Quote from: papajo_r on March 22, 2021, 07:57:55 PM

Can somebody explain to me WTF is going on with my GPUS? Tongue

so using teamredminer latest version while having my vega 64 (nitro+) on stock bios core1100@ 820mV mem 1100 @ 850 mV

with memory timings cl 22 ras 35 rcdwr 12// rcab 44 rcpb 44 // rrds rrdl 5// faw 20 //REF 17550//RFC 247

I get 51Mh/s @ about 140 watts.

When I downgrade my bios to a vega 56 one (so from vega 64 TO vega 56, not the other way arround!)

1) the memory timings mentioned above dont work

2) this memory timings work 0 --RAS 26 --RCDRD 13 --RCDWR 12 --RC 38 --RP 12 --RRDS 3 --RRDL 5 --RTP 6 --FAW 12 --CWL 8 --WTRS 4 --WTRL 9 --WR 14 --REF 20600 --RFC 244

and with the above ("2") and core at 1100 810Mv and mem 1045 @ 82Mv The damn thing runs at 55Mh/s while consuming less power!!!

unfortunately I cant got more than 1045 though (despite when being on vega 64 bios I cat get easilty to 1100)

but the question is HOW DOES THIS WORK? can anybody give me a kinda technical explanation as to why with a downgraded bios + slower mem frequency the GPU performs so much better ?

Can I make it even better?

Hi, I have 2 Vega 64 Nitros. I tried flashing Vega 56 Pulse BIOS (because of the lower TDP I thought it would be e better idea) - it killed it. The card cannot be seen neither in Safe mode (no win driver) neither in DOS with bootable flash drive with ATI flash on it. Luckily there's a BIOS switch. I downloaded the Nitro+ version for 56 bios that has the same device ID 1002 687F - 1DA2 E37F and 180W TDP (socket and battery) the other ones had 220W as far as I remember. Guess what - I killed the second 64 as well. So I wonder which BIOS did you use to flash your Vega 64 nitro to Vega 56? I read somewhere else that another user killed his gigabyte 64 when attempting to flash 56 BIOS on it, so I guess this is not universal and there is trial and error involved.

On your other question. memory performance is a combination of Mhz and timings. It is really complicated to measure which exact parameter makes what difference. If you look at the timing numbers from your 2 cases - they are VERY different. I assume ethash uses the memory in a very specific way, but some timings compensate for others. So you can get similar results with very different combinations. I believe this could end up being so complicated that you actually need an AI to test and evaluate all the 100000s of combinations.

55 is close to a record high! If you get that stable I guess it makes only sense to look for ways to lower power consumption. Or if you dont care - simply dont mess around with it anymore

Having reached the max of a rig is a moment where I'd be pleased to open up a beer or two, sit back and relax Cool

Fujitzo

Posts

Merits