AMD Locked Radeon RX 7000 Series Power Play Tables, Limiting Overclocking

News
By Aaron Klotz
published

One popular method of GPU overclocking won't work.

AMD RX 7000 Series Reference Card
(Image credit: AMD)

Overclocking expert @Buildzoid1 on Twitter recently shared a post discussing his frustrations with AMD's overclocking capabilities on the RX 7000 series. Apparently, AMD has locked power play table manipulation on its new GPUs, preventing enthusiasts from customizing power and frequency curves outside their official specifications.

See more

Power play tables provide an alternative method of changing a GPU's clock speed, power, and voltage behavior through the Windows registry instead of a BIOS mod or a BIOS swap. This can be a safer way to run GPUs outside of their official limitations since it prevents any risk that might occur when flashing the BIOS.

If you are wondering why AMD's overclocking feature set in the Adrenalin software isn't good enough; adjusting power play tables - or manipulating the BIOS, has a couple of benefits. One is that users can bypass firmware-related limitations on the GPU hardware, and push the hardware as hard as it will go. Another is to optimize the voltage/frequency curve for better efficiency. Either way, it allows diehard overclockers and enthusiasts to tweak their GPU's behavior in an extremely detailed way that is not accessible with normal overclocking software.

Unfortunately, we don't know exactly why AMD has shut down power play manipulation on its Radeon RX 7000 GPUs, especially when its previous generation cards have this ability. But we might have a clue as to what's going on from an AMD engineer's who is on Reddit. According to a Reddit post discussing RX 7000's power consumption issues, user Falk_csgo asked about driver limitations related to GPU overclocking; AMD engineer AMD_PoolShark28 responded by saying "Architecturally RDNA3 has some changes to how PPTables are manipulated..."

At the very least, we now know that the way users can edit and access RX 7000's power play tables has been changed since the RX 6000 series. To speculate a bit, this is probably the reason why the power play tables were locked on RX 7000 series.

Hopefully, this means the power play table lockout is only temporary and AMD will restore access at some point. But we really don't know. 

Aaron Klotz
Aaron Klotz
Freelance News Writer

Aaron Klotz is a freelance writer for Tom’s Hardware US, covering news topics related to computer hardware such as CPUs, and graphics cards.

See more GPUs News
16 Comments Comment from the forums
  • red.deugemo
    I think it's so the aftermarkets card aren't way faster than the OEM ones.
    Reply
  • -Fran-
    I mean... AMD just got out of the gate a new way to make GPUs. It should be expected they can't expose everything from the very first design to people, or is that too stupid/unreasonable to expect nowadays?

    Buildzoid's knee-jerk reaction is childish this time, I'd say. Has AMD said they won't enable finer grained control ever again? If so, then the reaction is justified, but until then, overblown.

    Regards.
    Reply
  • SyCoREAPER
    Probably because they are squeezing every damn Watt out of it from the factory there isn't much overhead left.
    Reply
  • shultzyssand
    Agree w Fran. Chiplet-based; two different processes across the two chips; new registry entries, etc.

    Most likely independent power ramping logic for the MCD and GCD on the board, so more stuff in the registry as well.

    For a product release this important it makes perfect sense the tables are initially locked. I don't think it'll be permanent.
    Reply
  • digitalgriffin
    Fran is likely onto something.

    Is it possible the power play tables aren't done yet as this is new process tech with separate chips? I can see the old ones being left behind till they finalize the new data layout. Like the prefetch, maybe it just wasn't ready in time.
    Reply
  • UWguy
    Two words: spontaneous combustion
    Reply
  • Soul_keeper
    I miss being able to flash a custom firmware on the cards.
    I would write my own ADL software to overclock/undervolt and test stability, once I found the best setup i'd create a custom firmware and flash it. This is the ideal solution in my mind. Having to constantly apply settings every boot sucks. And the powerplay tables are hit or miss, the methods and control change with every new card. I still can't find offsets for 6000 series to modify the ppt. And even if I did manage to guess which offsets applied to which voltages/clocks I would still not be allowed to reflash a bios with the modified ppt.

    The original GCN 7000 series were the last completely unlocked cards we had from AMD.
    Reply
  • Vanderlindemedia
    Solder on some resistors and bypass the way the logic deals with how much current the card is pulling. What kind of a hardcore overclocker are you if you need powerplay tables?
    Reply
  • Vanderlindemedia
    Soul_keeper said:
    I miss being able to flash a custom firmware on the cards.
    I would write my own ADL software to overclock/undervolt and test stability, once I found the best setup i'd create a custom firmware and flash it. This is the ideal solution in my mind. Having to constantly apply settings every boot sucks. And the powerplay tables are hit or miss, the methods and control change with every new card. I still can't find offsets for 6000 series to modify the ppt. And even if I did manage to guess which offsets applied to which voltages/clocks I would still not be allowed to reflash a bios with the modified ppt.

    The original GCN 7000 series were the last completely unlocked cards we had from AMD.


    Yeah but new cards are pretty much already operating to it's best possible. There is not alot of headroom if you start manually OC'ing and undervolting. Best way in my opinion is to make a working profile in AMD software, and load it upon every boot with Windows. Thats how i do it.
    Reply
  • hannibal
    Quite possible that this has similar problems as 5800x3d. Balancing the memory chiplets and main gpu can be quite complicated and free tinkering can cause a endless nest of trouble…
    Based on what we have seen so far from memory chiplet desing, there are bottlenecks it causes and allready there are memory OC problems partially because bad AMD OC program (MSI program work much better) maybe partially because the separater memory controller and the fabric in between need to be balanced. We know too little so far to be sure, but new architecture, new restrictions combination definitely suggest to that.
    Reply