GeForce RTX 4060 leak: lowest-power mainstream GPU since Pascal

The mainstream model of the Ada Lovelace generation

We recently heard about the GeForce RTX 4070, but a card that should be significantly lower in price is also starting to take shape: the GeForce RTX 4060. Kopite7kimi delivered information on the chip it will be based on and the general parameters. It looks like the new “sixty” model could be quite a bi more “lightweight” than previous cards, dropping power draw to significantly lower level not seen since the GeForce GTX 1060 from 2016.

According to Kopite7kimi, the news coming with the GeForce RTX 4060 is that the traditionally mainstream “sixty” model will be produced with a smaller (and potentially lower-power) chip. While previous cards have been based on GM206, GP106, TU106 and GA106 chips, the new GeForce RTX 4060 is set to be based on the smallest known chip of the Ada Lovelace generation so far, the AD107, and use a PCB designated PG190. The AD106 chip will likely only be used in the desktop realm in the GeForce RTX 4060 Ti.

Both the AD106 and AD107 are said to have, according to specifications obtained from hacked source code, the same 128-bit wide memory bus (while the AD102 is 384-bit, AD103 256-bit and AD104 192-bit). However, they will differ in the number of computing units and thus raw performance. AD106 contains 36 SM blocks or up to 4608 shaders / Cuda cores / FP32 units. The AD107 will be much smaller with only 24 SMs, giving a maximum of 3072 shaders.

The GeForce RTX 4060 is said to have the full configuration of this GPU (branded AD107-400) and its 3072 shaders, or 24 SMs, 24 RT cores and 96 tensor cores. We don’t know the clock speeds yet, but it really should be a very low-power graphics by today’s standards, the TDP is now said to be planned at just 115W. So roughly between the Radeon RX 6600 and the RX 6500 XT, but thanks to the 4nm manufacturing node and new architecture, this card should obviously be more powerful.

Memory will have a capacity of 8 GB and Nvidia is said to be planning to use GDDR6, but at 18.0 GHz effectively, which would mean a bandwidth of 288 GB/s. This means that the memory is expected to have the same capacity and bandwidth as the more powerful GeForce RTX 4060 Ti (which, however, will have higher raw performance with 4352 shaders and 160W TDP). Kopite7kimi also mentions the L2 cache capacity, which we did not know before. It will be 24 MB in size and this capacity will function similarly to the Infinity Cache on Radeon GPUs, allowing the graphics card to achieve slightly better performance than what would be otherwise possible without it given the memory bandwidth available.

The price is not yet known, not even approximately. According to the latest rumors, the more expensive GeForce RTX 4070 should be released in April, so we can’t expect the release of the cheaper GeForce RTX 4060 SKU before that date. It’s that some time will pass between the release dates of the two models (not to mention that the RTX 4060 will likely also have to wait for the launch of RTX 4060 Ti as well). Thus, the GeForce RTX 4060 probably won’t go on sale until sometime in the spring, if not early summer, perhaps before Computex.

That is, if Nvidia’s usual policy is to be maintained. But of course, the company may decide to change their usual practice and release the card earlier or even out of the usual order. Even so the launch won’t come in the immediate future though, as we would have seen more leaks by now in that case and we would also have already learned more detailed specs, including clock speeds.

Sources: Kopite7kimi, VideoCardz

English translation and edit by Jozef Dudáš


  •  
  •  
  •  
Flattr this!

Leave a Reply

Your email address will not be published. Required fields are marked *