regarding china, etc: the current political climate will end very soon. in the US, quasi-nationalist trump will be replaced with a pax-americana democrat. xi jinping will be replaced by a pro-west faction who will want to do business with the US. this will mean scuttling all their nationalist projects in favour of buying things from the US ("we're friends with the US now, no need for this second rate domestic stuff"): homegrown riscv replaced with arm cores, homegrown gpu stuff replaced with nvidia, etc. same thing goes for russia, and i dont think i even need to mention the middle east at this point in time.
10 years from now riscv will join the ranks of openrisc, but in the meantime i suppose there's no harm in 'distributing' academic grants and investor money.
The ISA war is really a two front issue: compute, and control. Control implementation vastly outnumbers compute. Heck, for any given compute motherboard with 4 or 32 cores, there is 10-50 control processors, each with 1-4 cores. A car has hundreds. Many lightbulbs have one or two cores. Basically everything electronic has an MCU these days, because a microcontroller is actually cheaper than a nand chip. Even advanced transistors as used in automotive switching often have a microcontroller, albeit a very rudimentary one.
In this world, riscV is eating ARM and everything else alive slowly but surely.
RiscV will take longer to mature into a good application processor, but it inevitably will. It will take over Arm for the same reason Arm is eating away at x86. My prediction is that in 20 years, it will be the predominant architecture for most sequential computing applications.
I think the driver here is cost. You don’t see 50mhz ARM cores for $0.07 for a reason, and that reason is that the licensing of the ARM ISA does not make that possible.
So at the low end, riscV is 1/2 to 1/3 the cost for comparable performance based on licensing costs alone. In a market where discrete components sell for thousandths of a dollar, a few cents is huge and a dollar is catastrophic. The difference between ARM and riscV can easily be a 40 percent reduction in BOM cost, or significantly improved features and performance at the same cost.
Being Chinese there has been some suggestion that they haven't actually paid for the CM0 core, but it's supported in Keil and I think listed somewhere on the main Arm site too, which tends to rule that out.
(WCH has already introduced successors to the CH32V003 with more RAM and flash at the same $0.10 price -- and even the CH570D with 96 MHz, 12k RAM, 240k flash, USB, and a 2.4 GHz packet radio -- but they are not as widely available yet)
God save the ARM
It's just an instruction stream
There is no future
In England's dreaming
Don't be told what you want to want to
And don't be told what you want to need
There's no future, no future
No future for you
But then the money and experienced talent started pouring in. Tenstorrent "Ascalon" 8-wide core comparable to Apple M1 or AMD Zen3 is taping out right about now, they want to see it widely used, including in laptops (which I'd expect to see next year).
That's five years or so behind Apple, but pretty close behind Qualcomm.
They say they'll catch Apple / Intel / AMD performance in 2027. That'll be cores. SoCs and stuff you can buy in a shop will be a year or two later. But this decade.
There are a number of others hard on their heels, including open-source XiangShan from China.
Will they be running Windows or MacOS? No. But they'll be running Android, Fedora/RHEL, Ubuntu/Debian and they'll be in data centres and phone and tablets.
Heck ... there are new phones introduced in 2025 with only Arm A53 cores from 2012 e.g. using the Mediatek Helio G50 SoC with 8x A53 such as the Infinix Smart 9 HD, Tecno Spark Go 1S, Tecno POP 9 4G (Indian model).
The majority of RISC-V SBCs and laptops for sale at the moment are similar specs to those phones, but the better ones are 2x as fast with another factor of 2x or 4x coming in the next 12 months.
> [RISC-V] peaked in 2022, and has long since run out of steam.
Not a chance.
v5v3•7mo ago