Last week, we reported a discovery Nvidia uses its streamlined GA104 graphics processing unit (GPU) for GeForce RTX 3060 It should be a graphics card based on GA106 GPU. The “new” board has the same performance characteristics and power consumption as the “old” board. This week, two graphics card vendors — Gainward and Galax — listed suitable products, indicating that this GA104-based GeForce RTX 3060 motherboard will be available soon.
Only in China.Currently
At present, Gengsheng and Galax only market GeForce RTX 3060 graphics cards based on GA104-150 in China, so this version of the product may eventually become another exclusive product in the country.At the same time, before Nvidia GeForce RTX 1650 And in At least one GeForce RTX 2060 Model number, so it’s not surprising if Nvidia decides to use GA104-150 for GeForce RTX 3060 motherboards sold in other parts of the world.
Only when the chip is not enough to power the GeForce RTX 3060 Ti/3070/3070 Ti graphics card, it makes sense to use the reduced GA104 GPU instead of the GA106 for the GeForce RTX 3060, because it has too many flaws. GA104 is a fairly large 392 mm² chip, containing 17.4 billion transistors and up to 6144 CUDA cores, and the manufacturing cost is quite high. In contrast, GA106 is much cheaper. It has an area of 276 mm², integrates 13.25 billion transistors and up to 3584 CUDA cores.
Usually, chip developers like AMD and Nvidia will try to sell defective bulk chips through OEMs and/or specific suppliers in specific markets (but they can still work perfectly with some non-functional units disabled) because It is not easy to support a large number of SKUs all over the world.
However, if a company has a large number of chips that cannot work in the correct configuration due to defects, they must either invent a new model or reuse the chips for low-end SKUs available globally. It remains to be seen whether Nvidia has enough GA104 chips to even be used on GeForce RTX 3060 Ti motherboards (GPUs with 4864 CUDA cores), and the company will have to sell them outside of China.
Gainward has two (or four, depending on your calculation method) GeForce RTX 3060 12GB graphics cards, which can be based on GA106-300, GA106-302 or GA104-150: GeForce RTX OC / OCG Equipped with a large three-slot cooling system with three fans and GeForce RTX 3060 DU / DUG With dual-fan dual-slot cooler, as found @momomo_usAlthough Gengsheng did not explicitly point this out, it seems that the letter “G” in the part number indicates that GA104-150 GPU is used.
Galax is a close relative of Gainward and has four (or eight) GeForce RTX 3060 12GB add-on boards (AIB) based on GA104 or GA106: GeForce RTX 3060 Gamer MAX overclocking[FG], this GeForce RTX 3060 Golden Master MAX OC[FG], this GeForce RTX 3060 Xing Yao MAX OC Forever Edition [FG],as well as GeForce RTX 3060 MAX Xing Yao [FG], according to @momomo_usIn Galax’s case, the letter FG in the part number seems to indicate that the board is powered by GA104, but we cannot be 100% sure because the supplier did not clearly state this.
Gainward and Galax graphics cards use large printed circuit boards (PCBs), but we don’t know if they use the same PCB for the GeForce RTX 3060 12GB graphics cards based on GA106 and GA104. At the same time, if the PCB is the same, it means that Nvidia’s GA104 and GA106 are pin-to-pin compatible. The good news is that at least it seems that the GeForce RTX 3060 AIBs powered by GA106 and GA104 have different part numbers.
A blessing in disguise?
Although it is currently difficult to find GeForce RTX 3060 equipped with GA104-150 GPU, customers may be motivated to purchase these motherboards.
Since GA104 is much larger than GA106, but has many unused parts, it is easier to cool it to normal temperature, which means these parts can be very good overclocking compared to regular GeForce RTX 3060 cards with GA106 Device.
Nevertheless, overclockers must remember that overclocking capability also depends on PCB design and power supply. At the same time, the design cost of the GeForce RTX 3060 AIB is very low, so setting the GPU frequency record is not the goal that engineers intend to achieve with these products.