YouTube channel Benchmark Lab recently tested Intel's new Arc A370M (Alchemist) mobile GPU against Nvidia's GeForce GTX 1650 Mobile in 10 gaming benchmarks. The GTX 1650 won the matchup with a performance gap of around 15% to 20% in every single title.
Since the Arc 3 series launch three days ago, we've seen several user benchmark comparisons of Intel's A370M and A350M GPUs against AMD and Nvidia's newer and older entry-level GPU options (RTX 3050/Ti, GTX 1650, RX 6500M, etc.). Intel's GPUs lost their battles consistently, whether in synthetic benchmarks such as 3DMark or real gaming applications. In addition, AMD recently shared benchmark results comparing its own Radeon RX 6500M and Intel Arc A370M GPU, with the 6500M being nearly 2x faster than the Intel counterpart. Of course, we recommend a pinch of salt when looking at vendor-provided benchmarks.
The Arc A370M is part of Intel's new entry-level Arc 3 series of GPUs the company launched just a few days ago. The A370M sits as the highest model in the Arc 3 lineup, featuring 8 Xe cores and 4GB of GDDR6 memory on a 64-bit wide bus. Depending on the laptop manufacturer, power consumption can range from 35 to 50W.
The Arc A370M is similar to the GTX 1650 Mobile in the sense that both have 1,024 shaders and 4GB of GDDR6 memory at 12 Gbps. However, Nvidia's GPU does have a higher base clock speed and a wider memory interface to offer more memory bandwidth.
The laptops used in the gaming benchmarks include the Acer Nitro 5 packing a Ryzen 5 5600H and a GTX 1650, and a Samsung Galaxy Book 2 Pro features a more powerful Core i7-12700H Alder Lake CPU and an Arc A370M discrete GPU. However, it's not exactly a perfect comparison since the Ryzen 5 5600H is a last-generation chip, whereas the Core i7-12700H is the latest processor from Intel.
Games tested include Microsoft Flight Simulator 2020, Red Dead Redemption 2, The Witcher 3, CS:GO, Cyberpunk 2077, Fortnite, Diablo 2 RE, World of Warcraft, Horizon Zero Dawn, and Forza Horizon 5.
Despite Samsung's Galaxy Book 2 Pro having a much more powerful CPU than the Acer Nitro 5, performance was entirely in favor of the Acer notebook, with a 15% to 20% performance gain overall, depending on the specific title. It begs whether or not we're seeing severe driver optimization issues.
A few days ago, we reported alleged die sizes and transistor counts for Intel's two current Arc Alchemist GPU dies, the smaller G10 and the bigger G11. What we found was surprising, Intel's top-tier die, the G11, which is responsible for powering Intel's flagship mobile GPUs such as the A770M, is substantially smaller than both AMD and Nvidia's flagship desktop dies. As a result, its performance also suffers because of its size.
For context, the G10 is Intel's largest GPU die we know of right now and should be responsible for powering both desktop and mobile solutions. Unfortunately, it means Intel isn't even considering competing with Nvidia and AMD's flagship options and will probably compete with their mid-range GPUs instead.
If this is true, Intel's product stack effectively has fallen a few notches from where we once thought it would be in the GPU market. So in this scenario, we can put the A350M and A370M as competitors against Nvidia's MX solutions instead of its GTX product stack.
If you haven't heard of Nvidia's MX GPUs, they are effectively Nvidia's replacement for the "GT" nomenclature in the mobile space and are Nvidia's lowest discrete GPU options you can buy. Instead of targeting gaming performance specifically, the MX series is designed for hardware acceleration instead -- think of them as faster versions of integrated graphics.
Performance and specs support this idea, with the latest Nvidia MX series GPUs -- the MX550 and MX570, both run on Nvidia's older Turing architecture and feature heavily neutered memory configurations compared to the GTX 1650. Including a 64-bit bus and up to 4GB of GDDR6 memory. Ironically this is the same memory configuration as the A350M and A370M.
According to leaked benchmarks, the MX550 is also slower than the GTX 1650 and is more comparable to a GTX 1050 Ti. With the MX570 being on par with a GTX 1650's performance. So, in this case, the MX550, in particular, shows very similar performance to Intel's A350M and A370M GPUs. If this scenario is actual, it would undoubtedly reveal why the A350M and A370M are slow.
