GeForce RTX 4070 Ti's AD104 GPU Strips Naked And Cleans Up For A Glamour Shot

hero ad104 image
"But Zak," you cry, "there is no GeForce RTX 4070 Ti!" Not yet, there's not, but all information indicates that that's the name NVIDIA will use when it re-launches the un-launched GeForce RTX 4080 12GB. We've seen both rumors and leaks under that name, but what we hadn't seen until now was the GPU that goes on those cards. Well, you're looking at it: AD104-400-A1, in the flesh; the mostly-enabled AD104 GPU powering the GeForce RTX 4070 Ti.

With little to give context to the size, the image isn't that interesting, but we actually know the size, too, thanks to @harukaze5719. Their tweet from early this morning seems to share a snippet from an NVIDIA document that lists the AD104 GPU's size as 294.5mm², and its transistor count as 35.8 billion.

harukaze5719 tweet

For perspective on those numbers, the GA102 GPU that powers every Ampere GPU from the GeForce RTX 3080 on up to the RTX 3090 Ti "only" has 28.3 billion transistors. Despite that, it's over double the size, at 628.4mm². That's a testament to the quality of TSMC's 4N process more than anything.

For further comparison, the "GCD" of AMD's Navi 31 is right at 300mm², nearly the same size—although of course that chip doesn't include memory controllers or last-level cache. It's also fabricated on a slightly older process, TSMC's 5N. That chip powers the Radeon RX 7900 XTX, which AMD has positioned as a direct competitor to the GeForce RTX 4080.

In case you haven't been keeping up, this processor, the AD104, is actually the third-tier chip in the Ada Lovelace family. The GeForce RTX 4080 is based on a larger processor called AD103. While we suspect that AD103 will find its way into a scant few ultra-high-end gaming laptops, the majority of premium gaming laptops are likely to get GPUs based on AD104.

plaguetale
Benchmarks from NVIDIA.

NVIDIA already revealed some in-house performance data for the GeForce RTX 4070 Ti, albeit back when it was still being called "RTX 4080 (12GB)". Judging from NVIDIA's numbers, the raw (non-DLSS) performance looks to be somewhere between the RTX 3080 and the RTX 3090. That's not too bad for an "x70 Ti" card, although our opinion of the product will still hinge heavily on the final pricing.

The RTX 4080 (12GB) was announced at $899, and that amount is clearly too high if we judge by tiering. $899 is a solid $200 upgrade from the GeForce RTX 3080. Even if we take NVIDIA CEO Jensen Huang's words at face value, $900 for a GeForce RTX 4070 Ti will be a hard ask. Even $800 will be a tough sell, despite that the performance strikes close to the $1199 GeForce RTX 3080 Ti. We suppose we'll find out when everyone else does.

According to the same person who shared the AD104 image above (MEGAsizeGPU, @Zed__Wang on Twitter), the GeForce RTX 4070 Ti will be available on January 5th. That implies a release at or before CES 2023, which starts on January 3rd. That's barely over a month away, so keep your eyes peeled.