Qualcomm Adreno 640 vs NVIDIA GeForce GTX 1070 (Desktop) vs Qualcomm Adreno 643

Qualcomm Adreno 640

The Qualcomm Adreno 640 is a smartphone and tablet GPU that is integrated within the Qualcomm Snapdragon 855 SoC. The chip will be available from early 2019 and will be used mainly in high-end Android devices.

According to Qualcomm, the Adreno 640 GPU offers a 20% improved performance over the Adreno 630, its predecessor, which is integrated in the Snapdragon 845 SoC. The enhanced performance is achieved by the Adreno 640 having 50% more FP16 and FP32 compute units (ALUs) than the Adreno 630, while its 7 nm FinFET manufacturing process makes it 20% more efficient too.

The Adreno 640 will also support the Vulkan 1.1 API, something which is possible for other Adreno 600 series GPUs too.

NVIDIA GeForce GTX 1070 (Desktop)

The Nvidia GeForce GTX 1070 is the second fastest Pascal based graphics card in mid May 2016. It is a high-end desktop graphics card. The mobile pendant is also called GeForce GTX 1070, but features more shaders and a lower core clock (see this page for details on the laptop version of the GTX 1070). Both cards are based on a cut down GP104 chip as the GTX 1080 and manufactured in 16 nm FinFET at TSMC. More details on the Pascal architecture including all features and comparisons to Maxwell, can be found here.

Compared to the GTX1080, the GTX1070 features less CUDA cores (1920 versus 2560) that are clocked a bit lower (maximum Boost 1683 versus 1733). Instead of the new GDDR5X, the GTX 1070 uses slower GDDR5 graphics memory - but still 8 GB. The TDP is also reduced from 180 Watt to 150 Watt.

The performance of the GTX 1070 is between the old GTX 980 Ti and the GTX 1080. Therefore, gaming in 4k at high details is possible with the card and also VR gaming is possible.

Qualcomm Adreno 643

The Qualcomm Adreno 643 is an integrated graphics unit (iGPU) that is used in the Qualcomm QCM6490 and QCS6490. The SoC is not only intended for use in Android-based smartphones and tablets, but also for IoT products with Linux, Ubuntu or Windows 11 IoT Enterprise. In the QCM6490 and QCS6490 the 643 clocks at 812 MHz.