Breaking the RTX 2080 SUPER !! NVIDIA GeForce RTX 3060 Ti FE-Hardware HKEPC graphics card

Debut of the new NVIDIA GeForce RTX 3060 Ti card

▲ GeForce RTX 3060 Ti officially debuted

NVIDIA officially released the second GPU product “Ampere” at the performance level the first, the main code is “GA104-200” “GeForce RTX 3060 Ti”, claiming to be able to kill the high-end GeForce RTX 2080 SUPER model Previous generation RTX 20, using first NVIDIA model The second generation RTX architecture, Ampere, is equipped with more CUDA Cores, faster second generation RT Core ray tracing cores, and faster third generation Tensor cores. It offers incredible performance and features for the price of $ 399.

▲ RTX 3060 Ti performance comparison officially provided by NVIDIA

According to NVIDIA, “GeForce RTX 3060 Ti” provides 16 FP32 TFLOPs, 32 RT TFLOPs, and 130 Tensor TFLOPs. Performance is roughly the same or faster than the GeForce RTX 2080 Ti, but the price is half the price, compared to the previous generation of the same price GeForce RTX 2060. Super is 1.4 times faster and the performance level is Full-HD / 2K. Playing smoothly is no longer a dream.

8 nm process, GA104-200 graphics core

Unlike the flagship GeForce RTX 3080 and RTX 3090, which use the “GA102” graphics core, the “GeForce RTX 3060 Ti” performance level uses the small-scale “GA104” graphics core. Compared to the previous generation “TU104”, the difference in performance mainly comes from The FP32 arithmetic unit is doubled, the second generation RT cores are updated and the third generation Tensor cores are updated, so that the traditional raster graphics operation improved by 1.7 times and ray tracing performance improved by almost 2 times.

“GeForce RTX 3060 Ti” uses the “GA104-200” graphics core, NVIDIA custom process at 8nm, SAMSUNG foundry, with 17.4 billion transistors, the die size is about 392mm², a full GA104 chip has 6 GPC units, 24 A TPC texture processing cluster and 48 SM streaming multi-processors increased to 6,144 CUDA Cores, 48 ​​RT Cores and 192 Tensor Cores.

▲ NVIDIA GA104-200-A1 graphics core

However, some units of “GeForce RTX 3060 Ti” have been shielded and reduced to 5 GPC units, only 19 TPC texture processing clusters, only 38 SM streaming multi-processors, 4,864 CUDA cores, 38 RT Cores and 152 Tensor Cores .

In terms of core clock, the default clock of the GeForce RTX 3060 Ti is a base clock of 1,410 MHz, a boost of 1,665 MHz, and the highest TDP is 200 W. Furthermore, “GeForce RTX 3060 Ti” still uses particles of GDDR6 memory, memory capacity of 8GB, the memory clock is 1,750MHz, but the transmission speed is 14Gbps, with 256-bit memory bandwidth interface, and the total bandwidth is 448GB / s.

NVIDIA GeForce RTX 30 family

GPU codename	GA104	GA104	GA102	GA102
GPU architecture	NVIDIA Ampere	NVIDIA Ampere	NVIDIA Ampere	NVIDIA Ampere
GPC	5	6	6	7
TPC	19	23	34	41
Sms	38	46	68	82
CUDA Core / SM	128	128	128	128
Core / GPU CUDA	4864	5888	8704	10496
Tensor Core / SM	4 (3rd generation)	4 (3rd generation)	4 (3rd generation)	4 (3rd generation)
Tensor Core / GPU	152 (3rd generation)	184 (3rd generation)	272 (3rd generation)	328 (3rd generation)
Core RT	38 (second generation)	46 (second generation)	68 (second generation)	82 (second generation)
GPU Boost Clock (MHz)	1665	1725	1710	1695
Size and type of frame buffer memory	8192 MB GDDR6	8192 MB GDDR6	10240MB GDDR6X	24576 MB GDDR6X
Memory interface	256 bit	256 bit	320 bits	384 bits
Memory Clock (Data Rate)	14 Gbps	14 Gbps	19 Gbps	19.5 Gbps
Memory band	448 GB / sec	448 GB / sec	760 GB / sec	936 GB / sec
ROP	80	96	96	112
L1 data cache / Shared memory	4868 KB	5888 KB	8704 KB	10496 KB
L2 cache size	4096 KB	4096 KB	5120 KB	6144 KB
TGP (Total Graphics Power)	200 Watts	220 Watts	320 Watts	350 Watts
Transistor count	17.4 billion	17.4 billion	28.3 billion	28.3 billion
Matrix size	392.5 mm2	392.5 mm2	628.4 mm2	628.4 mm2
Price	399 USD	499 USD	699 USD	US $ 1499