More Intel than Nvidia
- The extended partnership between Arm and Meta will see all of Meta’s systems optimised for Arm’s AI compute subsystems (CSS), which paves the way for a custom chip from Arm, as well as to extend the appeal of both Meta and Arm in the open source community.
- The problem is that the more that Arm looks like a chipmaker, the more it incenses the company’s existing customers, making its strategy for growth a very delicate balancing act.
- The new agreement expands and, to some degree, formalises the collaboration in two core areas.
- First, Meta Platforms’ systems: meaning the servers upon which Facebook, Instagram, WhatsApp and so on run, which are predominantly the well-established x86 systems that have been around for some time.
- All of the AI systems that do the ranking and personalisation of Meta’s digital ecosystem will now be fully optimised for Arm, which both Meta and Arm claim will deliver better performance and lower power consumption.
- This paves the way for Meta to replace x86 systems with Arm-based systems as and when the x86 systems reach end of life.
- This is a significant development because the one thing that is keeping x86 in the data centre is all of the legacy software that has been built on it and is not available for Arm.
- Meta is one of the largest developers of data centres across the world, and if all of its software is now optimised for Arm, it could accelerate the trend and further weaken the position of x86.
- Second, AI systems: where Meta’s generative AI algorithms and runtimes will be optimised for the edge using Arm’s KleidiAI software library so that they run optimally on edge devices.
- Furthermore, by harmonising how the models run both in the cloud and at the edge, the process of offering dynamic inferencing between the cloud and the edge becomes simpler.
- These optimisations will also be made available to the open source, such that developers who want to use Llama to build an AI service will also be able to benefit from the optimisation.
- This is important because it increases the appeal of Llama for developers and also the incentives for developers to ensure that the services that they build are already optimised for Arm.
- Given Llama’s scale and despite recent hiccups, this will increase the appeal of Llama in the open source, which had become far more competitive of late.
- The net result is that Arm and Meta are moving closer together, bringing the possibility of a fully custom chip designed entirely for Arm closer to reality.
- Although a fully custom chip for Meta would be much less of a threat to Arm’s existing customer base, they are still not receiving even the possibility of this eventuality very well.
- What we are seeing at the moment is an explosion of deals, projects and experiments all aimed at increasing the efficiency and reducing the cost of building data centres to run AI.
- This will lead to a period of intense competition where only a few of the new projects survive, with the majority going under or being cancelled.
- Whatever happens, Arm is going to increase its presence in the data centre as most of the new CPUs used to manage GPUs and networking or just for regular compute, are based on its instruction set.
- The question is whether it will become a fully fledged chip maker or remain a developer and licensor of technology.
- Meta is also keeping its options open, as whatever it does with Arm for inference will probably be alongside other options and suppliers with the best products going forward and the others falling by the wayside.
- Consequently, I think that this announcement is more of a threat to Intel than it is to Qualcomm or MediaTek, which spend most of their time and effort participating in the merchant market as opposed to fully custom solutions.
- However, it has increased interest and investment in RISC-V, which remains subscale but is getting bigger with every passing year.
- If RISC-V hits critical mass and standards for higher chip functions begin to form and an ecosystem is created, this will greatly increase RISC-V’s ability to compete with Arm.
- Arm currently has complete dominance in many verticals and is growing share, but for this to continue with Arm as a chipmaker, it needs to strike a delicate balance between maximising its revenue opportunity, keeping its customers happy and ensuring that RISC-V stays in its box.
- Making its own chips remains a high-reward but high-risk proposition.
