Galax RTX 5070 Ti HoF (Hall of Fame) review – To the top with a monster board?

1 - Introduction, overview and technical specifications 2 - Test system and equipment 3 - Teardown: PCB and components 4 - Teardown: Cooling system 5 - Teardown: Material analysis and TIM 6 - Benchmarks: gaming performance 7 - Power consumption, transients, PSU recommendation 8 - Clock rates and overclocking 9 - Temperatures and thermal imaging 10 - Fan curves and noise with audio samples 11 - Summary and conclusion

Teardown

The teardown of the Galax RTX 5070 Ti HoF is relatively straightforward, as the cooler and housing design is modular and contains no hidden or glued elements. After removing the back plate, which is secured with a few screws, the massive heat sink can be separated from the board relatively easily. The cooling design consists of a large base plate with a direct contact surface to the GPU, flanked by neatly positioned thermal pads for the memory modules and the voltage converter areas.

The backplate itself is functional and features an illuminated HoF logo; it contributes to passive cooling, especially in the area of the rear of the power supply. The actual circuit board is clearly structured, typical of the Hall of Fame series, with a solid VRM assembly and high-quality components. The slot bracket is also screwed on separately and can be removed without additional effort. It is striking that all components are clearly separated and easily accessible, which facilitates both maintenance and modifications.

Board and components

As with NVIDIA’s reference design, the Galax RTX 5070 Ti Hall of Fame uses three large voltage rails and several smaller auxiliary voltages. However, the PCB is an independent high-end design and is not identical to the PCBs of other models such as the MSI RTX 5070 Ti Vanguard SOC. The power supply is particularly generous here, with a total of 21 phases, which are divided into 14 phases for NVVDD, 4 phases for MSVDD and 3 phases for FBVDD. I will discuss the active components used and their manufacturers separately later.

The voltage converters for NVVDD, i.e. the core voltage of the GPU, are tried and tested solutions from the high-end segment. What is new, however, is that NVIDIA is again providing separate voltages for the GDDR7 memory and the frame buffer, similar to current CPU platforms from Intel and AMD. While GPU core voltage and memory voltage are already established variables, the dedicated voltage regulation for the frame buffer represents a targeted technical advancement.

The frame buffer functions as a special memory area in which the pixel information of the currently displayed image is stored. It stores data such as color depth, transparency and resolution and is constantly updated by the GPU to ensure smooth image output. It is physically connected directly to the graphics memory, which is operated under the MSVDD voltage. This voltage supplies the memory chips themselves and influences their clock stability and speed.

The FBVDD voltage, on the other hand, ensures precise and stable data transfer between the GPU and memory, especially at high memory frequencies. As the memory logic and frame buffer are closely interlinked, MSVDD and FBVDD are coordinated so that image data can be exchanged efficiently. The separate voltage regulation allows targeted adaptation to the specific electrical requirements of the individual functional areas. The Galax RTX 5070 Ti Hall of Fame therefore differs not only from the reference design of the RTX 5070 Ti, but also from the Founders Edition due to its significantly more complex and powerful power supply.

The BPD93136, which is probably still unknown to many readers, is a digital multiphase PWM controller from Shanghai Bright Power Semiconductor, or BPS for short. It is designed for dual-rail supplies in high-current applications such as graphics cards and servers and meets NVIDIA’s current OpenVReg specifications. The component scales up to 16 phases, allows splitting to two outputs and supports the NVIDIA-specific PWMVID interface for the core rail. I²C or PMBus according to revision 1.3 and AVSBus are available for configuration and telemetry, allowing voltages, currents and temperatures to be read out and controlled. The current measurement has a flexible design and is capable of DrMOS-IMON, DCR measurement and evaluation of the low-side RDSon; a non-volatile configuration is also integrated, which facilitates commissioning without an external memory. Depending on the configuration, the switchable frequency range extends from around 250 kilohertz to 2 megahertz, which allows the design to be optimized between high efficiency and low inductance. The controller is offered in a compact QFN housing with 56 pins, in practice often in a format of around 7 x 7 millimetres, and is specifically advertised for GPUs of the current NVIDIA generation, as it fully supports OVR16 and, according to the company, also addresses the OVR4-22 specification. Typical areas of application are graphics cards, servers and telecommunications hardware where a dual rail supply with a high number of phases and close-meshed telemetry is required.

The BPD93204 is a four-phase digital PWM controller from Shanghai Bright Power Semiconductor designed specifically for modern NVIDIA graphics cards and fully complies with the OpenVReg OVR4-22 standard. It is optimized for use as a core voltage regulator and offers a precise voltage supply with high efficiency and fast response times to load changes. Here, however, it works for the provision of MSVDD. The controller supports the PWMVID interface for direct communication with the GPU and also features PMBus 1.3 and AVSBus, allowing voltages, currents and temperatures to be monitored and adjusted in real time. Current measurement can be carried out flexibly via DrMOS-IMON, DCR measurement or low-side RDSon, which enables precise load distribution and reliable protective shutdown. Several protection mechanisms are integrated to safeguard operation, including overvoltage, undervoltage, overcurrent and overtemperature protection. The internal regulation is based on a fast constant frequency architecture, which is specially designed to minimize output voltage deviations during load jumps.

The BPD80690 from Shanghai Bright Power Semiconductor is a highly integrated DrMOS module developed for use in high-current applications such as high-end graphics cards and impresses in practice with its combination of high continuous load capability, thermal robustness and compact design. It is officially specified for currents of up to 70 amps (although it can unofficially handle up to 90 amps) per phase and can also work reliably at high switching frequencies of up to 1.5 megahertz. These properties predestine it for modern multi-phase supplies where both precise voltage regulation and high power density are required. The TLGA housing with a footprint of 5 x 6 millimetres has a very low thermal resistance and therefore promotes even heat dissipation across the circuit board. This not only allows safe operation at high currents, but also minimizes thermal hotspots, which can lead to premature ageing in other designs. The internal driver and MOSFET structure is designed for minimal switching losses and efficient timing, so that the BPD80690 operates in an optimum efficiency range even under full load.

Integrated protective functions such as overcurrent, overtemperature and short-circuit protection ensure that no critical states occur even under unfavorable operating conditions. The fast response of these protective mechanisms ensures that load peaks or faulty contacts do not lead to permanent damage. Especially in combination with powerful PWM controllers, this enables a power supply that remains stable even over long periods of time close to the maximum load. One example of its use is the Galax Hall of Fame Edition of the RTX 5080, in which the BPD80690 is used in the GPU core power stage and which uses the same PCB. Here, the high current carrying capacity is fully utilized to reliably supply the GPU both with standard and greatly increased power consumption. The thermal properties of the module have a direct effect on the overall stability of the system, as less cooling reserve is required in the power stage area and the ageing of the components is slowed down. Thanks to this combination of high electrical load capacity, optimized thermal management and protection logic, the BPD80690 is one of the components designed not only for short-term peak loads, but above all for years of continuous operation under demanding conditions. Its role in graphics cards such as the Galax HOF shows that it is designed for both maximum performance and long-term reliability.

Also on the board is the NCP45492, a monolithic high performance IC specifically designed to simultaneously monitor bus voltages and currents on up to four high voltage power supplies. Its key features include precise translation and scaling of shunt and bus voltages, enabling accurate measurement of supply currents. By supporting up to four separate power supplies with just one device, the chip offers high flexibility in voltage monitoring.

Each of the four channels can be individually adapted by selecting external resistors, allowing precise adjustment to specific requirements. The device is also characterized by a fast settling time, which minimizes delays in voltage monitoring. A real-time display of the validity of all bus voltages ensures that critical voltage deviations can be detected immediately. These features make the NCP45492 particularly suitable as a supervisor for the 12V lines (12V2X6 and PEG) of the power supply, as it ensures reliable control and protection of these central supply rails.

The chip with the designation “GM009-B” (labeled MCU on the diagram above) is a special microcontroller from GigaDevice, which is used in this version as an embedded controller or system controller in graphics cards. Such components play a central role in controlling and monitoring the card, work independently of the GPU itself and are precisely matched to the respective board design via the firmware. Typically, such a controller takes care of tasks such as initializing the voltage converters at system startup, controlling and monitoring the fans, evaluating temperature sensors, controlling the RGB lighting and managing telemetry data, which is later reported to the driver software or the BIOS. In addition, protective functions such as overtemperature shutdowns, emergency fan control or the shutdown of individual voltage ranges can be implemented.

The identifier “GM009-B” indicates firmware adapted for a specific board partner or a specific series, meaning that this microcontroller is not necessarily identical to standard MCU versions from the data sheet in terms of its function and enabled features. The flash and RAM capacity integrated in the chip is sufficient for storing the firmware, processing sensor and control data in real time and communicating with the other components of the board via common interfaces such as I²C, SPI or UART. Due to its robust design and close integration into the board management, this microcontroller is designed for continuous operation in a thermally and electrically demanding environment. This means that it must work reliably even at high ambient temperatures and in continuous load operation in order to ensure the overall stability of the graphics card.

Criticism of the power monitoring on the connector

The handling of the 12V2X6 connector is a particularly critical aspect of the power supply, especially with regard to the implemented safety mechanisms. In the reference design, NVIDIA has planned to record the entire current flowing through this connector as a single supply rail. This means that all lines of the connector are monitored together instead of considering each one separately. This decision is clearly due to NVIDIA and not the board partners, but it does have safety-related disadvantages.

As there is no individual current monitoring of the cables, an uneven load within the connector can go unnoticed. If, for example, one cable is subjected to greater stress than the others due to uneven contact quality or different contact resistances, local overheating can occur without the card’s protective circuits intervening. This reduces the effectiveness of the thermal protection and increases the risk of damage due to inadequate load distribution or increased contact resistances.

Although Galax has provided the custom board with a powerful power supply, it is bound by the specifications of the reference layout. The real weak point remains NVIDIA’s decision not to measure each individual 12V line separately. Such differentiated monitoring could have detected irregularities at an early stage and initiated countermeasures. The current implementation does not exploit this potential and therefore represents an avoidable restriction of operational safety.

Who or what is BPS?

If you compare the current DrMOS modules and multiphase PWM controllers from Shanghai Bright Power Semiconductor (BPS) with the solutions from Monolithic Power Systems (MPS), which have so far been favored on NVIDIA cards, several interesting technical and economic differences emerge that are quite relevant in the high-end graphics card and server sector. This is because Galax is the first graphics card manufacturer to use the products on its HoF models. Reason enough to have a little chat and do some research. In recent years, BPS has focused heavily on the OVR standard from NVIDIA and offers DrMOS devices such as the BPD80690 and PWM controllers such as the BPD93136, which are specifically tailored to high currents, high switching frequencies and extensive telemetry. Compared to many MPS solutions, which are technically mature but often more generic in design, BPS components are already optimized for specific load profiles of modern GPUs during the development phase. This means, among other things, that the internal gate drivers, dead-time control and temperature derating are precisely matched to the thermal cycles and transient response of the target platform.

Another point is the flexibility of the layout. BPS DrMOS modules often use more compact TLGA or PQFN packages with very low thermal resistance and readily usable solder pads, which not only improves heat dissipation but also shortens the signal paths for high-frequency operation. In conjunction with a dedicated BPS PWM controller, higher switching frequencies of up to 1.5 MHz can be realized without EMC problems or timing weaknesses. In practice, MPS modules often achieve lower maximum frequencies or require larger filter and decoupling structures for stable operation, which increases space and costs. BPS now also offers a range of telemetry functions that are often only available in the more expensive MPS product ranges. These include native support for IMON evaluation, multi-point temperature measurement within the DrMOS, as well as precise current measurement via DCR or low-side RDSon, in conjunction with full connection to I²C, PMBus and AVSBus. This complete integration is particularly important in the OVR16 environment in order to meet the response times and monitoring limits required by NVIDIA.

BPS solutions also have advantages in terms of cost structure, especially for high volumes. As BPS manufactures in China and cooperates closely with local board partners, the total cost per phase in the design can often be lower than that of comparable MPS components without having to sacrifice thermal or electrical performance. Although MPS scores highly in terms of long-term availability and documented reference designs, it tends to be priced in the upper segment. The main advantages of BPS DrMOS and PWM controllers over monolithic power systems in GPU and server designs are the close adaptation to OVR standards, higher thermal efficiency in more compact housings, higher permissible switching frequencies with lower EMC risks, more extensive telemetry at comparable or lower costs and close cooperation with GPU board manufacturers in the design phase. Here is another table:

Here is a high-resolution microscopy view of all the important components:

VHX_000000

VHX_000001

VHX_000002

VHX_000003

VHX_000004

VHX_000005

VHX_000006

VHX_000007

VHX_000008

VHX_000009

VHX_000010

VHX_000011

VHX_000012

VHX_000013

VHX_000014

VHX_000015