Jensen Huang, the CEO of Nvidia, recently finished his first GTC speech of 2022, during which he introduced Nvidia's next-generation Hopper architecture. It'll debut in the H100, a powerful data center GPU, but the announcement also contains indications for the RTX 4080 and Nvidia's next-generation consumer graphics cards.
Nvidia didn't mention the RTX 4080 during GTC, and it isn't expected to feature the Hopper architecture, according to speculations. Rumors stated that Nvidia might employ the Hopper architecture for its RTX 40-series graphics cards a few years ago, before the debut of the RTX 30-series graphics cards. In 2022, Nvidia is expected to introduce two generations: Hopper for data centers and Ada Lovelace for consumers. But that doesn't rule out the possibility of gaining information from the announcements.
Nvidia has divided its consumer and data center products for the second time. Nvidia introduced the Volta data center architecture between Pascal and Turing. It served as a temporary fix, allowing Nvidia to switch to a smaller production process for its data center devices in preparation for the next generation of consumer goods.
With the RTX 30-series, Nvidia brought both product lines together under the Ampere architecture. In other words, there isn't much precedence for what Nvidia is doing. It's the first time two Nvidia architectures have coexisted.
We learned that Hopper would be manufactured using TSMC's N4 technology and that Nvidia aims for efficiency. Nvidia is said to be employing TSMC's N5 process for the 4080, rather than the smaller and more efficient technology used by Hopper GPUs, which is intriguing.
Although N5 and N4 are both members of the same family, N4 is somewhat more efficient than N5. N5 appears to be more plausible for the consumer range, based on speculations regarding the substantial power needs for RTX 40-series graphics cards. This rumor supports leakers' claims that the RTX 4080 will have significant efficiency issues.
We may be witnessing a repetition of the Pascal/Volta/Turing dilemma. To set up the generation after the RTX 4080, Nvidia appears to be leading with Hopper, which features a more efficient architecture. We may have moved on to a simpler technique by then, but consumer cards will continue to lag behind data center cards.
The manufacturing technique is the most significant advancement, but Hopper also has a few other hints.
With the fourth version of NVLink, Nvidia focused on scalability. This interconnect is now only relevant in Nvidia's data center, but Huang promised it would be available to customers and partners shortly.
Nvidia argues that by making NVLink open, other businesses will be able to develop semi-custom processors that operate with Nvidia's products. Nvidia's forthcoming consumer graphics cards may be affected by this. According to rumors, AMD's RX 7000 graphics cards will use a multichip module (MCM) architecture, which combines numerous distinct processing clusters on a single chip.
Opening NVLink might pave the way for Nvidia to do something similar in the future. According to rumors, AMD's RX 7000 graphics cards will overtake Nvidia for the first time, and this might be owing to the MCM architecture. It's unclear whether the RTX 4080 will have an MCM design, but Hopper's release implies it won't.
The H100 CNX, a variant of the H100 GPU that includes an Nvidia ConnectX-7 SmartNIC, provides the final indication. This indication reduces latency and improves throughput to the GPU, removing server CPU constraints.
That isn't relevant for a desktop GPU, but the RTX 4080 may use a similar approach. By attaching an SSD directly to the GPU, Nvidia and IBM have improved memory bandwidth and throughput. We imagined this was a far-off technological development, but GTC says it might arrive sooner than later.
But, the RTX 4080 remains a significant question mark in the end. We have a few leaks concerning performance and efficiency and a few indications from Hopper, but we'll have to wait till the card is released to discover everything there is to know about it. It's expected to be released this autumn, though Nvidia hasn't confirmed that.
As we mentioned at the beginning, this sighting of Lovelace graphics cards might imply that we're moving closer to their release. This potential is fascinating because it aligns with another speculation that surfaced a few days ago, namely that Nvidia might be on pace to beat AMD with the introduction of its next-generation GPUs.
In other words, Team Green appears to have a leg up on the competition with RTX 4000 devices reaching stores ahead of RX 7000 (RDNA 3) cards, and the addition of GPU support inside this hardware utility lends credence to that theory.
This next-gen GPU is undoubtedly another shaky strand of supposition, and we shouldn't get too excited about it. Another option is that HWiNFO uses these Lovelace GPUs as placeholders based on the current information leaked via the Nvidia attack.
Both Nvidia and AMD's next-gen families are expected to provide significant results. Still, the newest word is that Team Red may have delivered a double-whammy in terms of substantial performance advances and significant efficiency benefits.