Any word on amd vega hash rates?

[DevlishNo]

Does brnstead's Excel-based-macro soft power play registry creator work for Vega 56s as well? Or does it apply only to Vega 64?

I'm asking because i'm using an MSI Vega 56 OC Airboost which currently does not have any stock BIOSes for the Vega 64 airboost version (using the MSI vega 64 reference version results in crashes).

Anyway since no one seems to be replying in this thread anymore, i'd just like to report that i've gone ahead and tried this myself and it works. It works for both VEGA 64s or VEGA 56s, original bios mod or not.

If you're still around, thanks @brnstead!

Could you please link the original post of his macro ?

[1714781501]

1714781501

[1714781501]

1714781501

[1714781501]

1714781501

[1714781501]

1714781501

[1714781501]

1714781501

[wudafuxup]

What do soft power tables do?  I don't get why they're needed.

You can't change the voltage on states P0-P5 on the core or P0-P3 on the Memory without modifying the registry. In essence the tables don't give you any extra hashrate but lower the amperage draw of the card. You can save like 60W per card.

Most people don't understand that. They simply open AMD settings and bring the power settings down -20% and think they are running at max efficiency.. which they are absolutely not. Not even close.

[BroganBloodstone]

What do soft power tables do?  I don't get why they're needed.

You can't change the voltage on states P0-P5 on the core or P0-P3 on the Memory without modifying the registry. In essence the tables don't give you any extra hashrate but lower the amperage draw of the card. You can save like 60W per card.

Most people don't understand that. They simply open AMD settings and bring the power settings down -20% and think they are running at max efficiency.. which they are absolutely not. Not even close.

I have made tables that are lower for those states and it made absolute no difference to power draw.

[treanski]

What do soft power tables do?  I don't get why they're needed.

You can't change the voltage on states P0-P5 on the core or P0-P3 on the Memory without modifying the registry. In essence the tables don't give you any extra hashrate but lower the amperage draw of the card. You can save like 60W per card.

Most people don't understand that. They simply open AMD settings and bring the power settings down -20% and think they are running at max efficiency.. which they are absolutely not. Not even close.

I have made tables that are lower for those states and it made absolute no difference to power draw.

then you did it wrong

this is mine for example... 1448mhz/880mv gives me ~1401mhz@875mv
complete system with 4 vegas 690watts from the wall on cryptonight - stable

[BroganBloodstone]

What do soft power tables do?  I don't get why they're needed.

You can't change the voltage on states P0-P5 on the core or P0-P3 on the Memory without modifying the registry. In essence the tables don't give you any extra hashrate but lower the amperage draw of the card. You can save like 60W per card.

Most people don't understand that. They simply open AMD settings and bring the power settings down -20% and think they are running at max efficiency.. which they are absolutely not. Not even close.

I have made tables that are lower for those states and it made absolute no difference to power draw.

then you did it wrong

this is mine for example... 1448mhz/880mv gives me ~1401mhz@875mv
complete system with 4 vegas 690watts from the wall on cryptonight - stable

I am not so sure. Every Vega is different. I do not think many vegas would run at 800mv on HBM for a starters. You seem to be under the assumption that all silicon is the same. I have identical AIB vega 56's and there is big differences between them. Some can so 980MHz on HBM, others cannot go above 890MHz.

[heavyarms1912]

What do soft power tables do?  I don't get why they're needed.

You can't change the voltage on states P0-P5 on the core or P0-P3 on the Memory without modifying the registry. In essence the tables don't give you any extra hashrate but lower the amperage draw of the card. You can save like 60W per card.

Most people don't understand that. They simply open AMD settings and bring the power settings down -20% and think they are running at max efficiency.. which they are absolutely not. Not even close.

I have made tables that are lower for those states and it made absolute no difference to power draw.

then you did it wrong

this is mine for example... 1448mhz/880mv gives me ~1401mhz@875mv
complete system with 4 vegas 690watts from the wall on cryptonight - stable

I am not so sure. Every Vega is different. I do not think many vegas would run at 800mv on HBM for a starters. You seem to be under the assumption that all silicon is the same. I have identical AIB vega 56's and there is big differences between them. Some can so 980MHz on HBM, others cannot go above 890MHz.

800mV here is not to the HBM2 memory.  HBM2 memory has fixed voltage; 1.26V for Vega 56 and 1.36V for Vega 64.
Reference versions use the same Samsung memory.  That's why reference Vega 56 can be bios modded (unlocked to 1.36V) and achieve 1100 Mhz frequencies like Vega 64.  Also, Hynix HBM2 is inferior to Samsung's (till now).

[Instant_exit]

So, stock 56 bios? Did you try 64 bios, im thinking of going back to stock if it doesnt make a difference really in hashes/watts.

/J

[Shnikes101]

What do soft power tables do?  I don't get why they're needed.

You can't change the voltage on states P0-P5 on the core or P0-P3 on the Memory without modifying the registry. In essence the tables don't give you any extra hashrate but lower the amperage draw of the card. You can save like 60W per card.

Most people don't understand that. They simply open AMD settings and bring the power settings down -20% and think they are running at max efficiency.. which they are absolutely not. Not even close.

I have made tables that are lower for those states and it made absolute no difference to power draw.

then you did it wrong

this is mine for example... 1448mhz/880mv gives me ~1401mhz@875mv
complete system with 4 vegas 690watts from the wall on cryptonight - stable

I am not so sure. Every Vega is different. I do not think many vegas would run at 800mv on HBM for a starters. You seem to be under the assumption that all silicon is the same. I have identical AIB vega 56's and there is big differences between them. Some can so 980MHz on HBM, others cannot go above 890MHz.

800mV here is not to the HBM2 memory.  HBM2 memory has fixed voltage; 1.26V for Vega 56 and 1.36V for Vega 64.
Reference versions use the same Samsung memory.  That's why reference Vega 56 can be bios modded (unlocked to 1.36V) and achieve 1100 Mhz frequencies like Vega 64.  Also, Hynix HBM2 is inferior to Samsung's (till now).

Correct - The voltage you see there in overdrive and wattman isn't really HBM2 voltage. Not really sure what it is tied to.
Functionally, it works basically like a floor voltage for the gpu states. It's why people say you need to have gpu mV at or just above your p3 memory mV

[dragonmike]

What do soft power tables do?  I don't get why they're needed.

You can't change the voltage on states P0-P5 on the core or P0-P3 on the Memory without modifying the registry. In essence the tables don't give you any extra hashrate but lower the amperage draw of the card. You can save like 60W per card.

Most people don't understand that. They simply open AMD settings and bring the power settings down -20% and think they are running at max efficiency.. which they are absolutely not. Not even close.

I have made tables that are lower for those states and it made absolute no difference to power draw.

then you did it wrong

this is mine for example... 1448mhz/880mv gives me ~1401mhz@875mv
complete system with 4 vegas 690watts from the wall on cryptonight - stable

I am not so sure. Every Vega is different. I do not think many vegas would run at 800mv on HBM for a starters. You seem to be under the assumption that all silicon is the same. I have identical AIB vega 56's and there is big differences between them. Some can so 980MHz on HBM, others cannot go above 890MHz.

800mV here is not to the HBM2 memory.  HBM2 memory has fixed voltage; 1.26V for Vega 56 and 1.36V for Vega 64.
Reference versions use the same Samsung memory.  That's why reference Vega 56 can be bios modded (unlocked to 1.36V) and achieve 1100 Mhz frequencies like Vega 64.  Also, Hynix HBM2 is inferior to Samsung's (till now).

Correct - The voltage you see there in overdrive and wattman isn't really HBM2 voltage. Not really sure what it is tied to.
Functionally, it works basically like a floor voltage for the gpu states. It's why people say you need to have gpu mV at or just above your p3 memory mV

My softpp table looks fairly similar to Treanski's. However, if I leave mem P3 state at 900 (or thereabouts) the miner will crash after a few seconds. I need to raise that value quite significantly (currently set at 1000) to have stable hashing.

If the HBM voltage value is locked, can somebody knowledgeable explain to me why that is happening?

Currently my 6x Vega 56@64 rig is hashing at 11600 h/s combined, 1408@900 core, 1115@1000 hbm, 1300W from the wall.

[Shnikes101]

What do soft power tables do?  I don't get why they're needed.

You can't change the voltage on states P0-P5 on the core or P0-P3 on the Memory without modifying the registry. In essence the tables don't give you any extra hashrate but lower the amperage draw of the card. You can save like 60W per card.

Most people don't understand that. They simply open AMD settings and bring the power settings down -20% and think they are running at max efficiency.. which they are absolutely not. Not even close.

I have made tables that are lower for those states and it made absolute no difference to power draw.

then you did it wrong

this is mine for example... 1448mhz/880mv gives me ~1401mhz@875mv
complete system with 4 vegas 690watts from the wall on cryptonight - stable

I am not so sure. Every Vega is different. I do not think many vegas would run at 800mv on HBM for a starters. You seem to be under the assumption that all silicon is the same. I have identical AIB vega 56's and there is big differences between them. Some can so 980MHz on HBM, others cannot go above 890MHz.

800mV here is not to the HBM2 memory.  HBM2 memory has fixed voltage; 1.26V for Vega 56 and 1.36V for Vega 64.
Reference versions use the same Samsung memory.  That's why reference Vega 56 can be bios modded (unlocked to 1.36V) and achieve 1100 Mhz frequencies like Vega 64.  Also, Hynix HBM2 is inferior to Samsung's (till now).

Correct - The voltage you see there in overdrive and wattman isn't really HBM2 voltage. Not really sure what it is tied to.
Functionally, it works basically like a floor voltage for the gpu states. It's why people say you need to have gpu mV at or just above your p3 memory mV

My softpp table looks fairly similar to Treanski's. However, if I leave mem P3 state at 900 (or thereabouts) the miner will crash after a few seconds. I need to raise that value quite significantly (currently set at 1000) to have stable hashing.

If the HBM voltage value is locked, can somebody knowledgeable explain to me why that is happening?

Currently my 6x Vega 56@64 rig is hashing at 11600 h/s combined, 1408@900 core, 1115@1000 hbm, 1300W from the wall.

Did you change your SoC limit to 1200 in order to capture the 1115mhz on the memory? I was under the impression that higher SoC limits to capture that higher memory frequency required a bit more voltage.
Honestly, I haven't messed with my Vega rigs in forever and have probably fallen behind in my understanding of where people are with these cards now days.
I'd try dropping the memory down to 1100 or 1050 @ 900 or 905 and see what your results are. You may end up with a similar hashrate at lower power draw.
 
Assuming blockchain drivers on reference cards?
Hopefully someone else can help you a bit more.

[dragonmike]

SoC is at 1200 (hence why 1115 MHz is even possible).
Dropping mem clocks do drop hashrate quite linearly. Was using blockchain drivers initially, now on latest Adrenalin. Less headaches and same performance.

[treanski]

i think the voltage setting on  p0-p3 memclocks dont have any effekt on memory voltage. out of my understanding vegas calculate their vdcc from the p0-p7 clock/voltage setting..with sum all voltages / amount of voltages you come pretty close to what you get, but clocks are mixin in there too somehow

as you see my settings give me 1401mhz coreclock...if i raise voltage on p7 it gives me also higher coreclock. just open overdriventool +gpuz + start miner and play around, you will see how vdcc changes

important thing is, p0-p6 states voltage needs to be lower than p7 to undervolt properly...if you have p5 state@1150mv you wont be able to go low with vdcc like me
download a power softtable editor and do your own registry mod...its like with bioses for polaris gpus -> the public ones suck

or try my...its for vega 56 & latest adrenaline driver, but should work with 64 too

vega56treanski.reg

[Shnikes101]

SoC is at 1200 (hence why 1115 MHz is even possible).
Dropping mem clocks do drop hashrate quite linearly. Was using blockchain drivers initially, now on latest Adrenalin. Less headaches and same performance.

Interesting. I noticed a slight drop (50 or so per card) in the one rig I tested moving from BC to the newest Adrenalin. This was maybe a week or two ago. Will look to see if there is newer drivers.
At that time, it wasn't worth the headache of moving my other Vega rigs over to the new drivers.

[Truthchanter]

Does the new driver require you to go into device manager and disable and enable all the gpus like the blockchain driver? (to get the max hashrate)

[dragonmike]

Does the new driver require you to go into device manager and disable and enable all the gpus like the blockchain driver? (to get the max hashrate)

Nope. No more of these shenanigans.

[jmigdlc99]

SoC is at 1200 (hence why 1115 MHz is even possible).
Dropping mem clocks do drop hashrate quite linearly. Was using blockchain drivers initially, now on latest Adrenalin. Less headaches and same performance.

What miner are you using with the latest drivers? CastXMR?

I would be Interested to know if hashrate is consistent. The way it currently is that hashrates drop when your temp goes up or if you start a new instance of another miner. New drivers with no headaches would be a breath of fresh air.

[Mashy81]

I changed to cast xmr from xmr stak.
Much more simpler and steady hashrate. Hashrate is definitely higher than xmr stak at the same power useage.
Cast xmr , latest drivers 1750hs for each vega56 @140w at the wall

[dragonmike]

SoC is at 1200 (hence why 1115 MHz is even possible).
Dropping mem clocks do drop hashrate quite linearly. Was using blockchain drivers initially, now on latest Adrenalin. Less headaches and same performance.

What miner are you using with the latest drivers? CastXMR?

I would be Interested to know if hashrate is consistent. The way it currently is that hashrates drop when your temp goes up or if you start a new instance of another miner. New drivers with no headaches would be a breath of fresh air.

Using CastXMR indeed. I'm experiencing no hashrate drops (unless indeed temperature goes up - but that's not what I'd call a "drop", just normal throttling), so essentially working as intended.

I'm using the latest Adrenalin 18.3.4 driver. Hashrate is 99% that of the blockchain driver, just as Glph3k announced.

[rednoW]

I changed to cast xmr from xmr stak.
Much more simpler and steady hashrate. Hashrate is definitely higher than xmr stak at the same power useage.
Cast xmr , latest drivers 1750hs for each vega56 @140w at the wall

I'm getting 1850h/s with sapphire pulse vega56 (hynix) 1456/970@0.875v 140watt according to corsair link
SrbMiner

[JuanHungLo]

1995 h/s, cryptonight v 7
Sapphire Vega 56 reference (air) (samsung) - stock, SRBMiner-CN-V1-4-6, "intensity" : 112, "worksize" : 16, "threads" : 2, 1407/935, Power Play table mod to 900mV for all except the clock at 1407=935, win 10, blockchain AMD drivers,
145w