By: ⚛ (0xe2.0x9a.0x9b.delete@this.gmail.com), August 15, 2022 5:28 am
Room: Moderated Discussions
Brett (ggtgp.delete@this.yahoo.com) on August 14, 2022 5:11 pm wrote:
> You also need competition to overcome inertia, x86 is going to be losing badly
> on all fronts in just a few years, and it takes 5 years to design a CPU. X86
> needs to design a hail marry now to compete and win against ARM.
Given the fact that [at my location] I have only 2 CPU vendors to choose from (AMD or Intel) when buying a desktop machine, the issue here is beyond the level of just needing competition in the form of ARM or RISC-V for x86's market share to start decreasing.
The desktop CPU market isn't liberalized - it is basically a duopoly today. The capital required for a 3rd or 4th player to enter this market is extremely large. If market liberalization laws were in place in this sector of economy, both AMD and Intel would be bound by law to design their platforms so that it is much easier for (1) an alternative CPU provider, and (2) an alternative CPU core provider, to enter the market.
A major obstacle in the above-mentioned direction is the concept of the "main CPU". Ideally, there should be no "main CPU" in a desktop machine and the device that is today known as the "main CPU" should be just an ordinary accelerator board that happens to be efficient at executing general-purpose (GP) computations, enabling the user of the desktop machine to install multiple, potentially different, GP compute boards. (It is harder to avoid the concept of the "main CPU" in notebook machines.)
-atom
Brett (ggtgp.delete@this.yahoo.com) on August 14, 2022 5:11 pm wrote:
> You also need competition to overcome inertia, x86 is going to be losing badly
> on all fronts in just a few years, and it takes 5 years to design a CPU. X86
> needs to design a hail marry now to compete and win against ARM.
Most predictions of the sort "x86 is going to totally lose to ARM in just a few years" will most probably turn out to be mispredictions in just a few years. Two basic reasons for this are that (1) every next x86 CPU design increases x86 IPC and/or performance per watt by a non-negligible margin, and (2) the limits of x86 IPC haven't been reached yet (in other words: nothing - except for the mathematical complexity of the CPU design itself and the time required to solve this as a math problem - is preventing a single x86 CPU *core* to sustain executing 10 x86 instructions per clock on average in most workloads).
-atom
> You also need competition to overcome inertia, x86 is going to be losing badly
> on all fronts in just a few years, and it takes 5 years to design a CPU. X86
> needs to design a hail marry now to compete and win against ARM.
Given the fact that [at my location] I have only 2 CPU vendors to choose from (AMD or Intel) when buying a desktop machine, the issue here is beyond the level of just needing competition in the form of ARM or RISC-V for x86's market share to start decreasing.
The desktop CPU market isn't liberalized - it is basically a duopoly today. The capital required for a 3rd or 4th player to enter this market is extremely large. If market liberalization laws were in place in this sector of economy, both AMD and Intel would be bound by law to design their platforms so that it is much easier for (1) an alternative CPU provider, and (2) an alternative CPU core provider, to enter the market.
A major obstacle in the above-mentioned direction is the concept of the "main CPU". Ideally, there should be no "main CPU" in a desktop machine and the device that is today known as the "main CPU" should be just an ordinary accelerator board that happens to be efficient at executing general-purpose (GP) computations, enabling the user of the desktop machine to install multiple, potentially different, GP compute boards. (It is harder to avoid the concept of the "main CPU" in notebook machines.)
-atom
Brett (ggtgp.delete@this.yahoo.com) on August 14, 2022 5:11 pm wrote:
> You also need competition to overcome inertia, x86 is going to be losing badly
> on all fronts in just a few years, and it takes 5 years to design a CPU. X86
> needs to design a hail marry now to compete and win against ARM.
Most predictions of the sort "x86 is going to totally lose to ARM in just a few years" will most probably turn out to be mispredictions in just a few years. Two basic reasons for this are that (1) every next x86 CPU design increases x86 IPC and/or performance per watt by a non-negligible margin, and (2) the limits of x86 IPC haven't been reached yet (in other words: nothing - except for the mathematical complexity of the CPU design itself and the time required to solve this as a math problem - is preventing a single x86 CPU *core* to sustain executing 10 x86 instructions per clock on average in most workloads).
-atom