A pre-release beta version of the popular PC system information tool HWiNFO has telegraphed some eyebrow raising Nvidia GPU updates. HWiNFO version 7.21 Build 4725 appears to have added support for a string of Nvidia Ada Lovelace GPUs. There are also a pair of the newly announced Hopper architecture GPUs listed, and two more GPUs which we think belong to the Blackwell family. Whether the software can actually detect these GPUs or if they're merely placeholders for presumed future parts is unknown (i.e., they may be based on the Nvidia hack and subsequent leak of data).
Most of our readers will be particularly interested in the upcoming Nvidia Ada Lovelace GPUs, rather than the just revealed Nvidia Hopper, or the far future Blackwell, as Ada GPUs are designed for consumer graphics cards. Specifically, Ada Lovelace is the computer scientist-based codename for the architecture behind the expected GeForce RTX 40-series graphics cards, which are rumored to arrive this fall.
(Image credit: Future)
In the release notes, HWiNFO lists the following Ada Lovelace GPU names: AD102, AD103, AD104, AD106, and AD107. These five GPUs are likely to form the foundation of the RTX 40-series, spanning from a presumably (hopefully!) value oriented GeForce RTX 4050 up to a halo (and hella-expensive) RTX 4090 (Ti). These AD10x GPU names tally with those that came to light from the Nvidia Lapsu$ hack. However, that hack included mention of the AD10B silicon (yes that is a capital 'b', and not an '8' or a second variable), which we have no further clues about.
If the Ada Lovelace GPU series follows in the footsteps of Ampere with regard to GPUs and their associated graphics cards, and we more or less expect it to, the AD102 should be the biggest consumer GPU and will feature in the GeForce RTX 4090 and likely RTX 4080, along with potential Ti variants. From there, the GPUs will trickle down into RTX 4070, 4060, and 4050 classes. Whatever the final model names, these AD10x GPUs will probably be used across the gamut of standard, Ti, Super, and whatever other adjectives Nvidia may choose to add, and will also feature in laptops.
We don't have any truly reliable information with regard to GeForce RTX 40-series performance, but we expect it will be impressive. The most significant Lovelace leak we've seen recently points to the AD102 packing in up to 75% more cores than the GA102 (Ampere, RTX 3090). Thus we could be looking at 18,432 CUDA cores on the flagship RTX 40 graphics card. For perspective, the GA102 offers 10,496 CUDA cores. Ultimately we have to take this CUDA core count information with a dose of salt, and moving to a new architectures doesn't provide a clear picture of the generational performance uplifts we will be getting.
Nvidia Hopper and Blackwell
We are in the midst of Nvidia GTC today with lots of serious graphics technology aimed at data centers, AI, robotics, and scientific research. We have a lot more information direct from Nvidia about Hopper today, namely the GH100 GPU, along with some of its specs and capabilities. However, the GH202 GPU in the HWiNFO release notes is something of an unknown. There was also an Arm-based Grace CPU Superchip with 144 cores launched today, though we don't have the codename for it.
GH202 is odd, because the first digit indicates some sort of second generation of Hopper. Perhaps this will be an inference-tuned offering, or it may be for some of the Grace Hopper Superchips that are slated for next year. It's also possible Nvidia will have a second tier of Hopper GPUs that aren't quite as large or power hungry, given the 700W TDP of the H100, but again, that doesn't explain why it's not GH101/GH102 as opposed the GH202.
Beyond the scope of the current GTC, today's mention of the GB100 and GB102 GPUs is likely premature. The Lapsu$ leak did reveal the Blackwell codename, in honor of David Blackwell. But given Nvidia's more recent GPU release cadence, we wouldn't expect to see Blackwell tape out or start shipping until some time in 2024 at earliest. Maybe HWiNFO has insider information on some other GPU project Nvidia is working on, but if so it hasn't provided any details yet.
Unfortunately, while HWiNFO may now be able to identify some of these upcoming GPUs, it doesn't provide any details on the actual hardware configurations. We'll have to wait to find out more in the coming months.
Mark Tyson is a news editor at Tom's Hardware. He enjoys covering the full breadth of PC tech; from business and semiconductor design to products approaching the edge of reason.
