Future of CPUs?
Future of CPUs?
Hello, it’s hard to foresee what lies ahead, only to dream it. Recently I reflected on processors—those are the things that spark my thoughts when I’m lying in bed. We’ve moved from single chips to chiplets, and now I’m pondering why we don’t push efficiency even further. My computer sits idle at roughly 60-80 watts while I write this... if we spread that demand across millions of users, we’d need a huge number of power sources. I understand operating systems don’t require a supercharged processor or graphics card, so I’m wondering why we don’t combine ARM and X86 cores into one CPU? We could handle simple tasks like OS on ARM, letting the scheduler manage heavy workloads on X86—games, for example. If that wasn’t enough, we might integrate a SOC GPU similar to what phones use for the entire user interface. You can stream 4K YouTube videos or watch Netflix smoothly on a phone without effort; it shouldn’t be a problem. When you start a game or edit videos, dedicated GPUs would handle those intensive jobs. This approach would resemble a phone’s SOC running on a CPU with X86 cores, and we could leverage the chiplets technology to build everything from smaller parts. I really appreciate Apple for creating such powerful yet compact CPUs. Or at least giving us the option to disable certain cores—like turning off 12 or 24 threads—to save power.
taking into account the electrical benefits of switching to LED bulbs, the efficiency of today’s devices has minimal impact on the power grid when compared to the lack of use in air conditioning or other appliances. this highlights a minor issue with basic assumptions about CPU operation—CPUs don’t waste much energy when idle, as they naturally produce heat only when active. This explains why CPUs never get excessively hot during inactivity. Generally, the range below 100 watts isn’t considered high power consumption; it’s roughly equivalent to two porch lights from the early 2000s. While income varies and some people struggle with electricity costs, most can easily manage the expenses of low-power systems.
I wouldn't anticipate this gaining momentum for a while due to several factors. The main concerns are that building an x86 translation layer would be simpler than creating a chip with dual x86 and ARM cores, and the operating system would need a completely different design compared to today's systems. This would demand significant effort, making it more practical to push developers toward ARM directly instead of x86. ARM technology is already impressive—think Apple’s M1 and M2 chips outperforming many high-end Intel processors. If scaled appropriately, they could match or surpass current performance levels. The real bottleneck right now is software support for ARM on Windows, which remains very poor. If Microsoft aligned its Windows strategy with ARM and a different processor manufacturer emerged, adoption would likely accelerate faster than with a hybrid approach like x86/ARM. That said, x86 hardware could still be more power-efficient if Intel refined its efficiency cores. While Intel’s current designs waste a lot of energy (a 13900K is essentially an eight-core powerhouse), future generations might balance performance and efficiency better. Ultimately, the outcome may shift toward a new perspective on x86 or a completely different architecture—either ARM or RISC-V—taking over. Neither transition will be fully realized in ten years, but I’m not sure about the details. It’ll be fascinating to revisit this in a few years and see what’s accurate.
Current chips are already quite effective. Intel offers both high-performance and efficiency cores, while AMD is exploring similar options using chiplets. However, flexibility, modularity, and ease of use often come at the cost of some efficiency. Devices like smartphones run on lower power because they have less hardware and components tailored for minimal energy use. They lack features such as SATA controllers, USB ports beyond a few, PCIe slots, and robust VRMs designed for wide power ranges. Their power supplies must handle higher voltages (110V/220V ±10%) and convert them to lower levels like 12V, 3.3V, and 5V—tasks that are far more efficient. Phone makers can fine-tune DC-DC converters knowing typical usage stays within a narrow range, usually 10–30 watts.
In contrast, phones use much smaller power supplies optimized for minimal consumption. They don’t need to support high TDP processors or complex expansion boards. Their chipsets are streamlined, often built on advanced processes like 5-12nm, and integrate essential functions directly into the processor. This design reduces power draw significantly compared to PCs.
The motherboard’s internal components also differ: phones lack a full chipset, instead relying on integrated processors, while PCs use discrete chipsets with multiple specialized modules. Network cards, audio, and memory all have their own power management strategies—phones prioritize wireless connectivity and eliminate bulky external parts like stereo jacks or separate sound chips.
RAM usage varies too. Computers typically use dual or quad channels for better performance, especially in gaming. Phones usually use single-channel LPDDR memory, which is less efficient for demanding tasks.
VRMs on motherboards must handle a broad range of power demands, whereas phone designs are more constrained. Power management software can fine-tune consumption based on actual usage patterns, making phones more adaptable.
Overall, while phones excel in simplicity and battery life, modern processors continue to improve efficiency, especially with advanced manufacturing and intelligent power control.
I’m planning to host my PC inside a DIY camper (Class B model) and running 100W continuously seems demanding, especially if powered by a battery. Simply saying we don’t need it isn’t the right approach. This is why Americans emit twice as much CO2 as the entire Europe despite similar populations—we have ample oil reserves, so why not use a 6L Hemi truck for a quick toast? Sure, some operations would struggle on those x86 processors, but the whole system could run smoothly on ARM-based components. Probably not zero performance loss, just a tiny fraction (around 0.01%). AMD has developed chiplets, so I don’t see much reason to change the controller or add an ARM module. It wouldn’t be much work. For Intel, 100% ARM would seem unlikely. Translation layers add unnecessary overhead without real benefit.
I pictured the CPU I mentioned would function like Intel’s e-cores—executing identical tasks on separate cores uses more power than running a full workload at 100W for an hour. There are plenty of YT videos showing tests that confirm this. The key difference is I’d prefer e-cores over p-cores, assuming they’ll be far more efficient.
Another point: Intel uses monolithic chips (one chip), whereas AMD relies on chiplets. I don’t mind chiplets; the efficiency gains matter more. Looking at the bigger picture, many people still favor x86 PCs because of familiarity and performance expectations. Fast-forward to today—apps like Hello World load in seconds, opening a browser feels effortless. That’s impressive compared to 20 years ago when servers were the norm.
Back then, we were more efficient since real programmers optimized everything. Now, with tools like Electron, launching a simple app is nearly instantaneous. This shift also reduces performance overhead because it doesn’t require heavy OS cores.
As for SATA controllers, they’re typically found on the motherboard, not the CPU. The mainboard handles them, so that’s not a major concern. Similarly, IDE controllers used to be on the CPU, but modern systems separate these functions.
You have the option to install a 3000ge on an A320 chipset board, which should suffice for watching movies and uses between 20 to 30 watts. Alternatively, you might opt for a laptop equipped with a low-power CPU (10-15 watts) and an AMD or Intel processor, ensuring smooth movie playback without the need for complex power conversion.
In your vehicle, a laptop would be preferable since it avoids dealing with inverters and power supplies that convert AC to DC. Regarding SATA controllers, they are found in both CPU and chipset designs on AMD systems, but ARM-based chips may lack them entirely, relying instead on PCIe lanes or eMMC modules. The CPU’s memory controller is integrated, yet modern processors allocate more space for dual channels and require enhanced design to handle longer traces.
ARM chips typically feature simpler memory controllers that accept only a few specialized chips, which are arranged differently internally. These aren’t built to support multiple RAM modules as general-purpose CPUs can. ARM processors achieve efficiency by reducing silicon usage—cutting unnecessary components and using fewer transistors. For instance, some models omit floating point units or integer division, relying on clever bit manipulation instead of hardware division. This approach minimizes transistor count while maintaining performance, allowing more RAM to fit in compact designs.
I’d bring up ARM and X86 again, but it looks like you’re avoiding the details... In short, I think explaining this would be tough for you. TL;DR—I feel like chatting with you is like trying to push against a wall... floating-point calculations work better on X86 than ARM. Threads can be closed since I don’t see much value in them.