Topic: AMD Mem Tweak XL - Read/modify timings/pp/straps on the fly - page 17. (Read 58827 times)

Dear hulk999, thank you for your input. I can confirm the Gigabyte RX Vega 64 Gaming OC is problematic for Ethereum mining in particular. The memory configuration (mine is also Samsung), seems to refuse overclocking or higher performance straps. The best I could achieve was 990mhz on the memory using stock straps. In other words just over 38 mega hashes, like you. Yes, I have upgraded my bios from version F1 to F2.

Thank you for referring to team red miner and their vega timing options. I now for the first time am achieving 2200 hash mining Monero. Although I would prefer to mine Ethereum at 50 mega hashes!

So, if anyone in this thread has any clue why some of us with Gigabyte Gaming OC Vega's have trouble with Ethereum tuning please respond. Good day to all.

@Eliovp

I'm not sure if it's posted here earlier or not, but does this tool suppose to support 5700 series?
I've tried various github releases with 19.9.3 and it was weird...

Any plans to support?

this is just plainly wrong. here's v56 sapphire pulse with no changes to the bios running @950 and producing over 42MH



and here's the vega56 refs giving about the same hash



Also note that the v56 pulse is a head gpu with display dummy in use (more stability needed to support display)
and v56 refs are not really optimized now (these can do close to 44)


And here's fine-tuned vega64 ref for comparison:

1) OK. Thanx. I got 2 rigs at 6 cards each, all cards have same problem. My cards stable at default timings at 950Mhz Memory only and give 38Mh ETH. Overclock 1000Mhz Mem at default timings and system hangs  in ETH mining. weak or conceal timings at 950Mhz Memory and system hangs in ETH mining.  in CNR-mining everything is OK.
2) Nobody has the same cards here in ETH mining with memory tweak?
[/quote]

You are not using any parameters in the claymore bat file for straps. Are you?
I have Sapphire RX Vega 56 (Flashed to 64 bios) with weak timings and 16*16:AAA gets me around 2150 H/s in CRv8 and 42-43 MH/s in ETH. (1100MHz mem)
Didn't face any issue like this but it seems that 38 MH/s is right for 950MHz mem.
[/quote]

1. No extra parameters in Phoenix 4.7 or Claymore 15.
2. for sure 38 MH/s is right for 950MHz mem. But this mem almost doesn't overclock in ETH at any timings.
3. in CNR-mining the same rig works at 1408 Core -1050Mem Mhz and another rig works at 1408 Core -1100Mem Mhz , everything is OK

You are not using any parameters in the claymore bat file for straps. Are you?
I have Sapphire RX Vega 56 (Flashed to 64 bios) with weak timings and 16*16:AAA gets me around 2150 H/s in CRv8 and 42-43 MH/s in ETH. (1100MHz mem)
Didn't face any issue like this but it seems that 38 MH/s is right for 950MHz mem.

OK. Thanx. But DAG generation goes every several hours or less on Nicehash so that delay option is worthless. Any other suggestions? Are there no owners of Gigabyte VEGA64 GAMING-OC at all here? Am I the only person who owns such a crap? All my 12 GPUs have same problem with memory tweak and ETH mining
[/quote]

make your cards stable without applying tight timings, core should run at or below 850, load the cards with default timings, run winamdtweak --current > current_mem_timings.txt and improve from there.
btw 12gpu rigs are way harder to manage - takes somewhat exponentially more time to fine-tune comparing to two rigs with 6 cards each
[/quote]

1) OK. Thanx. I got 2 rigs at 6 cards each, all cards have same problem. My cards stable at default timings at 950Mhz Memory only and give 38Mh ETH. Overclock 1000Mhz Mem at default timings and system hangs  in ETH mining. weak or conceal timings at 950Mhz Memory and system hangs in ETH mining.  in CNR-mining everything is OK.
2) Nobody has the same cards here in ETH mining with memory tweak?

make your cards stable without applying tight timings, core should run at or below 850, load the cards with default timings, run winamdtweak --current > current_mem_timings.txt and improve from there.
btw 12gpu rigs are way harder to manage - takes somewhat exponentially more time to fine-tune comparing to two rigs with 6 cards each

OK. Thanx. But DAG generation goes every several hours or less on Nicehash so that delay option is worthless. Any other suggestions? Are there no owners of Gigabyte VEGA64 GAMING-OC at all here? Am I the only person who owns such a crap? All my 12 GPUs have same problem with memory tweak and ETH mining

you may have memory not stable enough for intensive stage of DAG generation. simple solution - put delay of 2 minutes or so, in the Task Scheduler for execution of AmdMemTweak
my vegas are stable on ethash from 42(non modded 56) to 49.5. you may have shitty gpu:)

for Polarises (I have all bios modded, so most of the timings already tight) two major params should be modified:
REF(delay between memory refreshes in ms) and RFC(number of clocks per each refresh)
same as  with Vegas: first capture the current memory timings with --current option  

on ETH I don't go above REF = 10ms, normally I get about 31MH per 570, sometimes it can go up to 32+ (if I have patience)

Hello
Did anyone use Gigabyte VEGA64 GAMING-OC dualfan WINDFORCE Samsung Memory  with memory tweak and ETH mining?, what are your settings? I use ppt, With CNR mining for my cards everything  is OK for settings Weaker Vega 64 or flashed 64 (Samsung) from CN_MAX_YOUR_VEGA.txt from Teamredminer i.e. --CL 19 --RAS 30 --RCDRD 12 --RCDWR 6 --RC 44 --RP 13 --RRDS 5 --RRDL 5 --RTP 4 --FAW 18 --CWL 6 --WTRS 4 --WTRL 9 --WR 15 --WRRD 1 --RDWR 18 --REF 17000 --RFC 248 and my voltage is 840mV for Mem and Core, 6 cards at 1408 Core-1050Mhz Mem and give 13600 h/s, about 2250 h/s per card at --cn_config=14*14:AAA, absolutely stable for weeks.
With ETH mining my cards work only with default  timings and  give max 38Mh per card at 990Mhz core- 950Mhz Mem in Phoenix 4.7 and Claymore 15. I tried different timings, latest bios from official gigabyte site and also set higher voltages up to 950mV for Core and Mem, but eth miner with not default timings just hangs in a minute or less after Dag generated and rig restarts . When i set Mem more than 950Mhz, miner hangs in ETH mining at any timings, including default ones. My Gigabyte VEGA64 GAMING-OC have same default timings as any Vega 64 Reference.
My 2xPSUs are OK, they are 1kW GOLD. I have same problem with ETH mining using only one card from that rig

The stop line for your card seems to be 930 MHz. Don't OC much as it can cause stability issues or memory errors.

0:1500, 0:1800
Timings with '0' start are active when you have Samsung memory in the card.
1:1500, 1:1750, 1:1800
Timings with '1' start are active when you have Hynix/Micron/Elphida memory in the card.

So you have to modify strap timings starting with '1'. Also, check if they are applied successfully.

Hi
i has pulse rx5700 hunix
mining eth phoenix miner 53+ mh
mclock 930, and if up 940+, miner drop to buck to 930
where this stop line? changing possible?


timing has 2 position:
1500 1800
and 1500 1750 1800
why?

change both 1500 to 1800, reload driver - not effect, mining 53+

READ THE POST ABOVE YOURS--

My post just above yours should also be applicable to RX 480 cards.       --scryptr

Hello, masters) Could you help me with amdtweak settings for my rx480 samsung 8 gb? Thanks in advance)))

GOOGLE "--REF 30" --

Most non-Vega RX cards are at a default "--REF 5".  If you simply increase this single setting from "--REF 10" to "--REF 30", you should see a noticeable increase in hash rate.  Some have tried much greater increases, but I think they are less stable.

Personally, I use the string "--REF 30 --RC 70 --RFC 157" on my RX 560s, 580s, and 590s.  I'm not using any 570s currently, but I think the string would work.  It gives a small boost (~1+MH/s) per card.  Vegas are different.

--scryptr

any settings for a rx580 with orginal bios, and samsung memory??

Well... I'm not sure its still relevant (apparently it was for me), as I've missed good few months since this wonderful tool has been released.

Anyway it seems there is some confusion with applying settings to the GPUs so let me state my findings:
I use it under win10 (1607 Anniversary edition) with 18.6.1 driver for both Vegas(with PPTs) and Polarises(modded bioses).
For both I also use ODT for clocks and voltages, also I don't like disabling any stages.
Basically by switching default REF=5 for Polarises to 10 and reducing RFC a bit I can get over 3% increase in ethash output.

What I noticed however is that while REF is applied more uniformly (I can 'probably' do it before running the miner - before GPUs are in their target frequencies range)
The RFC is not applied. So in order to get it properly working I have it(CLI version) scheduled to run after some delay, when the miner is already running. 

@Eliovp

Great tool, thanks a lot!

Something weird though: I was updating my Polarises in remote location using cli version.
And without taking into account much higher default tRFC on Hynix I updated a mixed bunch of GPUs with "inappropriately" low tRFC of around 130 cycles.
And then I got myself a dozen "bad" GPUs with Hynix memory - these GPUs started to produce almost exclusively bad shares. Whatever I did after that, nothing helped: restoring the values to the default, restarts (including resets from watchdogs), driver reinstall - nothing. The only thing that actually solved this problem was power down/up cycle. (and I know this for sure, because I had exactly the same issue on 4 different machines and my usual help was away for a few days)

So I'm curious, can you clarify how this could be possible?

Ok, I will try. Thanks for help. Hope devs will improve this feature soon.