I tried your straps and unfortunately they didn’t do that great for me. I think the key difference is that you go for lower RC while sacrificing RCDRD, but I suspect RCDRD gives slightly better results. In my testing too low of RC can cause a crash, while too low of RCDRD only gives invalid shares.
Here are the straps I use for Samsung:
amdmemtweak --CL 20 --RC 37 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 20 --REF 65535 --RFC 248 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12
Here are the straps I use for Hynix:
amdmemtweak --CL 20 --RC 35 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 22 --REF 65535 --RFC 239 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 13 --RCDWR 12
In both cases I increase RCDRD to remove invalid shares.
And lastly, these are the “special” straps I use for my poorly performing Hynix:
amdmemtweak --CL 20 --RC 37 --RP 13 --WR 14 --CWL 8 --FAW 12 --RAS 22 --REF 65535 --RFC 239 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 20 --RCDWR 14
Hi, Great information. I am looking for a legend or something like it that explains what adjusting the different amdmemtweak values are and what they do. I have not been able to find any info.
So, after all this time, I finally got a gigashit card. They are awful haha. The one that I have can’t really take OC settings, but I have it set it 1050/875/930. That gets me a stable 44.5 without errors.
Anyone have any suggestions for a Vega 56 XFX with Hynix? Any straps I tried made it crash. Memory was running way too hot also. I’m going to replace the thermal paste and pads and see how it does after. Anytime the mem got to 95c the GPU would crash. I was doing everything I could to try to cool it so I could play with the settings and straps. It seemed like it might be stable around 42 mh/s without any strap mods. It was running around 102w in the software.
It seems like you should try changing the thermal pads first. I bet you’ll see a big difference there. Also, start with the most conservative OC profile you can find for finicky cards. I have one gigabyte card that I also can’t get to take straps, and I have it running at 44 stable.
Yup I replaced the pads and paste and the temps dropped by 25c on the mem, but I still couldn’t get it to stop crashing. I took off all the straps and tried to run it as conservatively as possible. It would mine for a few minutes then just randomly crash. Very frustrating. I’m not sure exactly what I can do.
I’ve got 2 Vega 56 Sapphires running. One with Samsung and it’s a beast doing 55 mh at 98w in the software.
The other is Hynix and a bit difficult to deal with but I got it at 48.4 mh and 86w in the software. Finally had it running stable for a couple days but it has the most invalid shares out of all my cards. Only 6 invalids but still it bothers me to see it lol. It runs extremely cool at 39c and 54c on mem.
You’re doing very well to have the memory running at 980, most of my Hynix can’t go above 960. At this point your best bet is to maximise efficiency by lowering power usage.
I’m currently running efficient settings and was able to get my core clock down to 1050 and VDD down to 775~790 for 99W in the best case scenario.
And if you’re really game, try setting this extra config in your flightsheet “–eth_config=B448” to shave off a few more W.
I don’t have it in my motherboard right now and I’ve forgotten exactly but it was something like 1000 core and 900 on the mem with 900 VDD. It would run for a few min and then just crash. I didn’t see any error messages to figure out exactly what was causing the issue.
I am using eth_config=a still if you use the"b" don’t use a number hive os will tune to a new number. Also turn down the core by about 25-50 since it uses all the memory and less core means less watts. Note: config b get unstable over time on some systems.
I am having a heck of a time trying to get my vega 56’s stable.
Right now one of them seems semi-stable, although not sure about long term. The other used to crash but I think I finally have it just getting a lot of invalids. I have tried changing basically everything, voltage, core, mem etc but nothing seems to even have an effect on the invalids.
