Flagship GPUs tested

Modern game lighting now traces the path of every bounce of light, and that is pushing even the fastest graphics cards below smooth 4K in some new releases. In DSOGaming’s April 13 tests, an Nvidia GeForce RTX 5090 averaged 31 frames per second in *Pragmata* at native 4K with path tracing enabled. (dsogaming.com) Path tracing is a heavier version of ray tracing: instead of faking most lighting, it calculates reflections, shadows, and indirect light together. Nvidia says *Pragmata* on personal computer uses Deep Learning Super Sampling 4, ray reconstruction, and multi-frame generation to make that workload playable on GeForce RTX hardware. (nvidia.com) In that same *Pragmata* test, switching the RTX 5090 from native 4K to Deep Learning Super Sampling 4 Quality raised performance from a 31-frame-per-second average to a base range of 53 to 58 frames per second, and adding frame generation pushed output to 100, 150, and 190 frames per second depending on mode. The test system used an AMD Ryzen 9 7950X3D, 32 gigabytes of Double Data Rate 5 memory, and Nvidia driver 595.97. (dsogaming.com) That result stands out because the GeForce RTX 5090 is Nvidia’s top consumer desktop card, with 32 gigabytes of Graphics Double Data Rate 7 memory. Nvidia markets it as the flagship of the GeForce RTX 50 series, but new rendering modes are already leaning on image reconstruction and generated frames to hold 4K. (nvidia.com) Memory, not just raw chip speed, is now part of the bottleneck. Notebookcheck reported on April 15 that MSI’s new Crosshair 16 Max HX E2WGXK will ship with a previously unannounced GeForce RTX 5070 Laptop graphics processor carrying 12 gigabytes of Graphics Double Data Rate 7 memory, up from the 8-gigabyte RTX 5070 Laptop configurations MSI had sold before. (notebookcheck.net) MSI’s own April 15 announcement confirmed the new Raider and Crosshair line and said select Crosshair models reach up to 200 watts of combined central processor and graphics power. Notebookcheck said MSI did not give timing or technical details for the 12-gigabyte RTX 5070 Laptop beyond naming it in the product stack. (msi.com; notebookcheck.net) The current GeForce RTX 5070 Laptop specification most widely listed by reviewers is 8 gigabytes on a 128-bit memory bus. Notebookcheck’s November 2025 review of the MSI Crosshair 16 HX AI said 1440p ultra gaming was workable, but “VRAM-intensive titles” were a challenge on that 8-gigabyte frame buffer. (notebookcheck.net; notebookcheck.net) Desktop cards lower in the stack are under the same pressure. TechPowerUp’s review of the GeForce RTX 5060 Ti 8 gigabyte card said performance drops show up most clearly in “VRAM-heavy titles,” especially as buyers move toward 1440p, while the 16 gigabyte version has become much pricier in real stores than Nvidia’s launch spread suggested. (techpowerup.com) Digital Foundry’s early *Pragmata* coverage was more upbeat about the top end, saying a “top-spec RTX 5090” can run the game well at maximum settings and that an RTX 4060 test machine stayed above 50 frames per second across the first two levels. That split reflects how much these results now depend on whether a test is measuring native rendering, upscaling, or generated frames rather than one simple frame-rate number. (digitalfoundry.net) The short version is that the newest graphics modes are no longer a showcase only for bigger chips. They are also a test of how much memory a card carries, and of how much help players are willing to accept from Deep Learning Super Sampling and frame generation to reach 4K. (dsogaming.com; notebookcheck.net; techpowerup.com)