Alleged NVIDIA GeForce RTX 3050 8GB Die Swap Brings Power Consumption Way Down
The GeForce RTX 3050 8GB, as we (and everyone else) reviewed it, is based on the GA106-150 GPU. That's a cut-down version of the very same GA106 GPU used in the RTX 3060 and the professional-grade RTX A2000. If that's the case, then why does it come with a PCIe 4.0 x8 interface instead of the x16 interface that those two cards have?
The answer, according to Igor's Lab, is that NVIDIA planned all along to bring out an RTX 3050 based not on GA106-150, but instead on the GA107 GPU. GA107 isn't a new chip; it's already been used in the mobile GeForce RTX 3050 and 3050 Ti, as well as in the MX570, the A2 accelerator, and confusingly, the GeForce RTX 2050.
As you've surely assumed by now, the GA107 GPU only comes with eight PCIe lanes. This makes the die significantly smaller, and probably has an extremely marginal impact on its performance (compared to a full x16 bus), at least when used on a platform with PCIe 4.0 support.
Assuming Igor's Lab is correct about the details, the difference in the two GPUs performance-wise should be nil, as they have the same configuration and allocation of resources internally thanks to the cuts made in the -150 variant of the GA106. However, it's reasonable to assume that the physically-smaller GA107 will draw less power.
Even if you're not into overclocking, the smaller die should still mean less power consumed, and less heat dumped into your case. Unfortunately, it's possible that it will be difficult to tell from the outside which cards come with the salvaged GA106 GPU and which cards use the smaller GA107. We'll have to see if vendors make separate SKUs for the two GPUs.
