HWiNFO, the popular PC system information utility, has hinted at potential Nvidia GPU changes in a pre-release beta version. Support for a series of Nvidia Ada Lovelace GPUs appears to have been introduced to HWiNFO version 7.21 Build 4725.
There are additionally two Hopper architecture GPUs listed, as well as two more GPUs that we believe belong to the Blackwell family. It’s unclear whether the software can detect these GPUs or whether they’re just placeholders for projected future parts (i.e., they may be based on the Nvidia hack and subsequent leak of data).
Because Ada GPUs are meant for consumer graphics cards, most of our readers will be more interested in the impending Nvidia Ada Lovelace GPUs than the recently announced Nvidia Hopper or the far-off Blackwell. Ada Lovelace is the codename for the architecture underpinning the upcoming GeForce RTX 40-series graphics cards, which are rumoured to launch this fall, according to a computer scientist.
HWiNFO specifies the following Ada Lovelace GPU names in the release notes: AD102, AD103, AD104, AD106, and AD107. These five GPUs are expected to be the foundation of the RTX 40-series, ranging from a presumably (hopefully!) value-oriented GeForce RTX 4050 to a halo (and insanely pricey) RTX 4090. (Ti).
The names of these AD10x GPUs match those revealed by the Nvidia Lapsu$ hack. However, the AD10B silicon (yep, that’s a capital ‘b’, not a ‘8’ or a second variable) was mentioned in that hack, and we have no further information on it.
If the Ada Lovelace GPU series follows in the footsteps of Ampere in terms of GPUs and graphics cards, which we more or less predict, the AD102 should be the most powerful consumer GPU and will appear in the GeForce RTX 4090 and possibly RTX 4080, as well as possible Ti variations. The GPUs will then be divided into the RTX 4070, 4060, and 4050 classes.
Whatever the final model names are, these AD10x GPUs will almost certainly be utilised in laptops and will come in normal, Ti, Super, and whatever additional adjectives Nvidia chooses to add.
GH202 is unusual because the first digit denotes a second-generation Hopper. Maybe it’ll be an inference-tuned product, or maybe it’ll be for some of the Grace Hopper Superchips coming out next year. Given the H100’s 700W TDP, it’s also likely that Nvidia may release a second tier of Hopper GPUs that aren’t nearly as huge or power hungry, but that doesn’t explain why it’s not the GH101/GH102 instead of the GH202.
The mention of the GB100 and GB102 GPUs today is likely premature, as they are outside the scope of the present GTC. The Blackwell codename, named after David Blackwell, was revealed through the Lapsu$ leak.
However, given Nvidia’s previous GPU release schedule, we don’t expect to see Blackwell tape out or begin shipping until at least 2024. HWiNFO could have insider information on a different GPU project Nvidia is working on, but it hasn’t revealed anything yet.