Oh, I taught it helped in keeping the core stable which in term kept the memory stable thus also less invalids
If I remember correctly it sets all powertables to the clocks and voltages we set in the OC.
Hey guys, thank you for sharing your results and progress during the whole time! I’ve been lurking for the past 2 weeks for proper straps and settings for Vega 56 cards, tweaked, and achieved really stable results by now. I thought I would share my settings with you guys to help anyone who has troubles with a Vega 56 Card (Samsung & Hynix memory). All my cards run on stock bios by the manufacturer. I should note that I DON’T use HiveOS since I don’t have many cards at my disposal and I just got started. However this forum and you guys provided tons of useful information so I am just returning the favor!
Cards:
2 x PowerColor Radeon RX Vega 56 Red Dragon 8GB (Memory Type: Samsung)
1 x PowerColor Radeon RX Vega 56 Red Dragon 8GB (Memory Type: Hynix)
Performance:
Samsung 1/2: +55.3MH/s @ 115W = 0,481 MH/W (P7: Core 1200MHz/813mV P3: Memory 1020MHz/820mV)
Samsung 2/2: +56.0MH/s @ 121W = 0,462 MH/W (P7: Core 1200MHz/870mV P3: Memory 1025MHz/820mV)
Hynix: +53.2MH/s @ 112W = 0,475 MH/W (P7: Core 1160MHz/850mV P3: Memory 970MHz/850mV)
Straps:
(see attachments)
Cooling:
AirCooled, however I adjusted the fan curves in a way that the HBM2-memory stays below 60°C. It might be a bit of an overkill, however I prefer my cards to stay as cool as possible and avoid any thermal degradation and bottlenecks.
OS:
Windows 10
Drivers:
AMD Radeon 21.5.1 (most recent)
Software:
AMD Memory Tweak by Eliovp & A. Solodovnikov (v0.2.3)
HWiNFO (v7.02)
Miner:
TeamRedMiner (v0.8.2.1) TRM seems to be the most efficient and fastest miner so far. I personally use the A-Mode since the B-Mode tends to be a bit more unstable. However, you should give B-Mode a try to gain a bit more performance and efficiency and see if it works for you!
Algo:
Ethereum (ethash)
In General:
Feel free to ask any questions concerning the straps and clocks. As you may know, every card is different, and my settings might not directly apply to your card specifically. You might need to adjust your Core/Memory voltages as well as your Core/Memory frequencies. (i.e. my seemingly identical Samsung cards work on different core voltages).
For that matter you should read the Tuning Guide “ETHASH_TUNING_GUIDE.txt” by TeamRedMiner for a better understanding.
If you see any space for improvement or optimization let me know!
Shoutout to @ss-silver @Smining570 @Rocco1180 @vsimovic for providing awesome straps and info, thank you guys! Keep up the good work, cheers!
5 Likes
Nice break down that’s cool and great performance on those!
I finally got my Vega 64 (GPU 4) to 56mh, under 120w today pretty excited about that 
Same strap, just new core and mem.
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 18 --RCDWR 12
Also found where the original REF 65535 strap came from (he ran a 64 flashe dto 56 for a day at rcdrd 12 for 57mh but threw 123 invalids:
While doing some testing I was able to hit 57.5 with rcdrd 18 mem 1100, 825mv but started throwing invalids after 2 min. Tried Rcdrd 20 and it was still 57mh but invalids after 5 min.
3 Likes
I see, yeah the tREF 65535 seems off, since it should always be a whole multiple X of the stock tREF.
I tested it too, and it looks like increasing tRCDRD dramatically reduces the hashrate and by increasing the MEM clock to compensate for the dropping hashrate the GPU gets unstable. I bet on a longer run the GPU would simply crash.
Happy the Vega64 worked out for you!
I will try to find more info on straps in the next view days. If i find somethingthing useful i will let you guys know.
2 Likes
What i have find u are right about RCDRD. I have managed to rised Mem with higher RCDRD value but without higher Mem hash goes down… its joggling with different values & strap combos. Thanks for doing some rechearch, have tried allso but most of it goes beond my paygrade when trying to apply in real life what iv learned about the straps by reading about theory how they affects gpu. Please do let us know what u find out 
I’m a little stuck with my MSI Vega 56 Airboost. 46.16 MH / s does not increase. I flashed it, strapped it, but it doesn’t change. can you help me.
from the mem info it looks like the vbios might be wrong one xxx.xxx.xxx. Maybe if u doubble check that its the right vbios and try to reflash the card? …remember that flashing can brick your gpu 
The MSI Airboost vbios gives this xxx-xxx-xxx version so it might be correct.
How did you apply the straps? By the looks of it it wasn’t in the Tweak section of the OC window; did you add them to the command line or in the old xinit.user.sh?
I see you have very high RCDRD of 22; is that the lowest you can get it without loosing stability?
I have this strap on my MSI Airboost:
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 18 --RCDWR 11
Make RCDRD higher or lower according to long term stability
1 Like
I see you’ve got them running allright 
Now you should play a bit with the OC’s and straps to get the GPU’s stable and/or the hashrate higher.
New to HiveOS but has been troubleshooting this for a few days now.
My Vega 64 FE is working well, but regular Vega 64 is reporting the following
May 14 12:18:06 CF002 kernel: [ 31.259832][ T948] [drm:amdgpu_atom_execute_table_locked [amdgpu]] *ERROR* atombios stuck executing D7E2 (len 279, WS 16, PS 4) @ 0xD8E6
May 14 12:18:06 CF002 kernel: [ 31.260645][ T948] [drm:amdgpu_atom_execute_table_locked [amdgpu]] *ERROR* atombios stuck executing A7FA (len 219, WS 8, PS 4) @ 0xA8C5
May 14 12:18:06 CF002 kernel: [ 31.261482][ T948] [drm:amdgpu_atom_execute_table_locked [amdgpu]] *ERROR* atombios stuck executing 9BB6 (len 381, WS 0, PS 8) @ 0x9C3B
May 14 12:18:06 CF002 kernel: [ 31.262343][ T948] amdgpu 0000:00:0a.0: amdgpu: gpu post error!
May 14 12:18:06 CF002 kernel: [ 31.262871][ T948] amdgpu 0000:00:0a.0: amdgpu: Fatal error during GPU init
May 14 12:18:06 CF002 kernel: [ 31.278891][ T948] amdgpu: probe of 0000:00:0a.0 failed with error -22
May 14 12:18:06 CF002 hive[851]: Saving /sys/class/drm/card1/device/pp_table
May 14 12:18:06 CF002 kernel: [ 31.302693][ T1009] pci 0000:00:0a.0: No more image in the PCI ROM
May 14 12:18:06 CF002 kernel: [ 31.308347][ T1009] pci 0000:00:0a.0: No more image in the PCI ROM
May 14 12:18:06 CF002 systemd[1]: Started Hive Watchdog Service.
May 14 12:18:06 CF002 hive-watchdog[1021]: Watchdog disabled in config (/hive-config/rig.conf)
May 14 12:18:06 CF002 hive-watchdog[1021]: MINER RESTART: minutes
May 14 12:18:06 CF002 hive-watchdog[1021]: REBOOT: minutes
May 14 12:18:06 CF002 hive-watchdog[1021]: MAX LOAD AVERAGE: .0 (min 13.0, def 28.0)
May 14 12:18:06 CF002 hive-watchdog[1021]: OK LA(5m): 0.34 < 28.0, LA(1m): 1.40 < 56.0
May 14 12:18:07 CF002 hive[851]: Cannot find DRI instance for pci:0000:00:0a.0
May 14 12:18:07 CF002 hive[851]: Cannot find DRI instance for pci:0000:00:0a.0
May 14 12:18:07 CF002 hive[851]: ERROR: can not match card id for GPU 00:0a.0
Also, i see people refering to an “Vega HiveOS”, is there a seperate place to pick up that image?
Looks like you guys here are flashing your VEGA 64’s with 56 ? Should i try and do that to resolve above issue?
Also, should i flash my Frontier Edition Vega 64 with 56 as well?
So original question, I have bought many (15) Vega cards and none get over 960 mem, can you enlighten me on what I am missing? Above 960 they drop to 8-9mh
1 Like
some more info about your cards would be apprciated as requested before…
indeed; which straps have you applied on those?
straps and clocks go hand in hand; some straps work better with lower clocks, some straps work better with higher clocks
I have had best luck with amdmemtweak --CL 20 --RC 36 --RP 12 --WR 13 --CWL 6 --FAW 12 --RAS 22 --REF 65000 --RFC 249 --RTP 4 --RRDL 5 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 11 --RCDWR 5
I tried 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 18 --RCDWR 12
but just got lower hashrate and still go down to 8 mh at anything above 960 mem