@Zr7s the 64s use quite a bit more wattage, which is why you see so many people flash to the 56, same performance with less wattage.
But if you are running the 64 bios you need to treat the card a little different. Mainly you need memory at 960 and leave SOC blank. with 960 mem that setting should get like 175-180.
@ikledude Ya I agree with LC 1440 / 960 or 1407 /930 if you want to turn some cards up.
So just found something about the “bad cards” that is mentioned a bit in the TRM guide. If you run the command amd-info through the hiveos webportal
Then click it once it finishes, you will see that the cards that are under preforming have true effective core value much lower than what its set at.
Looking at these two card’s:
the hashrates make more sence when looking at effective core
Pretty sure this is the “absurd scaling” mentioned in the max hash section of the TRM guide. So only way to improve those will be increase core more.
GPU 3 is my worst example of underperforming the set OCs but even with effective core only running at 1267 it needs nearly 900 vdd, ill have to see how much power it takes to get effective core up to 1300, guessing ill set core to ~ 1380
Update: my cards running RC 37, mem 847 (2 & 5) do not benefit pushing effective core above 1280, 1280 and 1300 both hash at ~179, below 1280 hash drops. So running got those at 1280 & 1277.
GPU 3 running RC 36 did see +8Mh increasing effective core from 1267 to 1300, but it cost 7 watts so not great.
Can confirm that , my hynix msi 56 blower card - is a bad one.
i have also been doing some testing with straps
all other 4 cards are around the 190mhs - with these straps ( samsung memory )
amdmemtweak --CL 20 --RC 36 --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
I have tried turning up the Core/Mem on the samsungs with no success. i could try the VDD but im comfortable for the time being with the MHS and Watts - would allways like them lower but i find at they crash with less VDD .
I tried various combos this morning with the aim of 200 but only got up to 195.*~MH 1400/900/960.
190.4 to 195 MH pushed the Wattage up 11.5W per card (at the wall, 92W for 8 cards).
thus also increased temps by 2deg C. per card.
a few rejected and one card with hardware rejected (the usual suspect card/memory; I could change the --RC 36/37 to resolve).
This may go back to the initial findings a few months ago it wasn’t worth the extra power required.
My quick test this morning at 1420/1440 failed.
I still need to try VDD above 905 up and keep memory down to save on the errors and memory heat.
Before I continue I need to make sure I’m not pulling too much from one of the PSU’s…
Hello guys, I mined a little bit of Ergo in May, and wanted to continue the job.
All my Vega 64 @ 56 can work easy at 170MH/s, but my Vega 56 it’s only working at 80MHs. I think I’ll have to divide the rig in two… Because mining ergo a 80MHs is not worth, I prefere Etherem.
I’m not paying attention to what HiveOs says about Ergo, I only care about pool report. With high clocks I can go over 200MHs according to HiveOs, but the real performance goes down.
Ya I was messing with the 203 Mh settings yesterday but its 3-4 vdd settings higher then my 198 setting so 5Mh costing 12 additional watts on both cards i tested.
If you do the math its still profitable to increase MH even with that bad of a Hash per watt.
12w x 24 hrs = .288kw/hr x $0.11 per kw.hr = 2.9 cents additional cost
additional 5Mh revenue currently = 12 cents per day
But its too hot out so going with 198 and only on the cards with great cooling.
I found that the effective core has a sweet spot for each memory level and going beyond that provides no additional hash, but often being 10-20 core below the optimal level can result in 1 to 3 less MH at same wattage. More for my notes later but here are the sweetspots I found.
I dont have anything special, for the pool I use Wooly Pooly or Nano and just pick the closest server.
Might want to try making a new flight sheet from scratch, when updating existing flight sheets there are a couple things that can carry over and cause the miner to not initiate, in my experience its easier to make a new one.
So I only notice i get a bump when socmhz goes up.
Am i changing the wrong setting?
When i see someone post 1100/900/870
I’ve set core mhz 1100, core voltage 900, mem mhz 870? Is that wrong?
I played around last night, got it 170mhz but after 4 hours crashed gpu dead, lowered clocks and now it ran 15 hours with 5 rejected @ 165 mhash
I think I’m missing a setting somewhere why must my mhz be so high?
@bobber1988 I think its just OCs. Helps to know which cards died so you know which one to fix could just be one or two need a bit more power. Less core or more vdd is usually the fix for crashes. When switching to TRM and doubling the hashrate its likely they need more vvd to stay stable than on the previous miner.
64 settings look fine to me.
GPU 4 and 6 have too low mem for 1400 core so either drop that core closer to 1300 or bump up mem to 930.
If it doesn’t start mining after like a minute on ERG I typically just reboot the rig then it starts mining. Or if a reboot doesn’t start mining I plug in a monitor and watch, you can usually see that 1 card doesn’t initialize so the miner doesn’t start, and thats the card that needs more vdd or less core.
@Zr7s Yep you got it 1100/900/870 is core, vdd, mem.
So rejected are just stale share in ERG so that’s just a share submitted after that block was solved, typically its just latency between your miner and the pool. Using a pool server closer to you will reduce those rejected, I wouldn’t really worry about rejected unless it gets over 1%,
That 64 strap is likely the reason you need the high values, it is very safe. Try changing these 3 values:
RC 38
RCDRD 13
REF 65535
since those clocks are so high tighter straps might not work as well. Maybe that new strap on 1300 or 1400 core 960 mem.
ugh either RC or RCDRD is too low, or REF is too high. I would roll back and try updating them one at a time see which one you card doesn’t like, RC is my guess.
I’ve been playing with core and mem mhz, makes no difference.
Rc and rcdrd I can increase individuality each by 1 and try each time but what about REF since it’s such a large number? Is there a fixed value to try decreasing?
