Intel Arc B570 review: a budget GPU that's too good to be true?

One of the biggest pain points in producing budget GPU reviews is this - the chances of the card being paired with an older or less capable CPU are much, much higher than any other class of graphics card. This raises a highly pertinent question: to what extent is testing with the most powerful CPU actually relevant to the way the product will be used by the majority of its users? In terms of theoretical max performance, the Ryzen 7 9800X3D used to produce the benchmarks on the preceding pages will show you exactly what the card is capable of. But what happens if we use a more common, older CPU?

We chose the Ryzen 5 3600 for our tests, but why? Well, it seems that Intel has a driver overhead problem. The task of the CPU isn't just to run game logic, physics, simulation and the like - it also has to prepare instructions for the GPU and that's where the driver comes in. A more efficient driver means more CPU time for running game logic, a less efficient driver means much lower performance in CPU-limited areas of any game... but how do we test it?

Our methodology is straightforward enough: move over to our CPU benchmark suite for a couple of games and test both Intel Arc B570 and Nvidia GeForce RTX 4060. Of course, we should expect to be CPU-limited with the Ryzen 5 3600 and GPU-limited with the 9800X3D. The game itself will overwhelm the 3600 but the extent to which performance decreases should expose how much the driver is contributing to the problem. Put simply: the 9800X3D results here show maximum potential performance from the Nvidia and Intel cards tested here. The difference between results in the Ryzen 5 3600 results show the impact of driver efficiency in a CPU-limited scenario - and it's clear Intel loses more of its performance relative to GeForce. A final point: to ensure realistic workloads, we're using two games not at ultra settings, but rather at optimised or console-equivalent settings.

Looking at Cyberpunk 2077 first of all, we've set up the PC to run at the equivalent to PS5's performance mode. With the Ryzen 7 9800X3D, we are not CPU-limited on either RTX 4060 or Arc B570. Those results show the hardware running at its best. The Ryzen 5 3600 is clearly having issues and on both cards we are losing a lot of performance. At 1080p resolution, we lose around 60.9 percent of performance on Arc, while GeForce loses around 50 percent of its frame-rate. The sobering reality is that a 13 percent performance advantage for RTX 3060 with the 9800X3D becomes 45 percent on the 3600. The difference is vast, and by a process of elimination that must come down to driver efficiency - to catch up with GeForce in a CPU-limited gaming scenario, you'd need a much more powerful CPU.

At 1440p, the importance of the CPU diminishes a touch - we are GPU-limited for more of the sequence. However, there is still a clear GeForce driver advantage. An RTX 4060 advantage of around six percent with the Ryzen 7 9800X3D becomes 37 percent with the Ryzen 5 3600. There's an interesting situation at 4K, however. RTX 4060 performance even with the 9800X3D drops to the point where the Arc B570 is now 14 percent faster - which is possibly down to PCIe limitations. However, swap over to the Ryzen 5 3600 and GeForce beats Arc performance by 34 percent. Looking at these results, now we can begin to wonder just how relevant the past seven pages of benchmarks actually are to the majority of people who will buy a budget GPU.

Starfield shows that CPU driver overhead issues are problematic, but more to the point, there are scenarios where Arc performance is downright baffling and doesn't really make sense. At 1080p resolution, the RTX 4060's advantage in a GPU-limited scenario with the Ryzen 7 9800X3D is a mammoth 24 percent. And those issues are exacerbated when the B570 is paired with the Ryzen 5 3600 - the 4060 is now over twice as fast. The differentials are only slightly diminished when looking at the 1440p results too.

Bizarrely, perhaps, it's at 4K where the differentials are most stark. Comparing 9800X3D to 3600 performance, the RTX 4060 seems to have mostly hit GPU limits - there's just the loss of around two percent in frame-rate between the same GPU tested on two very different systems. The Arc B570 remains problematic though, with a 38 percent drop in output frame-rate.

There's a bit of a problem here with these Intel results. Benchmarks are all about apples to apples testing, right? However, in more real world scenarios where CPU performance is at a premium, the bottom line is that you'll need more processor performance to match GeForce (and based on other results seen online, Radeon too). The implications are that while Intel's GPU is cheaper by some margin, you may well need to invest those cost savings into a higher performance CPU to achieve fully like-for-like results with competing graphics cards. This is something Intel really needs to work on.

Intel Arc B570 Analysis

• Introduction
• RT benchmarks: Alan Wake 2, Avatar: Frontiers of Pandora
• RT benchmarks: Dying Light 2, F1 24, Hitman: World of Assassination
• RT benchmarks: Indiana Jones and the Great Circle, Metro Exodus Enhanced Edition, A Plague Tale: Requiem
• Game benchmarks: Alan Wake 2, Black Myth: Wukong
• Game benchmarks: F1 24, Forza Horizon 5, Senua's Saga: Hellblade 2
• Game benchmarks: Hitman: World of Assassination, A Plague Tale: Requiem
• Driver overhead testing: does Arc underperform with older CPUs? [This Page]
• Conclusions, value and recommendations