| General > General Technical Chat |
| What's up with SiFive? |
| << < (4/5) > >> |
| SiliconWizard:
They want to become like ARM? I don't want to sound ominous, but I personally don't think it's ever going to happen. I'm not going to go into all the details of why I don't, but one first major point is that RISC-V is nothing like ARM. It's an open-source ISA, and it's the main reason why it has gotten that much traction, especially in asian countries. It's a key to independence for everyone using it. ARM both owns the ISA and the cores. That's a very different business model. ARM has already had its own set of difficulties, but say they opened-sourced their ISAs and I don't give them much time before the beginning of the end for them. Yes, many are jumping on the RISC-V bandwagon, and many are doing that on their own rather than call SiFive. That's not something you could ever do with ARM. Even if SiFive was very, utterly, extremely good with their cores, customizations and customer support, that probably would never be enough to get to the point where ARM has gotten (and may fall from one of these days, probably the days are counted.) More generally speaking regarding startups in general, the business models of startups are almost never sustainable. That's not what a startup is. And startups also almost never become big, established businesses - they eventually get sold to a big fish, or die. That's what the destiny of a startup is. And in this particular case, while I don't have all figures about SiFive's results and how healthy they are, I wouldn't be surprised if they were not *that* healthy. Let's just recall that they "almost" got sold to Intel. Even if one can consider the failure a good thing here, whatever really happened, a failed acquisition for a startup is never a good event financially speaking. Just a few thoughts. I've known quite a few startups and have worked in a couple, and all this looks unfortunately pretty familiar. So anyway, good luck to them. Unless, of course, they get some "unexpected" help from the US, desperate to fight China, who knows. And curious again to see what's going to happen in the coming months/years. Also just curious to have some insider views. |
| coppice:
--- Quote from: SiliconWizard on October 25, 2023, 10:52:40 pm ---They want to become like ARM? I don't want to sound ominous, but I personally don't think it's ever going to happen. --- End quote --- SiFive appear to be backing away from producing a set of off the shelf, time tested cores, and putting their emphasis on custom cores. That would make them the very opposite of ARM. ARM hates customisation. They make it hard for their customers to have even minor customisation, unless they pay an ARM and a leg for the privilege. Have you seen the amount people pay for an "architecture licence" to create a custom ARM core? |
| jwet:
I'm not commenting on if they'll succeed, just what they're up to. Again read EE Times. I think they've got a long haul ahead but I think most people didn't really understand the press release. Below is a pretty good short read about RISC-V, etc from EET for a year back. https://www.eetimes.com/examining-the-top-five-fallacies-about-risc-v/ |
| ejeffrey:
--- Quote from: coppice on October 25, 2023, 10:59:14 pm ---SiFive appear to be backing away from producing a set of off the shelf, time tested cores, and putting their emphasis on custom cores. --- End quote --- That is claimed by the siliconangle article linked in the first post but appears be be just speculation and doesn't agree with any of the other articles or the statement from sifive which is that their existing product line will continue and they are reducing customization. Which seems kind of obvious. Even if they were going to reduce development effort on expanding their off the shelf portfolio, discontinuing your entire product line of already developed licensable IP would be a nonsensical thing to do. |
| djacobow:
It is possible to split the difference between totally custom processors and a fixed ISA. I worked at Tensilica back in the day and we built configurable and customizable processor IP. Honestly, the customization technology was far ahead of its time: you described the customization you wanted in a special language and the verilog, testbench, compiler, headers, semi-hosted simulator, System V models, everything came out ready to go. The key to any customization approach like this is putting the right constraints on what can be done. Xtensa processors could have custom datapaths, but also hardware FIFOs that led directly in and out of the processor, and various other interesting ways to communicate with the outside world independent of the load/store unit. Oh, those were heady days, and we had a lot of fun, but the idea didn't catch on for various reasons. In any case, if SiFive were pursuing some level of fully- or partially-automated customization, that would not be ridiculous. By the way, that article linked is written by David Patterson -- not exactly a nobody in this field. --- Quote from: jwet on October 26, 2023, 12:43:21 am ---I'm not commenting on if they'll succeed, just what they're up to. Again read EE Times. I think they've got a long haul ahead but I think most people didn't really understand the press release. Below is a pretty good short read about RISC-V, etc from EET for a year back. https://www.eetimes.com/examining-the-top-five-fallacies-about-risc-v/ --- End quote --- |
| Navigation |
| Message Index |
| Next page |
| Previous page |