MSI Afterburner developer adding 'triple channel voltage' support for future MSI RTX 50 graphics cards
The updated software allows access to two new voltage parameters.

MSI Afterburner's sole developer, Alexey Nicolaychuk, is working on a new update for the app that will expand its voltage support for overclocking enthusiasts. In an update on the Guru3D forums, Nicolaychuk revealed that he's working on "triple channel voltage" aimed at future MSI graphics cards that will expand voltage control beyond just core voltage manipulation.
Triple-channel voltage control will allow users to control two additional voltage parameters on future MSI graphics cards: memory voltage and aux (MSVDD) voltage. Core voltage control also gets an upgrade, boasting a direct PWM access mode featuring an expanded 100mV offset range for these cards.
This is a significant upgrade over Nvidia's default voltage controls found on its Founders Edition graphics cards and many third-party cards. GPU voltage controls by default do not allow access to memory or auxiliary voltage control, and core voltage control is limited to a 20mV offset.
This new tech will be limited to future MSI RTX 50 series graphics cards, at least for now. Nicolaychuk frustratingly explains that this tech can't be adapted to other graphics card models (including outgoing MSI RTX 50 series GPUs), due to limitations in Nvidia's default voltage controls. RTX 50 series graphics card models that use Nvidia's reference design blacklist I2C devices at the driver level, making voltage controllers invisible to software trying to access them through the I2C bus.
However, Nicolaychuk clarified that future GPUs other than supported MSI models could work with triple channel voltage control, as long as those GPUs don't adhere to Nvidia's reference design and feature modified software to access the I2C bus. We'll have to wait and see if any brands other than MSI decide to make RTX 50 series GPUs with these modifications. These GPUs will likely be cards focused on extreme overclocking.
Memory voltage control is arguably the most interesting addition of the new triple-channel voltage control. Modern Nvidia graphics cards can be heavily memory-bound depending on the application, and can gain as much performance from memory overclocking as GPU overclocking alone. Having memory voltage control will allow overclockers to boost the voltage of Blackwell's GDDR7 memory modules, something that hasn't been possible with previous graphics cards.
The improved 100mv offset range for GPU core overclocking could be promising, but Nvidia's latest implementation of GPU voltage offset limits users to the maximum voltage the GPU is allowed to pull at stock speeds. Limiting voltage offsets on the core to boost voltage earlier in the GPU's boosting table. So it is likely this feature won't drastically improve what the GPU offset slider does by default.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Regardless, the addition of triple-channel voltage should significantly improve Blackwell's overclocking headroom on cards that support it.
The only way overclockers have been able to gain serious performance improvements through overclocking on Nvidia's latest GPUs is by using exotic cooling solutions that drop the GPU temperature to ambient or sub-ambient temperatures and using modified firmware that allows the GPU to pull significantly more power than it's supposed to from the factory.

Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards.
-
_D_D On my own RTX 5070 core voltage can be increased up to 1.15V via software but saw huge leakage with idle PState 8 and low 2D clocks showing 180W of power being drawn before that. Not sure if this is a Blackwell problem or just my card.Reply
Earlier cards such as GTX 1660S could increase voltage up to 1.3V and RTX 3070 to just under 1.25V however the 3070 would power off the system if 1.25V set. 1660S doesn't show increased voltage via normal software monitoring except for I2C VRM monitoring of Vout or using a volt meter. 3070 can be read with special API function. Not tried 40 series. 50 series seems can be read normally.
5070 Graph of core voltage and power draw.
https://www.techpowerup.com/forums/attachments/lk2-png.409207/Wonder if MSI are going to see similar problems or if not if they will increase power limits to cater for increased voltage.
Memory overclock is typically limited to +6000MT/s which can be reached at default memory voltage so wonder if they will increase that too as has been seen with ASUS XOC VBIOS. -
jp7189
It seems every card I've had from the last few gens has been perf capped by vrel, so for me at least, undervolting is the best way to get more performance. Do the new MSI cards raise the vrel limit to go along with the voltage increases?_D_D said:On my own RTX 5070 core voltage can be increased up to 1.15V via software but saw huge leakage with idle PState 8 and low 2D clocks showing 180W of power being drawn before that. Not sure if this is a Blackwell problem or just my card.
Earlier cards such as GTX 1660S could increase voltage up to 1.3V and RTX 3070 to just under 1.25V however the 3070 would power off the system if 1.25V set. 1660S doesn't show increased voltage via normal software monitoring except for I2C VRM monitoring of Vout or using a volt meter. 3070 can be read with special API function. Not tried 40 series. 50 series seems can be read normally.
5070 Graph of core voltage and power draw.
https://www.techpowerup.com/forums/attachments/lk2-png.409207/Wonder if MSI are going to see similar problems or if not if they will increase power limits to cater for increased voltage.
Memory overclock is typically limited to +6000MT/s which can be reached at default memory voltage so wonder if they will increase that too as has been seen with ASUS XOC VBIOS.
I haven't tuned a 5000 series yet. Is it different this gen? That said, I just got a rog astral under water and even at stock it can pull 720w through the 12v2x6 connector even though the board claims its pulling around 600, so that's stopped my enthusiasm for trying to push it further. -
_D_D Undervolting can be a great way to improve efficiency and performance.Reply
AFAIK reliability (VREL) is just a flag and not a limit but regardless it seems to be broken on Blackwell along with some other flags.
IINM voltage increase will be done by offset at the intelligent VRM controller so if you set 1.000V with an offset of 100mV you get 1.100V but nvidia SW/FW thinks it's still 1.000V.
50 Series can be a little different depending what your doing (like memory overclock) but generally much the same.