Glonce

Engineer rebuts RISC-V critic with developing-world cost math

infrastructureThe Glonce editorsHacker News🔥73Glonce rating73Impact55Novelty60Relevance72Surprise60Players30Credibility82

Dmitry Grinberg's essay "RISC-V: They Should Have Known Better" reached the Hacker News front page and drew a separate argument on Lobsters. An embedded engineer, writing from Trinidad and Tobago and publishing on rvembedded.com, posted a long response arguing that the critique, while technically fair on individual points, misses the reason RISC-V matters outside the US and Europe. The author does not defend the RISC-V International committee itself, noting the group denied the author's own membership request, and concedes several of Grinberg's specific complaints about compressed store offsets and the separate Zicsr extension.

The author's central argument is about access, not encodings. Getting a development board shipped to Trinidad and Tobago means checking whether a seller ships there at all and what customs will do to the package; "Free Shipping" offers from Digikey, Mouser or other US and European sellers do not apply. The author reports paying US $60 to US $200 to ship one-dollar chips that customers elsewhere receive for free, and says a PCB company that had considered sponsorship turned the author down solely over shipping to that location. The author also cites Hussein Ali, a YouTuber from NorthridgeFix, as a separate case of Digikey refusing an order. The author frames students in Trinidad and Tobago, Nigeria and Bangladesh as the group this problem actually falls on: the difference between a ten-cent part and a one-dollar part decides whether a class of thirty students each gets a chip or watches one demo board, if any.

The author points out that Grinberg's own essay, before its complaints, derives from first principles what a cheap microcontroller core needs (low interrupt latency, small die area, good code density, no hardware divider or multiplier, no privilege separation) and arrives at RV32EC, then states directly, "I am 100% sure that RISC-V will own the cheap-as-dirt single-use microcontroller space eventually." The author calls it "almost satirical" that Grinberg derives the case for the chip and then spends the rest of the piece annoyed that the chip exists.

The disagreement the author calls strongest is with Grinberg's core claim: "the things a high-end CPU needs are diametrically opposed to the things a small cost-saving microcontroller core needs," and that no single ISA can serve both. The author answers with three RISC-V parts used firsthand. The CH32V003 is a ten-cent RV32EC chip with sixteen registers, no multiplier or divider, machine mode only, 2KB of SRAM and 16KB of flash; the author has shipped it in a bin monitor with a time-of-flight sensor and an air tag, and in an agricultural product that opens and closes a door on a schedule. The CH32H417 is a dual-core chip with a 400 MHz core and a 144 MHz core, 896KB of SRAM, 960KB of flash, and an integrated USB 3.2 Gen1 transceiver at 5 Gbps; the author has run a web browser, a GAN-based image generator and real-time facial recognition in under 150KB of RAM on it. The Baochip is a VexRiscV part with an MMU that runs the Xous microkernel (designed by bunnie Huang), with ports to seL4 and Linux also in progress; the author wrote its bare-metal C SDK, and the chip shipped inside the DEFCON 34 badge.

All three parts share the same base instruction set, register model, calling convention and toolchain, which the author says let the same assembly-level skills carry from the ten-cent chip up to the Linux-capable SoC within hours of setup each time. The author puts the total cost of exploring that entire stack, from the CH32V003 to the Baochip, at under $100 in under a year, excluding data-center compute: about $7 for 50 CH32V003 chips with a debugger, $20 for a CH32H417 evaluation board, and $35 for two Baochip Dabao boards including shipping (a single board was $9.50 on Crowd Supply). The OrangePi RV2, a RISC-V single-board computer the author unboxed and had running Linux in about five minutes, cost $30 with free shipping at release, though the author notes its price has since risen.

The author contrasts this with ARM: moving from a Cortex-M0 part like the STM32F030 up through Cortex-M7 to an application-class Cortex-A processor means substantially more relearning at each step, because they are different ARM profiles with different privilege, exception and system models. No Cortex-M microcontroller integrates a USB 3.0 SuperSpeed PHY; the nearest comparable dual-core ARM part, the STM32H747, does not have one either, carries a twenty-week manufacturer lead time in its TFBGA240 package, and costs about $60 to $100 or more to ship to the author's location. A Segger J-Link debug probe, the standard ARM debugging tool, costs about $600, or roughly $200 for an EDU-edition unit including shipping, with no chips or boards included.

The author's closing argument is that the gap Grinberg treats as an architectural fact, that a small core and a large core have opposed requirements, is on ARM chips actually a product boundary enforced by licensing, not physics: no one can simply add an MMU to a Cortex-M part. The captured text ends mid-sentence at this point, describing how RISC-V differs because "the ISA owner does not decid[e]" that boundary; the rest of the argument is not present in the fetched text.

Key facts

  • Shipping alone costs the author $60 to $200 to receive one-dollar chips in Trinidad and Tobago, against free shipping elsewhere; a PCB sponsor reportedly declined solely over shipping to that location.
  • Grinberg's own essay derives RV32EC as the correct core for a cheap microcontroller and states "I am 100% sure that RISC-V will own the cheap-as-dirt single-use microcontroller space eventually," which the author says undercuts his own complaint.
  • The author documents three shipped RISC-V parts spanning the range Grinberg calls incompatible: the ten-cent CH32V003 (16 registers, no multiplier or divider, 2KB SRAM), the dual-core CH32H417 (896KB SRAM, integrated 5 Gbps USB 3.2 transceiver), and the Baochip, which has an MMU and runs the Xous microkernel plus in-progress seL4 and Linux ports.
  • Total cost to gain hands-on experience across that entire RISC-V stack, ten-cent chip to Linux-capable SoC, was under $100 in under a year, versus about $600 for a single Segger J-Link debug probe used with ARM parts.
  • The author argues the small-core/large-core split on ARM is a "PRODUCT boundary... enforced by licensing" rather than an architectural necessity, since no Cortex-M part gets an MMU and the nearest ARM alternative, the STM32H747, carries a twenty-week lead time.

Why it matters

The RISC-V versus ARM debate is usually argued on encoding elegance and interrupt latency. This response reframes it around who can physically obtain working hardware: an open instruction set that any vendor can implement means chips at a range of prices and sources, rather than parts gated behind a single licensor's product tiers and a shipping network built around US and European customers.

Who it affects

Engineers, hobbyists and students in countries poorly served by standard electronics distribution, the author names Trinidad and Tobago, Nigeria and Bangladesh, along with anyone trying to teach embedded systems on a budget where the gap between a ten-cent and a one-dollar chip decides whether a class shares one demo board or each student gets a chip.

How to use it

The chips cited span a wide range: the CH32V003 at roughly ten cents each (about $7 for 50 with a debugger), the CH32H417 evaluation board at about $20, and the Baochip Dabao board at $9.50 each, or $35 for two including shipping, all available through channels like the official WCH AliExpress store or Crowd Supply without the ID checks the author says Digikey and Mouser sometimes require.

How solid is it

This is a single engineer's first-person rebuttal on a personal blog, not an independent study; the claims about products shipped, books written and demos built are the author's own account of firsthand work, and the article does not give its own name in the running text or a publication date. The captured source text also cuts off mid-sentence before the author finishes the closing argument about ARM licensing.

Risks and caveats

Costs cited are snapshots that shift: the author notes the OrangePi RV2's price has already risen from its $30 release price. The framing of ARM's small-core/large-core split as a licensing-enforced "product boundary" is the author's own characterization, not verified here against ARM's licensing terms. The Digikey refusal cited for Hussein Ali is presented only as a parallel example; the source does not say it stemmed from the same cause as the author's own shipping difficulties.

“Instruction set elegance is a thing you can afford to care about once the hardware is already on your desk. Whether the hardware can get to your desk at all comes first.”

— the responding engineer, on rvembedded.com