Leaving aside the insensitivity of this post towards laid-off staff - which is bad enough - it doesn't actually tell current / prospective customers what has changed about SiFive's business and products.
Clearly something has, but this provides no real clues. Completely counterproductive. But all is good as SiFive will revolutionize computing in ways we have not yet begun to imagine!
Layoffs aside, it's hard as an enthusiast consumer to get excited about RISC-V for the same reason it's hard to get excitrd about ARM. Availability is limited to prepackaged SBCs and tinkering-discouraged consumer products.
I'd be so excited if I could buy a chip and a motherboard and just drop them into a standard ATX case. But I don't think there exists a consumer-purchaseable RISC-V or ARM chip that isn't a whole system on a chip.
ATX boards exist but are not cost effective. The economies of scale of the x64 ecosystem make it impossible to beat. Are you really interested in a 1500$ board and CPU combo that runs slower than an average Best Buy laptop? AMD and Intel are incredibly _cheap_ for the power you get.
I've never understood why I as a consumer, even as a developer, would want to get interested in RISC-V, or any architecture, since I'm not sure what, if anything, would be noticeably improved or different to me.
Unless it's for curiosity, then that makes sense to me.
The only thing Apple Silicon improved on is battery life/power consumption/thermals and yet there are lots of people who think it's better across the board despite faster x86 chips existing on the market.
I think excitement about architecture among end users is often not based in reality as the whole point of operating systems and high level languages is to abstract it all away.
I'll say upfront that you are right; developers getting excited about new architectures generally does not make sense.
But here is why I'm excited about RISC-V: I write my software in C, and I want it to be as portable as possible. RISC-V is something different that will exercise new issues when my code is tested on it.
It's a dumb reason, yes. But I want a 32-bit chip and a 64-bit chip for testing. I want to have a server rack full of different chips with different architectures, RISC-V included, and set up a commit pipeline that kicks off a massive test suite on all of the VM's on those machines, running different OS's.
I want to test my code on hundreds of OS/arch platform sets.
RISC-V is a different architecture with different things. I want it.
The RISC-V instruction set is not encumbered by intellectual property. This means they are cheaper to make, licensing costs being 0. Cheaper to make can translate to the customer into cheaper to buy.
This seems to assume that development of ISAs is done for free, which is not the case.
When a firm pays Arm a licensing fee for the Arm ISA that fee helps support the development of the Arm ISA.
When someone who works for SiFive or Tenstorrent etc works on the development of the RISC-V ISA, they are being paid and those costs will be recovered through the price paid for the products they sell. Others may use the ISA at zero cost but someone is paying.
Now you could argue that costs will be lower in aggregate because of greater competition, but the cost isn't zero in total.
This does include a presumption, but not the one that you're proposing: that the ISA development is done for cheaper than when a single profit-driven entity is responsible for it.
AMD & Intel also pay $0 for licensing costs. For the phone markets & up through servers, licensing costs are a rounding error on what you pay. You're paying for the R&D of actually designing the chip itself, not the ISA, as well as for the actual manufacturing itself.
If you're comparing between two products and one is $1500 and the other is $1501, is that $1 really entering your list of pros/cons? No, of course not. It's irrelevant at that point, literally any other difference (even what color the dumb thing is) is immediately more significant. That's my point, there's a point at which the increased end-price as a result of the licensing is such a small % of the total price as to simply no longer be significant for the consumer buying the product. And phones are already expensive enough that the SoC being 1-2% more expensive simply doesn't matter in any meaningful degree to the end purchase price. Nobody is looking at the $1200 iPhone 15 Pro Max and going "damn, if only this was $0.50 cheaper, then it'd be all over it..."
> Consumer devices with ARM cores can be had for as low as 2 USD (sensors, breakout boards).
Yes, and that is the market where RISC-V is actually appealing. Once you go higher end, the appeal of a no license cost ISA rapidly drops off and stops being a significant factor.
That is definitely going to be nice, but not enough for me to really care too much about it now vs things that actually affect me like software tools, languages, etc.
Yep Intel just gives me free hardware LOL, yeah, no...
So anyway, the reason is whenever risc-v makes sense to switch to for cost and performance, I think I'll do it at that time. There is no reason I can see for me to think about it until then. Literally it's going to be exactly the same for me except maybe doing something like `apt install somepkg-risc-v` for some packages and I'm pretty sure that will be the only thing different for me.
x86 has turned assembly into voodoo while RISC-V is very simple to understand and teach. This is true at the basic level, but is also true for something like vectors vs AVX where one is quite easy to teach while the other is massively complex. More devs who can do this means we get more higher-performance code for cheaper. This will be a very important milestone for software.
Also leaving aside this post, solidrun's honeycomb boards are a mini itx form factor that is ARM SystemReady. You can run windows on arm on it, and it has UEFI if you want that, meaning no need for a BSP. Have booted several Linux distributions on it.
It's limited ports wise (has a single nvme slot, but a ton of networking) and if you want to use it for a desktop you need to replace the noisy fan that comes with it. But I can actually be done.
An awful lot of marketing puff sounds like GPT3 output.
One good consequence of generative "AI" might be that real people start discounting meaningless bafflegab marketing prose as automated nonsense, even if it was deliberately composed by a human. Which in turn might drive PR-type people to learn to write meaningful English.
Incidentally, I thought GPT3's summary of the post (upthread) was rather good.
It's interesting that that they are arguing strongly (vs Qualcomm) for retaining an approach in the ISA that probably requires macro-op fusion - whilst holding at least one patent that relates to macro-op fusion.
I don't think the argument is that much about macro-op fusion vs cracking, as both sides agree that highperformance implementations are doable and not hinderes much buy both.
I think the real dispute is how the opcode space and code density will evolve when more and more extension will need to be added.
Will more 32 opcode space and 64 bit instructions but no 16 and no 48 bit instructions in the longterm be a better choice than fewer 32 bit insrructions, but 16/48/64 bit instructions?
Neither side is correct IMO. Qualcomm wants 32-bit only instructions, but that leaves out many use cases that require longer instructions.
On the flip side, a 48-bit instruction takes FIVE bits just to specify the length which is a massive amount of overhead.
I think 64-bit instructions which contain between 1 and 4 instructions (not VLIW as no ILP is specified) would be a better solution offering more compressed instructions and much lower length encoding overhead for 48 and 64-bit instructions while still leaving a potential opening for very long instructions composed of several 64-bit groups. This beats out the Qualcomm proposal in terms of cache, code density, fusion, jumping, and many more without losing any of the benefits of compressed instructions (in fact, it increases the compression benefits).
Have to wonder if the business strategy here is to get Qualcomm to buy what remains for access to these patents thinking Qualcomm will then change their position over the macro op fusion, which would be a classic move by people detached from technical realities.
I still haven't yet figured out if anyone is actually switching any big platforms to RISC-V, or if everyone is just pretending to switch so they can get a free ARM license from ARM (which it seems you only get if you show a prototype of your stuff working on RISC-V and a few blog posts about how RISC-V is the future).
Yes, in embedded roles. Western Digital's hard driver controllers, NVidia's power controllers, stuff like that and probably a lot more too. Probably more that don't make the news, though this is mostly a thing for companies with the scale and silicon experience that customizing a core to their needs makes business sense.
They use a couple dozen Chinook of soft cores. These are supposedly small 64-bit cores, but ARM requires features they likely don't use and ARM requires payment for each of these cores. Switching these to RISC-V could save area and money without having any effect on their big cores.
SiFive claims nearly A78 performance in half the die area. In wearables, this represents a lot of savings in area and probably in power consumption too. This is another move Apple might make as ISA backward-compatibility isn't very important in that market while cost and power are.
That's Samsung 10nm LP vs something something 7nm. Not the best comparison, but one that would still imply some amount of area advantage for stock 7nm nodes.
A78 on SD845 used 64kb L1 vs 32kb L1 on P670 (a 32kb variant is available, but would get significantly worse performance relative to P670). Likewise, other things like the uop cache and 32-bit frontend add a lot to the size/power of A78 as ARM themselves bragged that their frontend got 75% smaller when they dropped the uop cache and 32-bit support in newer mid-core designs. I'd guess this frontend difference alone accounts for most of the area difference between the two.
They aren't claiming A78 performance, they're claiming A78 performance/frequency. Since there are microarchitectural tradeoffs between frequency and IPC in terms of things like how many FO4s to use in pipeline stages these claims aren't as significant as they could be.
If you have a super embedded system where your clock speed is tied to your memory frequency in a way where that's the limiting factor then performance/frequency can be a useful metic, but not in any system with caches.
Not sure who is that Patrick guy, but that is a horrible long piece of junk with literally no info in it. Other than the mention of AI, there is nothing concrete.
I don't know what is the motive, but clearly that Patrick guy is creating doubts and he seems to be pretty good at that. Maybe just another big mouth who wasted way too much time in some fancy corporate PR department?
Makes sense for a small enough company although this company sounds big enough for the president (COO?) to at least be differentiated. Then if there are investors you want a seperate chair?
Maybe I'm not the intended audience, maybe I'm not from a culture that understands posts like this, but this all seems so hollow and tone-deaf.
If I didn't otherwise know about the lay-offs it would to me paint an entirely misleading picture, it only hints at redundancies with euphemisms:
> need to realign our operations
> a need to refocus our priorities
> The resulting restructuring
And that's it, no other mention, not even an oblique one.
And yet the article is full of things like:
> SiFive today is in tremendous shape
> SiFive’s growth has never been stronger
It just comes across as complete bullshit? There's no sense of communicating difficulties, or any sense of humility. Just a lot of "Hoo hah aren't we amazing!".
> Maybe I'm not the intended audience, maybe I'm not from a culture that understands posts like this, but this all seems so hollow and tone-deaf.
This is written for other c-suite level people and board members, and sadly this hollow pointless communication is the norm for the c-suite and board members. They are not smart people, they don't use language to communicate important information, they don't make choices based on what is best for shareholders, employees, customers or society, they have no useful skill or insight. They are worse than politicians, they hurt shareholders, customers, employees and society. The only people they help are the politically connected Ivy League class, and they are a blight on society.
My impression, having seen many of these over the years, is that they are just that; hollow, tone-deaf, bullshit.
They're basically quotable material for the media. I don't think any ordinary person actually reads these and takes them sincerely; at least I hope not.
Yeah I don't understand who the target audience is for this. Who would read this and be reassured the company is in tremendous shape? It seems way too far removed from reality
No way a VC is going to be satisfied by something like this. And they're getting to look at the actual budget in any event, no need for vague platitudes about being funded. I'd suppose its for the remaining employees who need reassurance but don't get to see the numbers.
> There's no sense of communicating difficulties, or any sense of humility.
If there's any difficulties, that would be financial, no? Does SiFive publish any financial reports? And if so, do those show the company is in financial trouble? Are there well-informed insiders claiming so?
Unless that, I'll take the company CEO's word for it. Who writes:
"We are very well funded for years to come, with engaged, highly supportive investors, and we are realizing revenues and compounding royalties from many of the world’s leading semiconductor companies."
There. So if no money trouble, then what?
RISC-V ecosystem growth slowing down? No. SiFive's customers shopping elsewhere? Seems unlikely. Portfolio growing stale/uncompetitive? Hardly. Revenues not (yet?) matching expenses? Who knows. It's a young company still, in a rapidly evolving market.
More likely imho: SiFive transitioning from seed stage -> investment rounds & rapid growth -> some kind of 'maintenance mode'.
They have an IP portfolio now. So developing that is done. And decided to re-adjust the size of their workforce to better fit what's needed near-future. Leaving 'just' 100s of highly skilled people to work on SiFive's next products.
All in all, looks more like a storm in a glass of water to me. It's sad for the people who were laid off, but they'll have good opportunities elsewhere.
Or (their business) not growing as rapidly as they planned or expected and so they have fallen short on their own financial projections and need to adjust.
Need to be careful not to conflate SiFive with RISC-V (despite the best efforts of this blog post to do so). They have made their own decisions as to where to compete and that may or may not be the right strategy.
SiFive's success or otherwise has implications for RISC-V but RISC-V being a success doesn't imply SiFive will be.
This is what GPT-4 produced when asked to remove the bullshit:
SiFive Update - October 26, 2023
After years of growth, SiFive has reassessed its strategies and operations to focus on the most promising opportunities, leading to a restructuring. We believe this change is essential for our continued growth and industry leadership.
We are financially stable and continue to secure revenues and royalties from leading semiconductor companies. We are focused on maintaining our technological edge with investments in advanced R&D.
We recently launched the SiFive Performance P870 and SiFive Intelligence X390 products. Our commitment to our four product families: Essential, Intelligence, Performance, and Automotive remains strong, targeting sectors such as AI, ADAS, Electric Vehicles, Wearables, and Mobile.
With the increasing industry emphasis on AI and the need for adaptability in product design, SiFive aims to offer products that cater to these evolving needs.
SiFive has restructured to increase our agility, strengthen partnerships, and maintain our leadership in the RISC-V domain.
Half joking here but I wonder if what you posted is close to what Patrick originally scribbled down then ran through an AI prompt to blow it out. So like encryption, when you put the encrypted text through the cipher you get the original text back.
The Road Ahead discusses the progress and future of RISC-V and SiFive. RISC-V has grown rapidly from a basic ISA to powering high-performance applications. SiFive has also expanded significantly but recently restructured to focus on the most promising opportunities like AI. While this caused speculation, the CEO explains they remain well-funded and committed to delivering the best RISC-V products. As AI is integrated everywhere, RISC-V is becoming critical for its flexibility and ability to meet power and performance needs. SiFive will continue driving the RISC-V revolution across industries like mobile, automotive and datacenters. The CEO is confident SiFive will revolutionize computing in ways not yet imagined.
Layoffs followed by a round of hiring mean that a company thinks its engineers are making too much money and can be replaced by cheaper hires with the same level of skill.
This is one of many reasons why investment capitalism is a really poor method of technological development - a better option would be to reduce the control of external investors and adopt a more democratic internal hiring and promotion policy. Unfortunately four decades of neoliberal government and corporate policy has concentrated all the capital in a few hands, and the outsourcing of much of the hardware manufacturing base overseas has resulted in a decrepit technological base.
Thus, I imagine Chinese development of RISC-V is well ahead of whatever SiFive is up to and if they were directly competing, they'd be well behind the curve:
Counterpoint: the last four decades have seen some of the biggest explosions in technological progress in history. And based on what we've seen, not on anyone's imagination.
Who is to say that the rate of progress wouldn't have been doubled if the constraints of excessive intellectual property laws (a key component of the neoliberal outsourcing program) had been drastically limited? Indeed any analysis of the Chinese technological boom will conclude that their more laissez-faire attitude towards IP played a large role in that outcome.
A similar argument is that the open-source movement has become the main driver of software innovation (Linux in the server space etc.) and that without it we'd still be back in the early 2000s tech-wise.
> Indeed any analysis of the Chinese technological boom will conclude that their more laissez-faire attitude towards IP played a large role in that outcome.
That works for catching up; not for moving ahead.
> A similar argument is that the open-source movement has become the main driver of software innovation (Linux in the server space etc.) and that without it we'd still be back in the early 2000s tech-wise.
China also uses Capital to invest. They have one of the largest capital markets in the world. I think what you do not like is not capitalism but free markets or the extent of freedom of the agents in those markets. But no matter what capital is the best way to industrialize a nation and this has been proven in Japan, Korea and China.
>the outsourcing of much of the hardware manufacturing base overseas has resulted in a decrepit technological base.
This is nonsense. All this shows to me is that you are not aware of the current semiconductor supply chain of which the US is a large member. And I suggest you find out how democratic the "hiring and promotion policy" is in Korea, or how concentrated the wealth is there.
America closed Radio Shack stores that used to provide hobbyists with all kinds of electronic components (before it tried to become a little version of Best Buy). In contrast, China has Shenzhen. Interestingly they seem to be engaged in state-sponsored VC-like activities:
I'm not sure what point you're making. Driving the current state of the art in semiconductors requires something like a PhD in physics or materials science. There are certainly many PhDs in China (and being educated is more respected there than here) but Radio Shack and Shenzhen pcbs have little to do with this. They aren't growing the state of the art. Have you looked into the "democratic internal hiring and promotion policy" of those Shenzhen factories by the way? Shenzhen is as far from communism as you can get. Marx described workers being "a slave of the machine" as an inherent quality of Capitalism: and he is right (until these things can be automated).
I have no doubt that Chinese semiconductor output will increase in quantity and quality. I also have no doubt that it will be sold on the global market for capital and that capital will be the primary way the owners and workers of those companies are rewarded as well as the means used to create and grow the company.
Also, Radio Shack closed down because its customers moved online – that is simple supply and demand.
>SiFive today is in tremendous shape. [...] We are very well funded for years to come
Ah, so they did it for more money, not to save the company. Good to know. I feel they could have written a few more hollow words to their ex-coworkers though
But even that rings hollow. in this job market, you keep tech people and just reassign them. Hiring people later to execute on the new vision is hard and expensive.
Or maybe they're just not planning on growing, which is a brave strategy in the tech world.
Doubtful. In my experience this happens when companies are not in tremendous shape. The most likely sequences of events is:
1. Revenues are not as high as needed. If the current trajectory continues they will run out of money.
2. Therefore they need to change the plan. Changing the plan involves firing a load of people.
3. Everyone assumes (correctly) that that is because they are not doing so well.
4. Post a reassuring blog post that everything is fine. SiFive depends on customer confidence.
They probably are well funded for years (2?) to come. And they may be in "tremendous shape" *assuming that their new plan goes as they hope. It does seem a little desperate though. No information on how their new plan is different at all.
Every person you hire to help drive a period of growth is someone who will later be fired once that growth (or commonly just funding) runs out. It's very easy to just increase headcount to increase growth but it's not sustainable.
I feel like companies often think they're always going up an exponential curve rather than breaching the top of the 'S' and you really need to plan for the long term health of your company. This company obviously hasn't
This is only true if you are measuring growth in extremely dumb ways. Most metrics for growth are revenue-based and if the growth is creating revenue there is nearly always enough money to pay for the roles created to grow the company. If the growth project fails sure you'll need layoffs but you don't need to fire a $300k developer who creates $1M/year in revenue.
I would still love to hear SiFive say what business they're in. From the context they're licensing cores, but making a SoC from those cores is still up to the customer. Still... would be nice to get that confirmed by the company themselves.
"We're doing great! Loads of money and everything is wonderful! Some of you are asking why we fired loads of people and that was a hard decision to make. Everything is getting even better from here!".
So... it seems the effect of the post is to further piss off the departed and give the general impression the CEO is incompetent.
Any guesses what they wanted to achieve with this piece of marketing?
>> Any guesses what they wanted to achieve with this piece of marketing?
Sure. They're hoping to sell the company. To whom I'm not sure. It wasn't an appeal to any specific customers, but a lot about how great the company is and what a great position they're in. So that must be what they're selling - the company.
Maybe they had an important announcement to make somewhere hidden in paragraph 3/4/5, but it got edited out because it was too contentful, just leaving PR fluff?
I would suggest that coverage here is somewhat biased. Today, SiFive basically owns the market for high-performance RISC-V cores, and that is the market that really interests HN. That market is also really small.
On the low end, there is a vibrant ecosystem of chips and IP providers, and SiFive is actually increasingly uncompetitive. This seems to be where the money really is, but it doesn't make SiFive excited. This market doesn't get cited on HN.
You are looking at the relationship from the wrong end.
Everything SiFive does is about RISC-V. Everything that's important about RISC-V is not done by SiFive. (And if Qualcomm wins the current pissing contest for the future of the ISA, nothing that's important about RISC-V will be about SiFive, and that's fine. Qualcomm's ideas for the future are certainly better.)
It’s makes sense to me, showing you can cheerfully fire people shows you aren’t making decisions based on likability. If you were invested in this company wouldn’t you want to know the CEO values you more than his own employees? Makes perfect sense to me.
First everyone over-hired during covid since everyone else was, then everyone started laying them off after since they never knew why they hired them to begin with and everyone else is laying off anyway
It's also the case VCs don't give me to extend the runway. They want their money to be immediately deployed in building new features, larger headcount and so on. Startups just followed the script.
vanguard and blackrock do prefer to keep only their big tech? (apple/msft/google/etc)
Or they did come up with the realization that this Big Tech is toxic for humanity and do invest in stuff in attempt to clean up the mess, for instance trying to make RISC-V a success, that on the worldwide non-free modern ISA issue?
Large chunks of Vanguard and Blackrock are in index funds. So no amount of realisation about toxicity or heartfelt revelations on the part of Vanguard and Blackrock are going to change what those chunks are invested in.
They can always create a new index and a fund that follows it. But getting people to buy into such a fund is much harder then a fund following some well known index.
And new indices get created all the time, plus funds that follow those indices.
You often even get new indices specifically created to be easy to invest in (instead of optimizing for the older goal of being some measure of overall market performance.)
Vanguard and Blackrock don't necessarily dislike that their hands are bound with the indices. That avoids lots of thorny political problems.
Clearly something has, but this provides no real clues. Completely counterproductive. But all is good as SiFive will revolutionize computing in ways we have not yet begun to imagine!