Foresight: Intel has finally started releasing its mobile Arc GPUs and is set to release its first desktop Arc GPUs later this summer. Somewhere near the top of the stack will be the Arc A770, a fully loaded piece with 512 EUs, a 2-2.5 GHz boost clock, and 16 GB of memory.
Intel first announced the Arc flagship hardware as the ACM-G10 GPU, then launched it on paper as the A770M for mobile. Notebooks equipped with the A770M should appear on the shelves in the coming months. Its unadvertised desktop equivalent has now been found in the Geekbench OpenCL database.
It’s not a groundbreaking leak as the hardware itself has been leaked so many times, including once before in the Geekbench database. It is, however, the first time that it is officially called Intel Arc A770 Graphics to give it its full name.
This is also another point on the clock speed map: 2.4 GHz. Late last year we talked about 2.5 GHz rumors, then in February we saw 2.4 GHz for the first time. Intel itself teased 2.25 GHz last month, but may have been referring to a different model.
And, the critical point: aligning with past leaks, the A770 earned an OpenCL score of 85,585 points. A little more than the RX 6600 XT, about the same as the RTX 2060 Super and RTX 2070, and a little less than the RTX 3060. In other words, completely mid-range.
But that’s not the whole story. Geekbench breaks down the OpenCL score into its component categories, of which there are 11. Some architectures are better in some classes than others. For example, the 6600 XT is around 35% faster than the 3060 in the Gaussian blur test, despite being slower in eight categories and having a worse overall score.
Although the 3060 outperforms the A770 in more than half of the categories, the A770 takes a significant lead in particle physics and Gaussian blur tests. It loses by the largest margins in the Sobel, Histogram Equalization, Horizon Detection, and Canny tests, all of which are based on computer vision.
From there, you can see a bit of a trend: the A770 does poorly in memory-sensitive tests but is otherwise computationally strong.
Given the A770M’s on-paper specs, it’s a surprising result that the A770 likely shares: 16GB of GDDR6 clocked at 17.5Gbps and connected via a 256-bit bus. It’s not a bad memory subsystem, so maybe it’s an OpenCL quirk, or maybe the Alchemist architecture has a memory bottleneck.
Realistically, there would be no significant ramifications if that were the case. At worst, it could make the architecture more suitable for lower resolutions than higher resolutions and reduce its longevity. Still, it’s interesting to see what differentiates Alchemist’s architecture from Ampere and Navi.
In terms of gaming performance, the OpenCL scores don’t suggest much. As noted above, the 6600 XT is over 10,000 points behind the 3060, but in our review we found it to be faster in most games at 1080p and 1440p. Benchmark leaks before we knew its name put the A770 ahead of the 3060 and into the realm of the 3070 Ti.
