AMD Radeon 890M RDNA3.5 graphics are 39% faster than Radeon 780M in OpenCL test

AMD Ryzen AI HX 170/370 in first leaked benchmarks

AMD was rather reluctant to share details on the RDNA3.5 during the official keynote. What we know is that it uses an updated architecture and features more cores than last-gen products. Both can contribute to higher performance.

Over at Geekbench, one can find three scores featuring the Ryzen AI 9 HX 170 processor in CPU and GPU tests. As we explained in the previous post, AMD changed the naming schema for its Strix Point processors shortly before announcements, leading some companies to display laptops with older SKU names during their demos. Apparently, somebody got their hands on one of these units, which is likely how the leak surfaced.

RYZEN STRIX POINT SCORES, Source: Geekbench

The Ryzen AI 9 HX 170/370 is a 12-core Zen5 processor with a maximum boost capability of 5.1 GHz. AMD has not yet confirmed the exact clocks for the ‘denser’ Zen5c cores that this processor has, but we do know that it will feature 12 RDNA 3.5 Compute Units, which will boost up to 2.9 GHz.

What we do not know is what the TDP was set for these tests, but we do know tests were performed on an ASUS ProArt A16 laptop, which is an APU-only laptop without discrete graphics, so the APU wattage might be higher. AMD is only saying that the default TDP is 28W, but it may be in the 15W to 54W range.

Additionally, the Strix Point CPU core did not boost to 5.10 GHz; instead, the detailed result page shows a max boost of 4.25 GHz, meaning it didn’t reach the highest frequency. This could be due to a lower TDP or the fact that the HX 170 is not the final retail SKU but an engineering sample for demonstration purposes.

So, for this post, we will focus on the graphics, which are less prone to TDP changes once the APU hits 28W. The GPU scores 41,995 points in OpenCL tests, which is hardly a good benchmark when gaming is considered, but this is all we have for now.

In short, this means that the 890M (listed as 880M here) is 39% faster than the Radeon 780M (RDNA3) and 59% faster than the 680M (RDNA2). It also outperforms the GTX 1650 Ti Mobile, which was a popular marketing point for the 780M graphics when Mini-PC companies compared raw TFLOPS. Now, it’s a real synthetic score.

Of course, OpenCL tests are hardly something we would use for GPU comparisons, but this is a good sign that core count update aside, there is clearly some architectural improvement. Also, it is worth noting that both 890M and 780M can boost up to 2.9 GHz.

Ultimately, this could lead to a complete reshaping of the entry-level desktop and mobile GPU market. These cards make little practical sense anymore unless they are specifically added to systems for their encoding capabilities. Integrated GPUs are now faster and far more power-efficient, so it’s definitely exciting to see AMD bringing these APUs to desktops.

Source: Geekbench #1, #2, #3 via BenchLeaks