the 56 hynix and the 56 samsung are normally from MSI model, i have flash the bios with the bios Saphire
Anyone use MSI vega56 air boost(Ram Hynix)?
I can’t reach more than 44MH/s
Any timming suggestion for me?
core 1028 mem 830 on this timming when push mem to 840 alway gpu dead lower than 10 min
amdmemtweak --CL 21 --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 17 --RCDWR 12
When used 3 timming below this ram can go only 800-830 can get hast maximum only 43.85MH/s
amdmemtweak --RC 35 --RP 13 --RAS 22 --REF 65500 --RFC 148 --RRDL 4 --RRDS 4 --RCDRD 12 --RCDWR 4
amdmemtweak --CL 20 --RAS 26 --RCDRD 12 --RCDWR 12 --RC 38 --RP 12 --RRDS 3 --RRDL 5 --RTP 6 --FAW 12 --CWL 8 --WTRS 4 --WTRL 9 --WR 13 --REF 24000 --RFC 239
amdmemtweak --CL 20 --RAS 22 --RCDRD 12 --RCDWR 12 --RC 36 --RP 12 --RRDS 3 --RRDL 5 --RTP 5 --FAW 12 --CWL 8 --WTRS 4 --WTRL 9 --WR 13 --REF 65000 --RFC 249
yeah, i shaved 10mV vrom all vegas and 15 from all my 12x 5xx cards, total consumption dropped a bit, and what was suprising to me is that many gpu´s hash went up a touch with lover vdd.
I think im gonna let rigs just roll on, looks like eth is down and rewards are up for now so -
I´ll look from the side for a while if u guys come up with some new trick´s with straps
Nice, must be as low u can go! no way i managed to go that low…is that MVDD 1265 that makes that possible?? My hynix died at 815…vega´s! , too many factors effecting each others
good hash, can u please share your straps for reference. i have same card with different vbios, so im gonna give it a try with your vbios…this vbios must be why i havent been able to sqeeze more from that red devil…
If it’s true I shouldn’t have sold my Gigabyte Hynix running stable 52mh
Try these settings of my best MSI Vega64 liquid
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
Hi. Im going crazy with my vega 56. Ive noticed you have One with the same memory than mine, 113-376XL-UO2.
Every strap ive tried makes the GPU dead after some minutes miming
Can you Share the BIOS and straps you are using for this?
Best regards
amdmemtweak --RC 37 --RP 12 --WR 13 --CWL 8 --FAW 12 --RAS 22 --REF 65535 --RFC 239 --RTP 6 --RRDL 5 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12
amdmemtweak --RC 37 --RP 12 --WR 13 --CWL 8 --FAW 12 --RAS 22 --REF 65535 --RFC 239 --RTP 6 --RRDL 5 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12
Let me know how it works out
I tried several different straps and customized versions, your starps were too tight for this card…same card and same vbios,thats the way it is with gpu´s, u have a good one, mine is good allso but dosent tolerate high mem. Thats max mem and core. This is with different set of straps. im really happy that this samsung is on the same level with others now. Will see if it hold up in long term, i have had similar values on this one before witn 54.8mh, but those were not good for 24/365 ops and got too tight when ambient temp went up…all variables…
im really happy that this samsung is on the same level with others now. many thanks for sharing info
Perhaps too low voltage? You have to share your settings to let us help.
I’ve got one of those, but can’t get it over 48MH/s for now either.
My mem is higher though, but rcdrd is still high as 19:
core 1080 vdd 840 mem 940 mvdd 1270
strap amdmemtweak --CL 20 --RC 36 --RP 13 --WR 13 --CWL 7 --FAW 12 --RAS 23 --REF 65535 --RFC 248 --RTP 6 --RRDL 5 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 19 --RCDWR 11
I like the MSI Airboost vbios, but not their actual GPU’s
Thanks this very nice strap now i try on mem900 get 46.33 if stable more than a day i will increase.
once you found your max stable mem, you can try do decrease rcdrd step by step; this should keep you busy for the next few days
Runra and Oni,
i have one model of MSI Vega56 flashed with the bios of the saphire model
the straps are
and the result is
when i go above 955/960 the card crash invalid
Yes, that’s normal. Hynix memory doesn’t go as high as Samsung memory.
You can try lower the rcdrd for better hashrate.
Here my sapphire 56 hynix values, you can try if your card can take these
amdmemtweak --CL 20 --RC 36 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 22 --REF 65535 --RFC 248 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12
If crash change RC to 37
mem 960 vdd 830 core 1050
so could’t resist when i got an opportunity to aguire another vega. Radeon 56 ref, bit of issues when i added that to rig, ended up flashing hive again and after, all issues gone. Seems like a good gpu, i can understand now why some of u guys have whole rig full of these
#EDIT#
Below was a test on Samsungs, but due to lack of sleep with wrong straps to begin with, hence I couldn’t get their mem clocks higher. The test is thus not so relevant. Take from it what you want.
#/EDIT#
I’m experimenting a bit with the SOC Frequency field in the OC screen. It sets the SOC state of our GPU’s.
The normal Vega 56 SOC states are :
- State 0: 600
- State 1: 720
- State 2: 847
- State 3: 960
- State 4: 1028
- State 5: 1107
Mem state and SOC state are linked. If you set your memory clock to 940MHz, the SOC state will set to the state right above, so 960 in this case. If you have mem 1020, SOC will set to 1028; if you have mem 1028 it will not set SOC to 1028 (that’s not above but equal) but to 1107.
I have now manually set some of my Vega’s one state higher in the OC screen, so for mem 940 I have not set to 960 but to 1028; for mem 1020 I have not set to 1028 but to 1107; those who were already SOC 1107 I left alone.
I can already tell that all those that were set to 1107 didn’t last long; they crashed pretty fast.
But all my GPU’s with somewhat lower clocks between 900 and 950 for which I have set the SOC to 1028 are running for 24h now without rejected shares (as they did before), AND each with a few tenths of a MH/s higher!
So although it doesn’t seem work on the high frequency Vega’s, it is another way to squeeze some lower frequency ones
Needs more experimenting though.