Why The GeForce RTX 4060 Ti Could Be A Killer GPU For Compact Gaming PCs

geforce rtx 3080 pcb
Little PCs like Intel's Raptor Canyon NUC 13 Extreme are absolutely awesome, but you aren't going to be fitting a GeForce RTX 4080 in one of those things. You might get away with a Radeon RX 7900 XTX, but it'll still be a snug fit—and that a pretty good-sized chassis, as small-form-factor machines go. If you want to build something even smaller, you'll need a much smaller GPU. Apparently, the GeForce RTX 4060 Ti could fit the bill.

Over the last few days, regular leakster @kopite7kimi has dropped new specifications for both the GeForce RTX 4060 Ti as well as the standard GeForce RTX 4070. As a refresher, the card formerly known as the GeForce RTX 4080 12GB, which has now been "unlaunched," is expected to reappear under the GeForce RTX 4070 Ti moniker. These two cards would be the next two steps down from that model.

rtx4060ti kopite7kimi tweet

Looking at the rumored specifications, we see the GeForce RTX 4070 being cut down from the same AD104 die used for the RTX 4070 Ti. If the rumor is accurate, it takes big cuts to its compute capability, and likely to its raster graphics ability as well. Those are on the order of 30%. However, the rumor seems to indicate that it will retain the same memory bus width and memory transfer rate, giving it the same memory bandwidth.

The more interesting part is the GeForce RTX 4060 Ti, at least if kopite7kimi's predictions are accurate. It will supposedly be based on a mostly-intact AD106 die. He puts it down for 4,352 shaders, which if we assume a 2.6 GHz boost clock, gives it a peak single-precision compute rate of 23 TFLOPs. Not too shabby for an x60 part. However, AD106 is known to only have a 128-bit memory interface, which gives NVIDIA the option of 8 or 16GB.

ada vs ampere chart dec2022

Worse, the RTX 4060 Ti will apparently come equipped with lower-clocked GDDR6 memory instead of the GDDR6X that the GeForce RTX 4070 will apparently use. The combination of narrow bus and mediocre transfer rate gives the hypothetical GeForce RTX 4060 Ti just 288 GB/sec of memory bandwidth, less than RTX 3060, RTX 2060, GTX 1660 Super, and even the GeForce GTX 1080. The drastically-increased L2 cache of Ada seems to help with that, though.

The leaker notes that the reference board is "very short," and also that it still uses the CEM5 connector. That's another name for the 12+4-pin 12VHPWR connector. It's interesting to see it here; kopite7kimi seems to expect that the RTX 4060 Ti will launch with a total board power of 220 watts. While that is by no means a low-power part, relative to the extremely-thirsty models at the top of the product stack, it's fairly svelte. It will likely draw less power than that while gaming, too, just as we've seen with the RTX 4090 and its smaller sibling.