From what I understand that the brand of memory can be a factor so Hynix cannot go as high as Samsung and also it was not recommended to flash the Vega 64 bios onto cards that are supplied with Hynix memory.
As a note I did have a few issues at 950 so it is now running at 940 with no stability issues but will try a few more things later to see what the limit is
In order to get decent clock speeds under Linux, where overclocking tools are non-existent, I went ahead and flashed my Vega 56 with a 64 BIOS. (Yes I have a card with Samsung HBM2.) Results under Linux are pretty good, with the power limit increased to 220w, clocks topping out at around 1520 Mhz and a lot less fan noise than my manual tweaks under Windows. Clock speeds aren’t quite as high, but I’m very happy with the reduced fan noise.
The HBM voltage can not be adjusted on Vega, the memory voltage on wattman does nothing for mV, instead it seems to control the minimum voltage drop for the GPU. E.g. if you get mV to 1000 GPU voltage will not drop lower than 1000mV, if set to 950 GPU voltage will not drop below 950mV, etc. Only way to get better memory performance is flashing Bios to 64 as you said.
Buildzoid did some experiments with the slider in wattman. Supposedly had some sort of success with it but as you say its not a direct memory voltage setting. it does help somehow though, at least that’s what he found.
Either way, if you’re not on samsung memory and vega 64 bios you’ll likely run into other limits well before needing to fiddle with that anyway.
And as above, heat is the enemy. You need to keep the HBM (and core, as they’re on same heat sink) as cool as you can to maintain clock.
I really want to liquid cool mine; vega is crying out for a custom loop to get better performance imho.
I have a Asus strix vega 56 which I water cooled using a NXZT kraken G12 and corsair H55 which I had on my old R9 390. With a few modifications I was able to make it fit on the 56 no problems. Works a treat as my GPU and memory temps generally never go above 55C.
The Asus has dual bios so I flashed 1 bios to 64 and works great. The card has samsung memory but I think I lost the silicone lottery as it clocks to 1100 but unstable. 1075 is good enough though. GPU clocks above 1550 most of the time, so pretty much runs like a 64. P6 state is at 1050mV and P7 at 1100mV of the top of my head at stock 64 MHz. Power limit is at +50%.
I recommend downloading HWinfo as it gives you monitoring of your whole system. It gives you the GPU hotspot temps. which are normally read higher that GPU and HBM. I found this to be a better indication of how your card is going as GPU and HBM temps might be low but, if GPU hotspot is high, then the card will most likely throttle and not reach higher clocks. Also AMD has not provided any information on what exactly GPU hotspt actually measures.
If your card is Hynix powered, you shouldn’t flash the Bios as it might have an issue with the clockspeed of the HBM.
Can you not adjust the fan speeds atleast?
Not under Linux. Or, if there is a way, I have no idea how to do it. Of course under Windows, I can use MSI Afterburner, Wattman or a number of other tools to adjust the speed. If I do, the fan gets quite loud. With the default Vega 64 fan curve, temps get up to around 80C before the fan starts really kicking in. With the Vega 56 BIOS they seemed to ramp up around 65C - 70C. I guess MSI wanted their Vega 64 to run quiet and hot.
Hey I don’t have vega so I am just in imagination mode.
I read a few times that heat is the problem particularly for the HBM. I wonder if a custom, possibly jury rigged, water block or rather set of them would help. The problem is space. What I am thinking is effectively have two cold blocks attached to each other but independant water ins and outs. One that sits on the core part and then a second small one that sits on the HBM allowing you to over cool one more than the other.
Sounds a little too insane though and definitely too niche to find so it would have to be a DIY endeavour. But just running in the idea that cooling the HBM separately might get you extra performance. I just wonder if at the stage the core being in close proximity would heat soak into the HBM should you manage to.push the memory fast enough to boost the core higher consistently.
I don’t think you could do both since the HBM is literally on the edges of the core!
But you shouldn’t have too much of an issue cooling it either way.
That’s a real shame… If you could adjust fan speed then you could get so much more out of the card! Some OC tools need to be made for linux ASAP!!!
Still want to, still haven’t done it. Will order the cooler at the end of the month so i HAVE to do it.
In theory - agreed.
In practice, if you can get good enough contact with ONE big block and cool it effectively you could probably get the same sort of result for less complexity.
What you need is some soft alloy for the base of the heatsink (like maybe gold? 
) that is a good conductor (like say, gold) so that the clamping force can ensure a good mating surface - an issue with Vega (and probably also Fury, Volta, and other HBM cards.) is that the HBM and die may be slightly different heights resulting in imperfect heat sink contact.
Then maybe go sub ambient with a chiller on your coolant.
I still plan to put at least one of my Vegas on water. I really should buy a block for it.
today the morpheusII arrived, so I will actually do a mod on the Vega
Awesome. Looking forward to your results…
I will try
to have stock results and then modded results and then with my devember project as results.
i dont wanna spend much time on “benchmarks” tbh
I meant using scores from that benchmark thread for comparison against your overclocking script
I would run that benchmark on my stock linux und stock vega56, again with optimized linux + modded v56, again with everything oc+++
So people could see if the “performance gains” (if there are any) are worth the time and money
