NVIDIA GeForce GTX 1070 Ti (Desktop) vs Apple M3 8-Core GPU vs Apple M3 Max 30-Core GPU

NVIDIA GeForce GTX 1070 Ti (Desktop)

The Nvidia GeForce GTX 1070 Ti is the second fastest Pascal based graphics card in November 2017. It is a high-end desktop graphics card based on a cut down GP104 chip as the GTX 1080 and manufactured in 16 nm FinFET at TSMC. Compared to the GTX 1080, it uses the more available GDDR5 graphics memory, and compared to the GTX 1070 it offers more shaders (2432 versus 1920). More details on the Pascal architecture including all features and comparisons to Maxwell, can be found here.

The performance of the GTX 1070 Ti was in our tests right between the GTX 1070 and GTX 1080. However, the powerefficiency was worse than the GTX 1070 and near the GTX 1080. Nvidia also specifies the 1070 Ti and 1080 at 180W.

Apple M3 8-Core GPU

The Apple M3 GPU is an integrated graphics card offering 10 cores designed by Apple and integrated in the Apple M3 SoC. It uses the unified memory architecture of the M3 SoC (up to 24 GB LPDDR5-6400 with 100 GB/s bandwidth). The GPU uses a new architecture with support for mesh shading and ray tracing. Dynamic caching is also a new feature that should only us as much shared memory as needed.

The old M2 SoC showed a difference of around 15% between the 8-core and 10-core variant. Therefore, we expect a similar difference for the M3 models. This means, the M3 8-core version should be slightly faster than the 10-core GPU in the Apple M2.

The M3 chip is manufactured in the new 3nm process at TSMC (most likely N3B). The power limit should be around the same 15 Watt as the 10-core variant.

Apple M3 Max 30-Core GPU

The Apple M3 Max 30-Core GPU is a self-designed graphics card in the Apple M3 SoC and offers 30 of the 40 cores that are included in the chip. According to Apple, the GPU uses a new architecture and now offers dynamic caching, mesh shading and ray tracing acceleration.

The graphics card can access up to 96 GB LPDDR5-6400 unified memory via a 512 bit memory bus (max 400 GBit/s). Thanks to dynamic caching, the GPU only reserves as much memory as is required.

Since the SoC is manufactured in the current 3nm (probably N3B) at TSMC.