ASRock Radeon RX 9070 Steel Legend: Performance, Power Analysis & Noise Output

AMD’s RX 9070 series managed to disrupt the market, because of their decent prices, still notably higher than MSRP ones, and good performance. To put it midly, they were a pleasant suprise by AMD, and a tough opponent for NVIDIA’s RTX 5070 and 5070 Ti. In today’s review I will look at the ASRock Radeon RX 9070 Steel Legend model.

The AMD Radeon RX 9070 series consists of two members: the 9070 XT and the plain model. According to AMD, it sits between the RX 7900 XT(X) and RX 7900 GRE lines. The RX 9070 models have up to 64 CUs following the RDNA 4 architecture, and their GPUs consist of 53.9B transistors. The new GPUs are suitable for QHD and 4K gaming. The specs of both 9070 cards look similar, with the fundamental differences being on the CU units and the TDP. So, the 9070 XT has 64 CUs and 304W TDP, while the 9070 has 56 CUs and 220W TDP.

The 9070 series utilizes the RDNA 4 architecture, which includes updated ray tracing and AI accelerators. Combined with the 16GB of VRAM, these accelerators offer strong gaming performance. To improve ray tracing performance, AMD doubled the ray intersection rate. The ray intersection rate is the speed at which a system calculates how many rays intersect with objects in a 3D scene during ray tracing. AMD also used 2x ray accelerators per CU to handle all required calculations quickly and efficiently. According to AMD, the updated RT engine “doubles the performance for both ray-box and ray-triangle testing” compared to RDNA 3. Moreover, the RT engine uses 128KB of shared memory, which might look small, but it is okay for this operation.

Furthermore, AMD moved to BVH8, a type of Bounding Volume Hierarchy (BVH) used in ray tracing. It organizes objects in a scene into 8-node groups for faster ray intersection tests, improving performance by reducing the number of calculations needed to find collisions or intersections. Lastly, to improve RT performance, AMD introduced oriented bounding boxes to reduce or eliminate false positives in ray-triangle and ray-box intersections and a dedicated ray transform block to help with transformation. A dedicated ray transform block is specialized hardware in GPUs designed to accelerate the transformation and manipulation of rays during ray tracing. It handles tasks like ray-object intersection calculations more efficiently, improving overall ray-tracing performance.

We shouldn’t forget AMD’s new software technologies, like FSR 4, an upscaling technology that transforms low-resolution frames into higher-fidelity ones, increasing performance. The FSR 4 upscaling algorithm now uses an advanced AI model, which exploits the RDNA AI accelerator for improved performance. However, we will talk more about FSR 4 on the next page.

The tables below show the specs of the RTX 50/40 series, AMD, and Intel models.

NVIDIA Graphics Cards Major Specs

RTX 50 Series

RTX 40 Series

The table below contains the specs of AMD’s current-generation graphics cards for reference purposes.

AMD Graphics Cards Major Specs

Intel Graphics Cards Major Specs