These include the Cortex-A78 CPU core, Mali-G78 GPU core and the Ethos-N78 machine learning core. The company claims 20 percent, 25 percent and 25 percent performance gains over previous cores. In the first two cases these improvements include improvements due to manufacturing process while in the last case its just an area improvement versus the Ethos-N77.
So in the case of the Cortex-A78 that is a 3GHz A78 on 5nm FinFET versus a 2.6GHz A77 on 7nm FinFET process. That would suggest that more than half the improvement is coming from the manufacturing process. ARM says the Cortex-A78 provides 50 percent energy savings over 2019 devices at the same performance as Cortex-A77.
A fourth announcement is the Cortex-X custom program which allows an additional tier of performance but is only available to customers who sign up to the Core-X program. The first result of this program is the Cortex-X1 with an 8.3 percent increase in peak performance over the Cortex-A78, ARM claims.
The Cortex-X1 is expected to find use in flagship smartphones or those with large displays. Samsung would appear to be one customers that has joined the Cortex-X program.
“Samsung and ARM have a strong technology partnership and we are very excited to see the new direction ARM is taking with Cortex-X Custom program, enabling innovation in the Android ecosystem for next-gen user experiences.” said Joonseok Kim, vice president of SoC design team at Samsung Electronics, in a statement issued by ARM.
The Mali-G78 is based on the Valhall architecture as was last year’s Mali-G77. The Mali-G78 supports up to 24 cores whereas Mali-G77 supports 16 cores. Mali-G78 is also 15 percent better at supporting machine learning. For those that do not require premium performance in graphics ARM has introduced the Mali-G68 with the same feature set as the Mali-G78 and up to 6 cores.
ARM said it is partnering with video game developer Crytek GmbH (Frankfurt, Germany) to brings Cryengine software to Android mobile phones.
But for many smartphones dedicated machine learning support is required. For those cases ARM is introducing the Ethos-N78 neural processing unit (NPU) with a performance of from 1TOPS scalable up to 10TOPS. It offers twice the peak performance of the N77 and 25 percent more energy efficiency, ARM claims.
Related links and articles: