GPU comparison · 1080p benchmark
RTX 5090 32GB recorded 784 FPS against 287 FPS for RX 6800 XT 16GB. These are continuous 1080p benchmark averages, not predicted FPS in any particular game.
RX 6800 XT 16GB
287Benchmark FPS at 1080p- Our rating
- 70.3 / 100
- Graphics memory
- 16GB
- Released
- November 2020
- GPUsed cards tested
- 49
RTX 5090 32GB
784Benchmark FPS at 1080p- Our rating
- 100 / 100
- Graphics memory
- 32GB
- Released
- January 2025
- GPUsed cards tested
- 46
Change either GPU in the comparison tool
What does this comparison mean?
The RTX 5090 more than doubles the RX 6800 XT's result and doubles memory from 16GB to 32GB. The size of that benchmark gain does not establish that such a large upgrade is necessary for your games or applications.
RX 6800 XT 16GB has 16GB and RTX 5090 32GB has 32GB of graphics memory. Memory capacity, board dimensions, power connectors and the price of a particular used card are separate from this benchmark. Compare the exact item before deciding whether to upgrade.
The two GPUs in context
RX 6800 XT 16GB
The RX 6800 XT introduced RDNA 2 to this 16GB GDDR6 model in November 2020.
RTX 5090 32GB
The RTX 5090 launched in January 2025 as a Blackwell desktop GPU with 32GB of GDDR7. It is the fastest card in this GPUsed benchmark dataset, which makes its compressed rating 100.
Release and feature sources: RX 6800 XT 16GB source · RTX 5090 32GB source.
How the figures are compared
Both FPS figures come from the owner-supplied GPUsed benchmark dataset. GPUsed reports a continuous FurMark run at 1080p for its physical-card checks. The percentage compares the higher model average with the lower one using the original FPS values. Our rating uses a compressed scale, with the fastest GPU at 100; it is separate from the percentage difference. A benchmark result does not predict gaming FPS.
GPUsed reports its physical-card checks typically last 30 to 60 minutes while temperature and stability are checked. The owner supplied the model-level FPS workbook on 2 October 2026; this benchmark dataset was last reconciled 3 October 2026. Historical cards tested are separate from the number of FPS runs behind a model average. Full benchmark method.