Image: AMD
AMD has announced its next generation of graphics cards, the RX 7900 XTX and RX 7900 XT, which are the first cards powered by its new architecture, RDNA 3. That means they use a chiplet design, similar to the company’s Ryzen processors.
Both cards will launch on December 13th, with both AMD’s reference cards and board partner models set to ship on that date. The 7900 XTX will cost $999, and the XT is $100 cheaper at $899. That represents a significant price jump compared to the previous gen —the top-spec 6950 XT retailed for $849, while the base 6900 XT was $679. Nvidia’s RTX 4080, however, starts at $1,199, while the 4090 is an eye-watering $1,599.
There’s been a lot of buzz around AMD’s next generation of cards, with people waiting to see how it would answer Nvidia’s latest RTX 4000 series GPUs, especially considering that Intel’s newly released GPUs aren’t really a high-end contender.
Image: AMD
AMD’s performance claims for the RX 7900 XTX.
So, about those cards: the RX 7900 XTX is the company’s flagship, which it’s pitching as its card for 4K gaming that’s up to 1.7 times faster than its existing RX 6950 XT card at that resolution. It has 96 compute units clocked at 2.3Ghz and 24GB of GDDR6 memory with an up-to-384-bit bus. AMD appears to be focusing largely on performance per watt here. With a board power of just 350 watts, compared to the 450 watts on the RTX 4090, AMD might have an efficiency advantage here, though it does use more power than Nvidia’s stock RTX 4080.
The RX 7900 XT is a less powerful option — it has 84 compute units running at 2Ghz, 20GB of GDDR6 memory running at a slightly slower 320 bits, and a board power of 300W.
The company says it’s using GDDR6 memory instead of GDDR6X for both cards, mainly because it uses less power. You won’t need to use any special power connectors or adapters, as both cards will require two regular 8-pin connectors. Nvidia opted for a new 12VHPWR connector for the 4090 and has had problems with those melting, an issue that the company is still investigating.
The company’s promising a lot with RDNA 3: up to a 54 percent jump in performance-per-watt compared to RDNA 2, which used a more traditional GPU design. AMD says it’s capable of producing up to 61 TFLOPs, compared to RDNA 2’s 23 TFLOPs, though it’s worth noting that these numbers aren’t always comparable when you’re talking about different architectures. That means we probably won’t be able to use just the numbers to directly compare consoles like the PS5, Xbox Series, and Steam Deck, which use RDNA 2, to the company’s new PC GPUs.
At 4K, AMD is promising 62fps in Cyberpunk 2077 with its 7900 XTX card with ray tracing and AMD FidelityFX Super Resolution (FSR) enabled — the company says its new compute units have next-generation ray-tracing tech, which provide an up-to-50 percent bump in performance compared to its last-gen cards. Still, that’s less than what we’ve seen with the RTX 4090 at 4K with DLSS 2 or DLSS 3, so it looks like the 7900 XTX is positioned more to compete with Nvidia’s RTX 4080 instead.
Image: AMD
AMD performance promises for Cyberpunk 2077 and more.
When it comes to traditional rasterized performance, the story looks a bit better — though we’ll obviously have to wait and see how it performs when we get our hands on it ourselves. At “esports” (aka medium-to-low) settings and 1440p resolution, AMD says the 7900 XTX can max out the frame rate limits of Apex Legends, Overwatch 2, and Valorant, with 300, 600, and 833 fps, respectively. That’s when it was paired with a pretty beefy system Ryzen 9 7950X processor and 32GB of RAM on an AM5 motherboard.
At 4K (with seemingly unspecified settings), the company says the card can hit 295, 355, and 704 fps with those games.
Meanwhile, it says games like Call of Duty: Modern Warfare 2 can play around 306 fps as long as you’re using its FSR technology, which is similar to Nvidia’s DLSS tech in that it renders games at a lower resolution and then upscales them. Move up to 8K, and FSR does even more heavy lifting for 96 fps in Assassin’s Creed Valhalla.
Image: AMD
In games that support it, FSR does do a lot of heavy lifting.
AMD does say it’ll be improving FSR, with a new generation of the tech coming next year with “up to 2 times more fps” than FSR 2. It’ll use “fluid motion frames technology,” aka frame generation, similar to Nvidia’s DLSS 3 tech. We’ll have to wait and see how it really compares, though — Nvidia uses machine learning hardware on its cards to accelerate its tech, while AMD says the current iterations of FSR don’t because they’re meant to run on any graphics card, not just Radeon ones. The company wouldn’t say if the same was true for FSR 3, but given its promise of better AI performance on RDNA 3, it would be nice if it could give its next-gen scaling tech a bit of an advantage when it’s running on its latest cards.
Interestingly, AMD has opted for DisplayPort 2.1 support on both the RX 7900 XTX and RX 7900. That means higher refresh rate support at 4K and 8K compared to the old DisplayPort 1.4 port that Nvidia is using on its latest RTX 40-series cards. And just like Nvidia, AMD has simultaneous encode or decode for AV1, something that creators will be interested in for their streams.
Image: AMD
AMD seems very proud of RDNA 3, but it is slightly difficult to compare it to RDNA 2 just based on specs alone.
According to AMD, RDNA 3 is the first time a chiplet architecture (where a chip’s multiple parts are broken up and talk to each other via an ultrafast interface) has been used in a gaming GPU, and it says it decided to go with the design so it can use “the right process technology for the right job.” Basically, it means it can use cutting-edge 5nm tech for its graphics processing while relying on more “mature” 6nm chips for memory caching instead of having to use the fancy stuff for everything. AMD previously made waves with its chiplet-based Ryzen processors, and companies like Intel, TSMC, Samsung, ARM, and Qualcomm have all at least started looking into the concept.
The chips can talk to each other extremely fast, at around 5.3TB/s. For reference, Apple’s M1 Ultra inter-processor connection that connects two CPUs and GPUs runs at what was widely considered a blazing-fast 2.5TB/s.
When AMD first introduced Ryzen, it’s fair to say there were some growing pains as developers and hardware manufacturers figured out how to really take advantage of its design. During a Q&A session, AMD said it expects the cards to be good right out of the gate, though it does admit that they’ll likely see some performance improvements over the next year or so as the drivers and software improve.
Image: AMD
AMD Hypr-RX one-click optimization.
Outside of raw hardware speeds, AMD was keen to stress software improvements for its latest Radeon cards. AMD is introducing a new one-click feature called “Hypr-RX” in early 2023. It’s meant to help you automatically have the best settings for various AMD technology like FSR or anti-lag without having to play around with them yourself to see what offers the best frame rates and responsiveness.
AMD is also launching SmartAccess video with its Radeon 7000 series GPUs, which works paired with Ryzen 7000 series CPUs. It essentially spreads the encoding and decoding workloads across Ryzen CPUs and Radeon GPUs, so if you’re doing batch transcoding, it will greatly improve how long it takes to complete a video. AMD says this will be supported in apps like OBS, Adobe Premiere Pro, and others starting in December.