According to a Twitter post by @Zed__Wang, Nvidia's (vanilla) RTX 4060 8GB will use the company's entry-level Ada Lovelace GPU die known as AD107, rather than the larger AD106 die from the RTX 4060 Ti. This won't affect the GPU's official specifications, but the fact that the RTX 4060 can be used on the AD107 die shows focused Nvidia is on increasing power efficiency this generation on mid-range and entry-level GPUs.
Nvidia's decision to use AD107 inside the RTX 4060 isn't a leak or a rumor, in-fact Nvidia told us directly that it will officially use the AD107 die in its new mid-range GPU. As a result, this will be the first RTX xx60 class product operating on an entry-level die. Previous generations of mid-range RTX GPUs like the 3060 and 2060 worked on the GA106 or TU106 dies, one tier higher than their entry-level equivalents ending with the number 7.
We believe the GPU will use a fully enabled AD107 die. As a result, there won't be another GPU model surpassing the RTX 4060 with better specifications on the same die, including core count, memory bus configuration, ROPs, and cache. There could be an "RTX 4060 Super" with higher TDP and clocks, but there will not be another GPU with higher physical specs on this die.
|Graphics Card||RTX 4060||RTX 3060|
|Process Technology||TSMC 4N||Samsung 8N|
|Die size (mm^2)||158.7||276|
|GPU Cores (Shaders)||3072||3584|
|Boost Clock (MHz)||2460?||1777|
|VRAM Speed (Gbps)||17||15|
|VRAM Bus Width||128||192|
|TFLOPS FP32 (Boost)||15||12.7|
|TFLOPS FP16 (FP8)||121 (242)||102 (sparsity)|
|Bandwidth (GBps)||272 (453 effective)||360|
|Launch Date||Jul 2023||Feb 2021|
For reference, the RTX 4060's specs include 3072 CUDA cores, 24 SMs, 96 Tensor Cores, 24 RT Cores, 24MB of L2 cache, 48 ROPs, 96 TMUs, and a 128-bit bus. In addition, boost clocks are rated at 2460 MHz, and the GPU will feature a power target of 115W.
AD107 gives Nvidia several advantages with the RTX 4060; one is that the GPU will be cheaper to produce since it will be operating on the smallest and easiest-to-produce GPU die in the Ada Lovelace family. Another is that Nvidia can offload production to another GPU die in the future if it needs to improve yields. For example, Nvidia might opt to leverage both AD107 and AD106 for the vanilla RTX 4060 in the future to enhance yields and reduce waste on potentially bad AD106 dies that cannot be used with all of the cores turned on.
For enthusiasts and gamers, it is unfortunate to see the RTX 4060 being regulated to the AD107 die, with the performance potential it could have had with Nvidia's bigger AD106 die. But at least we are getting an incredibly efficient GPU that will be easy to power and cool.
AD107 "Ada" GPU Die Size = 146mm2 (TBD @5nm)
GA107 "Ampere" Die Size = 200mm2 (8.7 Billion Transistors @8nm)
TU117 "Turing" Die Size = 200mm2 (4.7 Billion Transistors @12nm)
I assume reason they made a "ti" version was to not relive the 1060 6gb vs 3gb issue where they were same tier yet vastly different they shouldnt of shared same name.
the ti looks fine but the non ti is questionable if its gonna be any better than 3060 outside dlss 3 enabled titles.
Ada is more powerful and power efficient than Ampere. Isn't that what people were crying for in the past?
Sometimes one may rise more than the other. That's just physics.
Nvidia next gen will be power efficient since they achieved that now. Therefore, they can focus on squeezing more performance out next time. But they will have a different design most likely to take advantage of more cache hopefully.
Now product segmentation is more deliberate and not driven by production process variability. (?)