AMD has finally pulled the wraps off its next-generation RDNA3 GPUs, and they’re aimed at 4K gaming at high frame rates. At a live-streamed eventyesterday, the company unveiled its flagship GPU as well as a slightly cut-down version of the same chip. Both GPUs use its biggest Navi31 die, code-named Plum Bonito. This is the company’s first chiplet design for a GPU, and as rumored, it features a large graphics chiplet die (GCD) flanked by six smaller memory chiplet dies (MCD). The main die houses the actual GPU, with each smaller chip hosting memory controllers and Infinity Cache. The GCD was built using TSMC’s 5nm process, with the MCDs using a more mature 6nm node. Both GPUs will go on sale on Dec. 13 at $999 for the XTX and $899 for the XT.
The flagship Radeon RX 7900 XTX will sport 24GB of GDDR6 20Gb/s memory. The XTX will offer 6,144 streaming processors, which are “dual issue” now and line up with rumors of there being 12,288. AMD’s slides stated it has 96 compute units, compared with 80 for the 6900 XT. As for clock speeds, the XTX will offer a 1.9GHz base clock, a 2.3GHz game clock, and a 2.5GHz boost clock. The transistor count has gone from 26.8 billion for the previous generation to 58 billion this time around.
Each of the six MCDs house a 64-bit memory controller for a 384-bit memory bus. Additionally, each chip has 16MB of Infinity Cache, for a total of 96MB. This is less than the 128MB found in the previous generation. AMD is also reportedly working on a future card with 3D-stacked memory chips that could allow for a doubling of the available cache. The company says it has increased memory bandwidth 2.7X over its previous architecture.
AMD says the XTX is designed to handle 4K gaming, and can even do 8K on some titles. For its 4K benchmarks, it says it can deliver 353fps in F122 with FSR, as well as 295fps in Apex Legends. At 8K, it can run the newest Assassin’s Creed Valhalla: Dawn of Ragnarok expansion at 96fps with FSR. One of the reasons it can offer such crazy frame rates at 4K is thanks to the inclusion of DisplayPort 2.1. This is a notable benefit compared with Nvidia’s 40-series cards, which use the older 1.4a interface. Those cards are limited to 4K at 240Hz. AMD’s cards can double that to 480Hz. DisplayPort 1.4 is also limited to 8K 60Hz, but 2.1 boosts that to 165Hz. This is one of the clearest advantages AMD has this round, as some gamers were not happy with the older 1.4 interface appearing on a next-gen graphics card like the RTX 4090. Note that you will need a DisplayPort 2.1 compatible monitor for this, and those won’t be out until early 2023.
One of the most unsurprising — but still kind of surprising — parts of the presentation is the XTX is only a 355W GPU. That was hinted at before, but now it’s official. Both GPUs require only two eight-pin PCIe connectors, so there will be no “cable gate” for AMD at launch. AMD says it has improved performance-per-watt by 54 percent over RDNA2, a stunning achievement. The XT version requires 55W less power than the XTX at 30W. Both cards are also relatively small compared with Nvidia’s latest GPUs. The XTX measures just 287mm with a 2.5-slot design. The XT is 276mm and is also a 2.5-slot card. That’s a size that will fit in pretty much any case, which is not, um, the case with the RTX 4090s. AMD recommends an 800W PSU for the XTX and a 750W model for the XT.
Other ancillary features include a 1.5x improvement in ray tracing performance over RDNA2. AMD says Cyberpunk 2077 will run at 62fps in Cyberpunk at 4K with ray tracing set to “ultra” and utilizing FSR. The new GPUs are also designed for content creators and have built-in video encoding, including AV1 encode and decode. It also said it’s working on its third generation of FSR, which will offer twice the frame rates of FSR 2.0. However, it won’t be available until next year.
The Radeon RX 7900 XT is a slightly downgraded version of the flagship. Though it uses the same Navi31 die, it offers a bit less of everything. The XT sports 5,376 streaming processors, for an effective total of 10,752. It reportedly has one less memory controller chiplet, for a memory bus that is slightly narrower at 320-bit. It also has less memory than the XTX at 20GB of GDDR6 20Gb/s. Its infinity cache is relegated to 80MB. It has 84 compute units compared with the XTX’s 96.
Overall the rumors about these cards seemed to be right on the money. The XTX is clearly not as fast as an RTX 4090 or else AMD would have said so. However, it’s $600 less expensive, which is hard to believe. They claimed it was the “world’s fastest GPU under $1,000,” which doesn’t sound like too much of a stretch. AMD seems to have settled on a lineup that delivers very good performance at 1440p and 4K while not consuming a ton of power. Its chiplet design also let it save a lot of money, resulting in a less expensive GPU.
Now comes the hard part, which is waiting for third-party benchmarks. Since the GPUs won’t go on sale for another five weeks, we might be seeing them any time soon. AMD seems to be attempting to own the sub-$1,000 market, which is probably a bit larger than the “over $1,000” market Nvidia is operating in with the RTX 4090 and 4080. However, that’s just our assumption and not based on any market research. It’ll be very interesting to see what kind of numbers a 350W card can put up though. AMD had said it was going to be pushing board power up for RDNA3 but it clearly didn’t go as far as Nvidia on that front. That will make a lot of gamers quite happy, assuming the benchmarks are promising.
Now Read:
- AMD Engineer Confirms RDNA3 AMD Noise Suppression For Radeon GPUs Accidentally Leaks
- MSI Yanks Video Showing Off AMD’s New AM5 Socket (Updated)
- AMD’s RDNA3 7900XT Flagship to Offer 384-bit Memory Bus, 24GB of VRAM