AMD Radeon RX 10800 XT Reportedly Beats RTX 5090 by 15-25% in 4K Gaming and Local AI
An unverified GameGPU report says AMD's next flagship beats Nvidia's RTX 5090 by 15-25% in 4K gaming and local AI inference, a lead AMD has not held in years.
A report from the hardware outlet GameGPU claims AMD's unannounced Radeon RX 10800 XT outperforms Nvidia's RTX 5090 by 15% to 25% in 4K gaming and local AI inference. If accurate, it would be the first time in years that AMD's top consumer card has clearly beaten Nvidia's flagship on raw performance. Neither AMD nor Nvidia has confirmed the card exists, let alone the numbers attached to it.
The margin, if it survives scrutiny, is not a rounding error. The RTX 5090 has been the default pick for buyers chasing maximum 4K frame rates or running large language models on their own machines, and a 15-25% deficit at the top would put that default in question. GameGPU's figures cover both rasterized 4K gaming and local AI inference, the latter of which has become a second front in flagship GPU competition.
That second front matters more each cycle. Enthusiasts running quantized Llama and Mistral models at home weigh VRAM capacity and memory bandwidth above ray tracing features, and AMD's recent generations trailed badly on CUDA-dependent software. A claimed lead in AI workloads implies either a serious software push from AMD or a hardware spec large enough to overwhelm the software gap. Neither has been demonstrated publicly.
The report is unverified, and early leaks of this type have a long record of overstating gains before retail silicon ships. Reviewers will want to know whether the 15-25% figure holds across a broad game suite or comes from a handful of titles that favor AMD's architecture, and whether it persists with ray tracing enabled. Pricing and final specifications are also unknown, and both could change the comparison entirely.
If the numbers hold, the fallout reaches past gamers. Nvidia's pricing power at the top of the consumer stack rests on the 5090 going unchallenged, so a faster AMD card would force a response on price or a quicker refresh. AI developers who buy consumer hardware instead of datacenter GPUs could also shift build decisions toward a cheaper, faster card with enough VRAM.
The open question is verification. Independent benchmarks, official specs, and retail pricing are the only things that will settle the 15-25% claim, and none of them exist yet. Until cards reach reviewers, this stays a rumor with real momentum behind it.