AMD's Navi 48 graphics processor is based on the RDNA 4 architecture that the company briefly demonstrated yesterday and is much smaller than its RDNA 3 generation predecessor. The small die size may be evidence that the company does not intend to position its RDNA 4 GPU to compete with Nvidia in the high-end market.

According to calculations by blogger David Huang, the die size of the AMD Navi 48 GPU produced using TSMC's 4nm-class process technology is about 390 square millimeters. The GPU is slightly larger than Nvidia's AD103, which is manufactured using similar process technology.

The Navi 48 GPU is monolithic and does not come with any memory cache die, unlike its predecessors Navi 31 and Navi 32. This makes it impossible to compare its die size with that of its predecessors. However, if we combine the size of the Navi 3x graphics compute die (made on the N5) and the memory cache die (made on the N6), we can estimate that the Navi 32 uses about 346 square millimeters of silicon, while the Navi 31 uses 529 square millimeters of silicon. These measurements roughly correspond to the positioning of the GPU in the performance mainstream and high-end market segments, respectively. That being said, it's safe to say that AMD's intention is not to create an expensive GPU with its Navi 48.

TSMC's 4nm-class manufacturing technology, such as N4P, belongs to the same process development kit as the foundry's 5nm-class manufacturing process. These technologies offer about 1.06x higher transistor density, up to 11% better performance (at the same power and transistor count), or up to 22% lower power consumption (at the same speed and transistor count). While N4P has clear advantages over N5, don't expect significant performance gains simply from adopting a new node. Therefore, any performance improvements in RDNA 4 or the Radeon RX 9070 XT over existing products are likely the result of architectural enhancements rather than adopting a new node.

In theory, AMD could go with N4C and achieve higher transistor counts and/or reduce costs by up to 8.5% in an area of 390 square millimeters. However, this would come at the expense of performance, so it is unlikely that GPUs will adopt this node.

When AMD first showed off the Navi 48 GPU yesterday, it was shown as the company's Hawk Point Refresh processor. However, it does resemble the Navi 48 GPU in AMD's press materials, not the Hawk Point CPU. We asked AMD to clarify, but the company declined to comment. So, logically this is actually the debut of RDNA 4 compared to the company's other press materials.

AMD explains why RDNA 4 wasn't announced at CES

Perhaps the biggest surprise from AMD's January 6 CES keynote was the lack of mention of the Radeon RX 9070 series of desktop graphics cards and the RDNA 4 graphics architecture. This was largely because the CES teaser slides AMD provided to the press included material about Radeon, but were missing from the actual onstage presentation. Immediately after the event, AMD leadership, including David McAfee and Frank Azor, agreed to sit down with a small group of tech journalists for a roundtable discussion, and we got a chance.

In this article, we'll try to share what we can. The presentation began by addressing the elephant in the room, namely why AMD omitted Radeon-related announcements from Jack Huynh's keynote, and the company confirmed that providing the press with information teasing Radeon was intentional and that they knew it would not be included on the main stage.

AMD said the 45-minute keynote did not allow enough time to officially unveil the Radeon RX 9070 series and RDNA 4. The company noted that other announcements, such as the Ryzen Z2 series of gaming handheld chips, were omitted from the keynote. The company said Radeon-related announcements would require more time in the keynote to detail product naming changes (more on that later), the RX 9070 series' positioning in the market, advancements made with RDNA 4, and new technologies like FSR 4.

AMD said the Radeon RX 9070 series could arrive in the first quarter of 2025, and the company will host at least one online media event to unveil them. The hour-long presentation will give the team a better chance to dive into the details of RDNA 4 and the RX 9000 series. The company wouldn't confirm a specific date for when it plans to release these graphics cards, but we know from a brief preview up front that the launch is in the first quarter of 2025, and that all of AMD's add-in board partners have their custom-designed graphics cards ready.

We then go down the rabbit hole of product naming, and why AMD chose to abandon the gaming GPU model naming scheme it established over three generations of Navi, starting with the RX 5000 series, followed by the RX 6000 and RX 7000. The RX 5000 series has been around for over 5 years now, so the market has learned quite a bit about this naming scheme. AMD first tried to explain why it skipped the RX 8000 series. The company said it wanted to align the model numbering with the Ryzen 9000 series of desktop processors.

In the second half of AMD's answer, the company tried to explain the change from a name like "RX 9700 XT" to "RX 9070 XT." The rumors are true, and AMD wants to make it easier for gamers to identify the competitive positioning of the SKUs by giving them a naming scheme similar to the "competitor" i.e. NVIDIA. This way, gamers will compare the RX 9070 XT to NVIDIA's GeForce RTX 5070 or RTX 5070 Ti; and the future RX 9060 XT to the RTX 5060 Ti or RTX 5060.

Afterward, a reporter asked what everyone was thinking at the time - if this was done to educate consumers, would this naming scheme stick? The Radeon RX "Navi" series is now the 9000 series, and the next generation will probably either be the "RX 10070" or bring another change. In response, AMD agreed that it could not commit to a naming scheme, and that it would constantly evaluate the market and consumer behavior and implement whatever worked.

The next question from a reporter was what AMD had to say about the decision tree that led to the development of RDNA 4. AMD responded that it wanted to pursue sales, mainly by identifying key market segments that drive sales growth, and then trying to win in these market segments by providing products with higher price-performance and efficiency. Most of the market is chasing the 70 series and 60 series (for example: RTX 4070 and RTX 4060), so AMD can save a lot of product development costs by focusing only on the potential market segments. The company does not have to prove its ability to create enthusiast-level GPUs because the market does not think so. The switch to the 70 series and 60 series naming scheme is more about easier exposure to market sales.

AMD was also asked about the reaction to Intel Arc B580 "Battlemage", specifically how Intel successfully won praise from the media and gamers at a price of $250. AMD's response sounded like they were very happy with the B580. "I think that proves that if you bring features and performance to a very valuable price point, it really resonates with consumers."They seem to agree that AMD's strategy will be very similar to Intel's – impress the largest market segment with high-performance/value products.

AMD said it received a lot of praise and criticism for the RDNA 3 generation and tried to replicate the things that got it the praise. The RX 7000 series only had a few home runs, such as the RX 7800 XT and RX 7900 GRE, and the company’s efforts will be to replicate the success of those products.

Attention then turned to FSR 4, and a footnote in AMD’s brief presentation slides said that FSR 4 is only available for the gaming-capable RX 9070 series that has integrated FSR 3.1. AMD says FSR 4 uses machine learning-based algorithms to reconstruct details in its super-resolution components, and RDNA 4 significantly improves MLops over previous generations of RDNA, so initially FSR 4 will be available on the RX 9000 series, but the company will evaluate specific SKUs from previous generations that it believes can provide the AI acceleration performance required for the FSR 4 SR algorithm without incurring excessive performance or latency costs. FSR 4 will not only see significant improvements in frame rates at a given preset, but will also significantly improve image quality.

We then asked AMD about its priorities for developing RDNA 4 and why they ranked certain things high or others low. Their first priority was to focus on improving performance in the areas that gamers care about most. You should see huge ray tracing improvements, MLops (AI acceleration) improvements, like FSR 4 and ML Super Resolution in this generation.

The second priority is that every component on the GPU is designed for efficiency, not just in terms of energy efficiency, but in terms of enabling AMD to price it competitively - the right manufacturing process, the right die size, the right memory type. In other words, AMD is preparing for a very intense price war with NVIDIA and potentially Intel, and it doesn't want quirky hardware design decisions to put obstacles in the way for the marketing team.

The third and smaller priority is getting to market. If you remember, AMD was significantly slower to market than NVIDIA with the Radeon RX 7000 series. NVIDIA had almost finished launching the RTX 40 series when the RX 7900 series was launched. AMD is looking to avoid that this time around, and you should see a fairly active lineup of product launches from AMD in the RX 9000 series.

Finally, AMD refuted the performance claims circulating on the web about the RX 9000 series GPUs. AMD quipped: "No one has the final drivers, not even the motherboard manufacturers, so don't trust the performance claims on the web."

Source: Content compiled from tomshardware

View more at EASELINK