NVIDIA GeForce RTX 4060 Spec Leak Suggests Fully Enabled AD107 With Meager Memory
Well, awkwardness continues, apparently, because if kopite7kimi's latest leak is accurate, the GeForce RTX 4060 will have just 8GB of video RAM onboard. Other relevant specifications include the memory type and clock—GDDR6 at 18 Gbps—as well as the news that the RTX 4060 will be using a tiny AD107 die with 3072 shaders and 24 MB of L2 cache. The leaker claims that NVIDIA is targeting a 115W power budget with this card.
The specifications are slightly suspect because they are nearly identical to those of the GeForce RTX 4060 mobile GPU, right down to the shader count and power budget. The only differences are that the mobile part, which releases next week, is running lower-clocked 16-Gbps memory, and that the leaked specifications for the desktop RTX 4060 only have 24MB of L2 cache.
That would normally imply a 96-bit memory bus, which makes little sense considering that the mobile part has a 128-bit bus with the expected 32 MB of L2 cache. So saying, something doesn't exactly add up here by our math; either the desktop card has a hilariously-narrow 96-bit memory bus—note that kopite7kimi doesn't elaborate on that spec—or the shader count isn't quite correct.
To clarify, on most of the rest of the Ada lineup, the L2 cache capacity is linked to the memory bus width: 8 megabytes of L2 per 32-bit GDDR6(X) memory channel. The only card where this isn't the case is the GeForce RTX 4090, and that's because it is cut-down from the full-fat AD102 that is rumored to wind up in a future "Titan Ada" or "RTX 4090 Ti" card. With that in mind, it's possible that NVIDIA could have simply disabled some of the L2 cache, with other components unaffected.
The only other notable part of the leak is the news that the GeForce RTX 4060 will use the PG190 board design. That continues the trend of Ada being pin-compatible with Ampere, as PG190 has been in use for some time now on GeForce RTX 3050, RTX 3060, and RTX A-series GPUs. If it ain't broke, don't fix it, we suppose.
Unfortunately, kopite7kimi doesn't give any clues as to when we might see the littlest Ada GPU on the desktop, but there's no real reason NVIDIA couldn't launch it as soon as next week, with the mobile parts. That seems unlikely to us, though. We'd expect to see an announcement in a month or two, followed by a release a few weeks after that. We'll be reporting it here, of course, so stay tuned for that information when it hits.
