Based on the GA100 amp core, NVIDIA has released products such as the A100 Accelerated Computing Card. Unsurprisingly, in the current quarter, GeForce gaming cards for consumer-grade users will also be available. Recently frequently broke the RTX 30 series graphics card message Kopite 7kimi also gave the message “listen to the public version of ampere core can touch 2.1 to 2.2GHz”, but may be afraid of the follow-up “face-kicking”, he said cautiously, “not quite sure.”
We know that the highest public frequency of the previous generation of RTX 20 series graphics cards was the RTX 2080 Super, which accelerates to 1815MHz. Although non-public super 2GHz is not difficult, but the RTX 30 series factory so fast, ampere architecture is called a leap, the ultimate potential is certainly even more amazing.
Previously, the same source has revealed that this generation of RTX 30 series graphics card GPU for cost reasons, based on Samsung 8nm, that is, 10nm modified version, so the frequency is not expected to be very exaggerated, two news superimposed to see the words, it seems more confusing.
Rumor has it that the RTX 3070 has three models, all with ga104 cores, while RTX 3060 is the GA106-300 core. The GA104-400 Core (SKU0) is currently designed with two PCB boards, one for PG141, 8GB GDDR6X, one for PG142, 8GB GDDR6 memory, all built-in 3072 CUDA, which is consistent with THE RTX 2080 Super.
There is also the RTX 3070 universal version of the GA104-300, with 2944 CUDAs built in, consistent with THE RTX 2080, matching the 8GB GDDR6 memory. If the pin bandwidth can touch 16Gbps, the total bandwidth under 256bit is 512GB/s, a significant increase.
The RTX 3060 Ti/Super has the potential to integrate 2304 CUDAs and RTX 3060 to integrate 2176 CUDAs.
Therefore, the performance of N card is definitely not to let us worry, the real people are entangled in the end is the price will not rise, the magnitude and geometry?