Comment by _yosefk | Hacker News Reader

Comment by _yosefk | Hacker News Reader

Parent Full thread

Well, currently shipped SPARC's volumes are relatively low AFAIK, and ARM and MIPS (now IMG) are in fact monopolies, they just happen to license the ISA & the implementation because they choose to do this. Intel does not choose to let AMD and VIA ship x86 CPUs and in this sense it's less of a monopoly. Of course you could argue that ARM is more open than x86 in some other significant sense; here I think that the PC is a more open platform than perhaps any widely shipped ARM-based system, in terms of the variety of available operating systems and software and the ease of developing such software, especially if said software is to have reasonably wide applicability and long shelf life, and also in terms of the degree of control that any single entity has over the platform.

In general I think that the "vertical disintegration" where you have fabs which are open to everyone making chips integrating cheap IPs, instead of companies with their own fabs, has destroyed "openness" - it used to be the case that you knew exactly what the hardware was doing, and it is now the case that you need to sign a draconian NDA to program most of the components in most chips, and this option is only available to select "partners." x86 is an older product and hence more open than most successful newer products; it was perhaps less open than its competitors at the time, which in part explains why it was more successful than those competitors... (A closed platform is a kind of a high-risk, high-reward game.)

OK. Well, we have to consider that. There's companies making money on the ISA's and implementations. Intel's main business is the ISA and implementation of x86. By that standard, they have quite a huge chunk of a oligiopoly with only 1 real competitor and a niche competitor in low-power space. For ARM, there are at least 10 companies that fully licensed and extended the I.P. plus academics doing their own. Additionally, Intel and AMD extend their own stuff in their semi-custom business whereas I can extend ARM or MIPS myself after licensing it. The PC builds on blackboxes from Intel and AMD plus a few others that differentiate at the board level. Barely a comparison to ARM, MIPS, etc differentiating at SOC and transistor level.

So, Intel is dominant in their ISA, its implementation, and in a profitable way. ARM and MIPS don't come close despite their monopolies on the ISA's themselves.

"x86 is an older product and hence more open than most successful newer products; it was perhaps less open than its competitors at the time, which in part explains why it was more successful than those competitors... "

This is news to me. Far as I knew, it dominated desktop due to backward compatibility and other strategies. It wasn't open in many ways. SPARC and PPC even adopted OpenFirmware while Intel still had closed microcode updates. It's still the least open ISA of the high-sellers.

"In general I think that the "vertical disintegration" where you have fabs which are open to everyone making chips integrating cheap IPs, instead of companies with their own fabs, has destroyed "openness" - it used to be the case that you knew exactly what the hardware was doing, and it is now the case that you need to sign a draconian NDA to program most of the components in most chips"

That I agree with. Fortunately, there's work from companies and academics to change that with actual silicon being prototyped. This will be an uphill battle but is becoming feasible. The main drawback is that the cost of EDA and prototyping means closed I.P. might stay the norm. Only exception is if we can get academics to put key I.P. in public domain as they develop stuff. PCI or USB here, analog stuff there at 45-90nm at least. Make it cheaper.

"Far as I knew, it dominated desktop due to backward compatibility and other strategies. It wasn't open in many ways." - we're agreeing here, I said "it was less open than competitors at the time, which was perhaps one reason for its success." The only part we're disagreeing about is, I think it's more open than newer products (as a part of the general trend of hardware getting more closed.)

I worked on chips which included both ARM and MIPS CPUs; I don't think it makes ARM or MIPS "open" in many of the ways that matter, because, for one, programming those chips without the consent of the company that made them is very problematic both technically and legally. So ARM and MIPS in practice are "open" to select partners who make chips which are in turn "open" to select partners, unlike the PC where you have not only Windows, Linux, BSD etc. but things like MenuetOS and TempleOS and it's both legal and technically feasible to do this kind of thing. Of course this isn't due to Intel or anyone else behind the PC being more benevolent than other vendors, it's just an older platform and closeness wasn't or at least didn't seem as important for profitability as it is today. Here BTW Intel's focus on backward compatibility is very helpful - low-level software for the ARM or MIPS becomes obsolete more quickly and developing working low-level software for newer version of these in a timely manner might require partnering with the vendor, AFAIK. Basically ARM and MIPS are way more open than x86 to me as a chip architect (although an ARM architecture license AFAIK costs $30M or so... meaning that a lot of chip makers ought to take whichever implementation ARM provides and you're at their mercy to a much greater extent than you'd often desire), but they are far from fully open and then they're way less open than x86 to the users.

And then both MIPS and ARM are pretty old... the trend of really closing everything, where you can't get any spec without partnering with the vendor, didn't take off until maybe the 2000s.

I wonder what it'd take for open IPs to gain significant traction in hardware. It's easy to come up with reasons for them to fail or succeed but it's not easy to predict which reasons will end up more dominant.

"I worked on chips which included both ARM and MIPS CPUs"

That's really neat. Especially as I work a little on methods for secure HW/SW design. Mostly done with digital side of ASIC methodology but still custom digital and analog stuff to figure out. Appreciate any tips you have on such threads where I'm exploring.

"So ARM and MIPS in practice are "open" to select partners who make chips which are in turn "open" to select partners, unlike the PC where you have not only Windows, Linux, BSD etc. but things like MenuetOS and TempleOS and it's both legal and technically feasible to do this kind of thing. "

So, back then, ARM and MIPS I.P. holders didn't allow a PC-like product to be built with their chips? It was custom negotiated for every product with no platforms available? I could see you calling that less open and the ecosystem hit it would have.

"I wonder what it'd take for open IPs to gain significant traction in hardware. It's easy to come up with reasons for them to fail or succeed but it's not easy to predict which reasons will end up more dominant."

Inherently safer or language-specific ISA's like jop-design.com or crash-safe.org. Microcontrollers where any royalties are eliminated to further lower costs. OEM's targetting hobbyists and/or idealists that want maximum freedom. Anti-subversion efforts where multinational pride kicks in with an ISA like RISC-V being remixed allows everyone to feel like the owner. Tools or projects for processor design in academia where they just want something to work with and build rather than screwing around with I.P. and legal risk. Wait, that last one already happened and I already named it. ;)

So, given you've done hardware, what do you think of my recommendation that market wanting OSS hardware just build on Gaisler's SPARC products? There's a configurable CPU (Leon3) and substantial I.P. library already GPL'd or licensable. Probably closer to MIPS than ARM in price. There's also a 4-core version. One could put them right in an ASIC with Pi- or Novena-style boards w/ SPARC ecosystem (esp BSD/Linux) leveraged immediately. Alternatively, really my touch, is swapping out SPARC-specific stuff (eg instruction handling) for RISC-V so all the I.P. just magically supports a RISC-V solution. Open I.P. can gradually be developed to replace it, maybe keep using Gaisler's proprietary source where compelling, stuff is 90+% ASIC-proven from start, people can inspect, paid get paid, everyone is happy.

Nowadays there is ARM SBSA which aims to resolve the problems and allow a PC-like product to be built (the main reason such a thing didn't exist before is lack of interest). In the meantime the GPL helps to ensure at least the source code for the drivers for the Linux kernel are revealed.

BTW, notice the "server" in the ARM SBSA name.

Regarding the PC ecosystem vs the ARM and MIPS ecosystem - I'm not sure what happened back when MIPS and ARM made chips, just that most of their volume came when they were IPs, and at that point chip makers were keeping things rather closed. I'm pretty sure both Intel and say MS do not see the PC as the ideal model for their business, they just happen to profit more from it than they would from an attempt to kill it.

Generally, today ARM will sometimes break OS-level code between releases - something Intel would never do - and I think ARM does this because it sells an IP, not chips, and so in any case the only way to sell a CPU to a consumer is through a chip maker licensing ARM and then porting OSes to the new chip - and the chip maker can handle ARM-related breakage while they're at it. But it does not follow that ARM is to blame for most chips being closed and/or hard for an OS unsupported by the chip maker to work on the chip across its various generations; I'd guess that most of the difficulty comes from the chip-level breakage and lack of documentation, not CPU-level issues of this sort.

So IMO it's not that ARM or MIPS actively prevent(ed) the emergence of an open platform like the PC while Intel was pushing for open platforms; it's that Intel happens to have succeeded mainly selling chips for an open platform, while ARM and MIPS happen to have succeeded mainly selling CPU IP for various closed platforms.

Regarding the SPARC IP that you mentioned - I'd have to look at it more deeply to have an informed opinion. Generally for a chip vendor who wants source code, as opposed to wanting an open source IP, MIPS is just fine, they'll give you the source. Otherwise what you want from your CPU vendor varies (ARM for instance gives you not just the CPU but everything you need to build say a cellphone application processor - probably not the best one on the market, but a completely functional one. At the level of the CPU itself, you might want a bunch of things - like hardware virtualization support, or cache coherence with other processors, etc. etc. - that different vendors support differently, if at all.) So as I said, I don't think I have a serious opinion about the product you mention.

Reply on news.ycombinator.com