IBM announced on Tuesday (8th) that it will launch a series of new data center chips and servers, focusing on more power-efficient than competitors and simplify the deployment process of AI in enterprise operations.
IBM released the new Power11 chip that day, which is the first major upgrade to the Power chip series since 2020. It is known that the Power11 processor is very cost-effective in energy use. It not only has higher core performance, but also provides an optional power-saving mode to bring higher efficiency and efficiency to the data center.
Although the number of cores per chip of Power11 is the same as the previous generation Power10, IBM says each core has a significant increase in efficiency. In addition, a new computer function called "resource groups" has been added this time, which can maximize system usage and improve overall efficiency without additional energy consumption.
Power11 chip also provides energy mode to expose some of the efficiency and change to 28% energy efficiency improvement. IBM also claims that the Power11 system is "double times that of x86 systems" in the performance-per-watt part.
Power chips have competed with Intel and AMD in the data center field for many years, especially in vertical application fields such as financial services, manufacturing, and medical care. IBM pointed out that the Power11 system will be available from July 25 with the goal of improving the reliability and security of the system.
For daily maintenance, the entire system is also designed to upgrade and replace parts without interruption. Power11 supports hot-plug fans, power supplies and I/O components, making it easier for technical personnel to repair without interruption.
At the same time, Power11 has real-time security protection capabilities, and the system can detect and respond to ransomware attacks in less than a minute, which helps companies respond quickly when facing security threats.
IBM also announced that the Power11 chip will be integrated with the AI chip Spyre launched last year in the fourth quarter of this year. The AI independent accelerator card Spyre consumes 75 watts of power, is equipped with 128GB of LPDDR5 memory and has 300 TOPS computing power. It is known that this acceleration card is similar to NVIDIA L4 in terms of power consumption and performance levels, but the memory capacity is more than 5 times, indicating that IBM aims to integrate high performance and large models in AI recommendation applications.
IBM Power System General Manager Tom McPherson pointed out that IBM's goal is not to compete with NVIDIA for the AI training market, but to focus on simplifying AI's application in the "inference" stage, that is, the practiced AI model is actually applied to accelerate business processes.
McPherson pointed out that AI functions can be integrated into the system for recommendation acceleration and helping enterprises improve process efficiency. Although these systems do not have the strong computing power required to train large models, they are excellent in their recommended performance and easy to integrate, which is expected to bring real benefits to enterprise AI applications.
IBM boasts new Power11 chips are stingy on power usage IBM rolls out new chips and servers, aims for simplified AI Extended reading: Trump will charge 50% tax on copper imports, and New York copper futures sold at 17% at a record high Trump is 50% tax responsibilities and import tax responsibilities 200%