that properly. The trfc and ras2ras for xryptonight is nearly useless, you lose about 10h/s when going to 150 for example. But the power cons drops by 5watts. Thats for cryptonight.

Also, all trcd values are not properly done, the A value should be the same as the non A one.

[R0land]

I had found a new better memory strap for Elpida RAM. More than 1kh/s cryptonight xmr !

RX 470 : 1020 h/s  , sgminer @1250 core, @2100 mem, 66W (gpu-z)

                   942 h/s , @1150 core, @2050 mem, 52W (gpu-z)

777000000000000022AA1C00AC615B3CA0550F142C8C1506006004007C041420CA8980A9020004C 01712262B612B3715

I don´t know the eth hash speed.   Maybe somebody can check it ?

This strap is bad. Both of my cards bsod hard with it. Also its slower and runs with more power. Trfx and ras2ras are too tight.

I often see TRFC and ras2ras for elpida straps at that value

Its actually default value for 1500MHz strap

Make it better

My bad, i didnt say that properly. The trfc and ras2ras for xryptonight is nearly useless, you lose about 10h/s when going to 150 for example. But the power cons drops by 5watts. Thats for cryptonight.

Also, all trcd values are not properly done, the A value should be the same as the non A one.

Make it better ! I test so much straps, and that are the best, I can find.

[Branko]

I had found a new better memory strap for Elpida RAM. More than 1kh/s cryptonight xmr !

RX 470 : 1020 h/s  , sgminer @1250 core, @2100 mem, 66W (gpu-z)

                   942 h/s , @1150 core, @2050 mem, 52W (gpu-z)

777000000000000022AA1C00AC615B3CA0550F142C8C1506006004007C041420CA8980A9020004C 01712262B612B3715

I don´t know the eth hash speed.   Maybe somebody can check it ?

This strap is bad. Both of my cards bsod hard with it. Also its slower and runs with more power. Trfx and ras2ras are too tight.

I often see TRFC and ras2ras for elpida straps at that value

Its actually default value for 1500MHz strap

Make it better

My bad, i didnt say that properly. The trfc and ras2ras for xryptonight is nearly useless, you lose about 10h/s when going to 150 for example. But the power cons drops by 5watts. Thats for cryptonight.

Also, all trcd values are not properly done, the A value should be the same as the non A one.

Make it better ! I test so much straps, and that are the best, I can find.

Good idea

And shame you have no merit, despite helping people a lot, so let me give you first one