News SiFive selects a faster Chinese-made RISC-V CPU instead of an Intel chip for its latest development board

DavidMV

Commendable
Nov 18, 2021
31
56
1,610
Probably because Intel terminated their RISC-V development project last year. I like the idea of RISC-V, but lets be honest, it is a couple decades away from wide adoption (if it ever happens). ARM is still gaining momentum and x86-64 has so much legacy that it is almost impossible for a third ISA to break into the market.
 

JamesJones44

Reputable
Jan 22, 2021
789
723
5,760
Widely adopted might be a bit of a stretch... Other than Espressif Systems SoCs from China and a few FPGA cores, they are few and far between.
I wouldn't be sure of that statement. Apple and Google are already deploying RISC-V SoCs in their devices and QNX supports a few different RISC-V variants which are supposedly in use by both Automotive and Telcos.

https://www.patentlyapple.com/2021/...cross-all-operating-systems-using-risc-v.html

https://jobs.apple.com/en-il/details/200530453/arm-risc-v-high-performance-team-manager
 
  • Like
Reactions: bit_user

bit_user

Titan
Ambassador
Probably because Intel terminated their RISC-V development project last year.
You're mischaracterizing that. The goal of Intel's efforts were to help jump-start the RISC-V ecosystem, which their efforts had achieved by the time they were wound down. At the time, Intel themselves said there were already better resources available than their configurator tool, for example.

I like the idea of RISC-V, but lets be honest, it is a couple decades away from wide adoption (if it ever happens).
It already has wide adoption in certain embedded markets. As faster & more capable cores become available, it becomes possible to talk about entering new markets with it.

ARM is still gaining momentum and x86-64 has so much legacy that it is almost impossible for a third ISA to break into the market.
ARM is helping drive the development & refinement of emulation technology that is creating an easy path for RISC-V to follow. Going from one ISA to two is much harder than going from two to three!
 

bit_user

Titan
Ambassador
Not in my personal opinion. It's better than SD card storage by about 2x, but still really poor when it comes to random reads and writes.
Would that not depend at least somewhat on the details of the storage chips used in the eMMC?

Anyway, I have eMMC in one of my SBCs (ODROID-N2+) and I find it to be quite adequate for casual use. For anything more serious, I'd want to use NVMe, if possible. Even at just PCIe 2.0 x1, that's still 500 MB/s, which is way faster than eMMC and still offers advantages over SATA (even if peak sequential speed is a little lower).
 

JamesJones44

Reputable
Jan 22, 2021
789
723
5,760
Would that not depend at least somewhat on the details of the storage chips used in the eMMC?

Anyway, I have eMMC in one of my SBCs (ODROID-N2+) and I find it to be quite adequate for casual use. For anything more serious, I'd want to use NVMe, if possible. Even at just PCIe 2.0 x1, that's still 500 MB/s, which is way faster than eMMC and still offers advantages over SATA (even if peak sequential speed is a little lower).
That's fair. I assumed the question pertained to personal use like a mini computer for work which I would not consider eMMC fast bootable. However, for non-time critical embedded and non-interface solutions like a home automation or HMS, etc. it's more than adequate (better than SD cards).
 

bit_user

Titan
Ambassador
That's fair. I assumed the question pertained to personal use like a mini computer for work which I would not consider eMMC fast bootable.
Coming from my prior experience of using SD cards on a Pi, it certainly felt fast. That board was the first such device I used that felt reasonable close to the responsiveness of a regular desktop computer. As for boot times, they didn't seem excessive.

Here's how Hardkernel characterized the eMMC performance of the ODROID-N2+

"Sequential read and write speed is over 150MB/s and 125MB/s respectively.
4K random access performance is reasonably fast too. iozone test result are as follows.

table-2.png


Source: https://www.hardkernel.com/shop/odroid-n2-with-4gbyte-ram-2/
On the same page, they included data for a UHS SD card and it's about half of the eMMC performance listed above. On my Pi, it was even way slower than that!

However, for non-time critical embedded and non-interface solutions like a home automation or HMS, etc. it's more than adequate (better than SD cards).
For such tasks, you can use just about anything.
 
Jun 14, 2024
2
2
15
Widely adopted might be a bit of a stretch... Other than Espressif Systems SoCs from China and a few FPGA cores, they are few and far between.
JH7110 is the most widely adopted Risc-V SoC (Vision Five 2, Milk-V Mars,Pine64 Star64) Lacks vector support but has decent GPU BXE-4-32.
T-Head TH1520 is next most common SoC (LicheePi 4A devices, Milk-V Meles) Has 2 channel memory and RVV 0.7.1 vector support along with decent imagination GPU BXM-4-64.
Spacemit K1 is the newest SoC in use (Banana Pi BPI-F3, Musebook, MusePi, Roma Laptop v2) and it is an Octa-Core design with full RVV1.0 vector support. Only single memory channel and more basic BXE-2-32 GPU holds it back though.

At this point you would have to be mentally impaired to bet against Risc-V, it is the future. The open source nature of the ISA and lack of licensing fees almost guarantees its success.

There are thousands of Risc-V products and projects worldwide. It's only a matter of time until main stream adoption happens. The launch of the Milk-V Oasis ITX board later this year should help.

User level OS support on the desktop still has a way to go. The best OS platform on current dev boards is Bianbu Linux for Banana Pi BPI-F3 and other spacemit based platforms. Debian runs very well on TH1520 boards. JH7110 is officially supported by Ubuntu but the installs are server targeted leaving the user to install and configure desktop GUI.
 
Last edited:
  • Like
Reactions: bit_user
Jun 14, 2024
2
2
15
  • Like
Reactions: bit_user