I’m likely not the first to tell you that the memory crisis has really done a number on PC gaming, but nothing quite exemplifies this like AMD releasing a graphics card with 4-8 GB of VRAM in 2026. That card is called the RX 9050; it starts at $279, and it is limited to certain regions. One early review suggests performance isn’t great, but its specs are looking a little funky.

That review is from Japanese hardware outlet Hermitage Akihabara, which managed to get its hands on an 8 GB-toting ASRock model. Testing the card in 3DMark, Hermitage argues it “falls short of the 30fps generally considered the benchmark for comfortable single-player gameplay. Furthermore, its performance is only about 75% of that of its competitor, the GeForce RTX 5050, and just over 60% compared to the higher-end Radeon RX 9060 XT” (machine translated).

The RX 9050 doesn’t lag behind the RTX 5050 in absolutely all tests, though. The RX 9050 matches the RTX 5050 in 4K testing on Rainbow Six Siege X, and it performed better in Doom Eternal at 1080p and 1440p, though this was with the help of FSR, which is not an apples-to-apples comparison. Still, the RTX 5050 wipes the floor with the RX 9050 in pretty much every other test conducted by Hermitage.

At 1080p, Hermitage saw 192.4 average fps in Apex Legends, and 147 average fps on the 9050. In Cyberpunk, on Ultra, Hermitage saw an average of 142 fps on the AMD card and 240 fps on the Nvidia card. In fairness to AMD, though, this is with both cards using the frame generation and upscaling models of their respective cards, so not a 1-1 comparison.

With the RTX 5050, and its 8 GB of GDDR6 VRAM, launching at $250, that means that the RX 9050, according to these tests, is substantially slower than the Nvidia card. Of course, that’s assuming you can get one at MSRP, but we can’t assume you will find AMD cards at MSRP either.

The RX 9050’s score of 4531 in 3D Mark’s Port Royal and 1873 in Speed Way places the RX 9050 close to the RTX 3070, as noted by hardware analyst Harukaze5719. They also note a specs page with 32 ROPs on the card, which is half of the advertised 64 on AMD’s own specs page. Fewer ROPs, or Render Output Units, in a card means worse performance, and that would be reflected in tests.

We saw a similar problem with eight missing ROPs in some RTX 50-series cards around their launch, which Nvidia claims only affected 0.5% of cards. If this s a defect in the tested card, one might expect better real-life performance.

That’s assuming those ROPs are actually missing and it’s not a defect in the software reporting itself, or some other problem in between. That’s all to say that while early figures for AMD’s latest budget card aren’t looking very good, we’ll have to wait for some more real-world testing from other outlets to see if it actually is so thoroughly outpaced by the RTX 5050.

Either way, early signs aren’t looking great for the budget card market. That’s especially true when you consider the AMD price increase rumours that dropped just today.