It's real fun to look at leaks of the RTX 4090 and its ludicrous
graphics-processing prowess, but the unfortunate reality is that most of us simply do not have $1500 or more to lay out on a graphics card. The majority of us are going to be chomping at the bit for the lower-tier GPUs, so leaks about those are arguably even more exciting. That's what we have today: a performance rumor from kopite7kimi about AD106, the second-smallest Ada GPU.
Tweeting this morning, the prolific GPU leaker posted that "AD106 is not very strong." He backs that up with a benchmark leak: a 3DMark
Time Spy Extreme score of "<7000." That's certainly not a world-beating score like the GeForce
RTX 4090 and
RTX 4080 leaks we've seen, but we do have to question his "not very strong" remark.
We don't usually run
Time Spy Extreme as a benchmark, but we probably don't have to tell an audience of hardware enthusiasts that one of the best features of 3DMark is its massive database of results. Perusing the results archive, and taking "under 7000" to mean "over 6000", we see the AD106 GPU in fine company: the GeForce RTX 3070, the GeForce RTX 2080 Ti, and the
Radeon RX 6750 XT. None of those are "weak" GPUs, capable of high-refresh 1440p gaming or even full 4K UHD on some titles.
Alright, sure—by comparison to the rest of the Ada Lovelace family, AD106 looks pretty puny. That's sort of the point, though. Where previous NVIDIA GPU generations have typically gone 102-104-106, we've got the AD103 GPU slotted in near the top-end this time around, and where the Ampere GA103 is fairly rare and only really used for high-end mobile GPUs, it sounds like AD103 might be the basis for the GeForce RTX 4080.
If true, that puts everything else down a tier, leaving the AD104 GPU for the GeForce RTX 4070 and possibly also the GeForce RTX 4060, in cut-down form, and if that's the case, the AD106 GPU could be relegated to strictly the GeForce RTX 4050 and lower-end parts, as well as mobile GPUs.
TechPowerUp! seems to think that's how things will shake out, and while we're not confident enough to make that prediction ourselves, it's not crazy.
With that in mind, a
Time Spy Extreme score of "under 7000" is pretty insane for an "x50"-tier GPU, especially when you compare to the GeForce RTX 3050's puny 2827 average score. Even if the AD106 GPU ends up in the RTX 4060 after all, it's still a nice step up from the RTX 3060's circa-4000 score. If we split the difference and call the AD106's "under 7000" showing as 6500, we're looking at a 61% increase over the RTX 3060—not actually all that far from the
79% gain of the RTX 4090 over the previous generation.
At the end of the tweet, kopite7kimi also notes that both of the two smallest next-gen NVIDIA GPUs will be limited to a PCIe x8 interface. If we believe earlier leaks, that'll be PCIe 4.0 x8, which has similar performance to PCIe 3.0 x16. The downside of this is that anyone installing one of these GPUs in a machine stuck with PCIe 3.0 will be limited to an x8 interface as well, but that probably won't matter much for most software.
Unfortunately, even though NVIDIA is purportedly poised to preview its next-gen graphics cards at its
Project Beyond event later this month, all rumors indicate that the reveal will likely be strictly for the RTX 4090 and possibly the RTX 4080. Lower-end cards will have to wait, and possibly a while—earlier rumors
seemed to say that we'll only see AD102 this year, with the more affordable cards coming early next year.