Nvidia explains that the RTX 4060 Ti uses 128-bit video memory: the L2 cache is increased by 16 times, which greatly improves the hit rate.

According to news on May 20, NVIDIA has now released the RTX 4060 Ti 8G graphics card, whose memory width is only 128 bits. In comparison, NVIDIA's old RTX 3060 Ti is 256bit wide, and the RTX 3060 is also 192bit wide.

NVIDIA responded to this in a blog post, saying that compared with the previous generation of GPUs with 128-bit video memory width, the storage subsystem of the new NVIDIA Ada Lovelace architecture increases the size of the L2 cache by 16 times, which is a significant increase. Improved cache hit rate.

As shown in the figure above, L2 cache bandwidth has increased significantly in Ada GPUs. This makes it possible to transfer more data between the processing cores and the L2 cache.

In addition, Nvidia engineers tested a special test version of the RTX 4060 Ti with 32 MB of L2 cache and one that only used 2 MB of L2.

Across a variety of games and synthetic benchmarks, 32 MB of L2 cache reduces memory bus traffic by more than 50% on average compared to the performance of 2 MB of L2 cache. This 50% reduction in traffic allows the GPU to use its memory bandwidth more efficiently, increasing efficiency by up to nearly 2x. So, in this case, isolating memory performance, the performance of an Ada GPU with 288 GB/sec peak memory bandwidth is similar to the performance of an Ampere GPU with 554 GB/sec peak memory bandwidth. Across a range of games and synthetic tests, the greatly improved cache hit rate boosted game frame rates by up to 34%.

Nvidia said that historically, memory bandwidth has been used as an important metric to determine the speed and performance level of new GPUs. However, memory bitwidth alone is not a sufficient indicator of storage subsystem performance. Instead, it's helpful to have a more complete understanding of storage subsystem design and its overall impact on gaming performance.

Due to the advantages of the Ada architecture, including new RT and Tensor cores, higher clock speeds, the new OFA engine and Ada's DLSS 3 capabilities, the GeForce RTX 4060 Ti is 256-bit more powerful than the previous generation GeForce RTX 3060 Ti and RTX 2060 SUPER The graphics card is faster while using less power.

 RTX 4060 Ti and RTX 4060 graphics card parameter information is as follows:

RTX 4060 Ti has 4352 CUDA cores, equipped with 8GB / 16GB 128bit GDDR6 video memory, TGP power consumption 160W / 165W, using PCIe 4.0*8 connection, priced from 3199 yuan, and will go on sale on May 24.

The RTX 4060 graphics card has 3072 CUDA cores, is equipped with 8GB GDDR6 128bit video memory, consumes 115W, uses PCIe 4.0*8 connection, priced from 2,399 yuan, and will be available in July.