News European Commission offers funding for CPU efficiency projects

So with ARM being bought by a Japanese company and us leaving the EU does that rule them out from something like this, if so it seems a shame as ARM (afaik) are fairly good when it comes to CPU efficiency.

 

I was thinking the same thing. Pretty stupid they decided to do this now. They also seem to be pretty clueless if they think someone can make a CPU that has better performance-per-watt than anything ARM, Intel, AMD, IBM, Nvidia, and MIPS can, have, or will make. Don't forget - it needs to be cost effective to mass-produce too. Sure, Intel is kind of slacking lately and we don't yet know what AMD will be releasing for servers, but for the most part it seems we have reached peak efficiency for silicon chips.

 

Not sure how that quote relates. I'll accept performance and efficiency improvements with open arms, but the only architecture I've noticed that made any real-world progress is ARM, and in doing so they're slowly selling off their RISC soul by becoming CISC.

Yes, what about those architectures? If they are deserving to win, why haven't they already?

RISC-V is, to my understanding, kind of an in-between approach of MIPS and FPGA. It's a great architecture, but I don't see it being all that useful in the average server/mainframe.

The Epiphany-V seems innovative but I can't really tell if it's any better than a GPU. Epiphany may have a more refined design, and it may also be more efficient since a GPU requires a CPU to tell it what to do (in other words, avoidable overhead). But in a business perspective, there are too many downsides. The roadmap for it isn't clear or promising, and it would take more effort to get desired software to run on it (compared to GPUs) and, more importantly, take advantage of its cores. There doesn't seem to be much of an upgrade path either.

Fujitsu's architecture also seems great, but nothing is stated about energy efficiency. Regardless, it doesn't seem to be easy to properly utilize. I have a hard time believing you can just simply re-compile your program for this architecture and suddenly see a 10000x performance increase. I have no doubt it could blow any other CPU out of the water, but it isn't realistically possible for companies to re-write whatever software they use to take advantage of the performance of this in a way where they can maximize it's potential.


So all that being said, yes, of course there are ways to get better performance out of a silicon CPU. But if just simply getting the best possible performance-per-watt was the only thing that ever mattered then someone could just get the funding for developing an ASIC. But the requirement was "specifically tailored for high-performance, low-power server-side applications" so if you have an architecture that is really picky about how you use it, I don't see that as being a winner and clearly the EC doesn't either (unless they're just ignorant).

 

If I were so blind then why has innovation for current PCs and servers stifled so much? Intel has real competition, maybe not against AMD, but elsewhere. They have more than enough money to create something competitive, they focus on more than just x86, and yet they still aren't keeping up. Intel makes for a sore loser, so the only realistic reason for them to give up in something (such as the smartphone market) is because they can't make something good enough.

ARM themselves doubted their own success. They succeeded simply because they offered something nobody else did - a cheap and energy efficient design that was very capable but simple enough be easy to develop for. The architectures you mentioned do not wholly fit that description.

Unless you didn't mention it in the article, then no, they did not in fact win. Not this funding from the EC anyway, which is what this is all about.

Seems you misinterpreted my statement as much as you think I misunderstood how it works. I explicitly said I have a hard time believing you can recompile a program for it. The fact that you actively pointed out that it isn't a general-purpose processor is exactly why it isn't a contender for what the EC is looking for, which was my point...

Irrelevant comparison. Back then, people didn't have a choice. There were no alternatives. The "boom" moments you mentioned were what made x86 a CISC and capable enough of being a general-purpose CPU. But as you can tell, x86 has become woefully complex, and the only way for architectures like ARM to stay relevant with modern software is to do many of the very same "boom" moments. It takes years to accomplish all of that, even with an already established architecture. Architectures such as the ones you mentioned are relatively primitive (with RISC-V being intentionally primitive). They may be great at what they do, but in a business perspective, they're not good enough in terms of time, cost, and capability or else the EC would have already funded them. If these architectures are to see success, they need such attention.


I'm aware what MIPS and FPGAs are, but I understand your confusion; I should've elaborated: I get the impression RISC-V is more complex and capable than FPGAs, but less so than MIPS. In other words, it seems you could run a modern OS on RISC-V, but it doesn't seem like it could be capable of being used as a PC or a general-purpose server. Not that it was meant to, but again, we're talking about what the EC is expecting.

 

Worlds fastest supercomputer means nothing by itself. Nvidia created the world's fastest supercomputer at one point too, but I think we can all agree they don't have the world's best architecture, even at the time they held that title. When it comes to supercomputers, money is the limit. I haven't found enough info on the SW26010 to know how much better each core or logical thread is compared to all competitors, so I wouldn't be so quick to say current general-purpose server-capable CPU technology isn't stifled.

Again, getting a little presumptuous here. At least according to the article you linked to, we don't know what the wattage is. Nothing clearly states it was designed with being "low-power" in mind, which was a separate requirement from PPW. But at this point this is getting into petty semantics.
Anyway to my knowledge, Fujitsu has European factories and engineers. I'm sure they could exploit some loophole using their European HQ if they really wanted to; many companies do such a thing to avoid stuff like taxes.

Has someone peed in your coffee this morning? What could have been an intelligent debate is turning into "you're stupid and so are your thoughts".
What I stated is based on what I know has currently been accomplished with RISC-V. That being said, if you continue reading that Wikipedia article, you'll find that support for the OSes you mentioned are very recently provisional. It requires a simulator to run a web browser... Not exactly something I would call an efficient approach.
Obviously, it's capable of being a PC or general-purpose server, but it isn't going to be good at it. For a 6-year-old project it sure hasn't accomplished much in being a PC or server, probably because it's better off focusing on specific applications where it will actually flourish.

I'm not trashing the architecture, but it seems to be more of a proof-of-concept than anything with a prosperous future.

 

Hmm> yeah, I'm not going to notice a link to a PDF when you added it after you already replied...
Again - I'm not denying it's efficient. Obviously, it isn't drawing 10,000x the power of a competing CPU, but 10,000x the performance is clearly also a best-case scenario. In some scenarios, it wouldn't surprise me if it performed worse than a 10-core Xeon. When AMD says Zen will be 40% faster than Excavator, those are probably cherry-picked results. Any performance claims that come directly from a manufacturer are to be taken with a grain of salt.

Remember - if it were really as good as they claimed for just about anything you throw at it, everyone (obviously in the server world) would be flocking to them.

Sure are getting nit-picky here... The end result is literally no different. As far as we're concerned, a web browser has not been successfully run on RISC-V without a simulator. It doesn't matter if the browser itself is in a simulator or the entire environment, the point is a simulator was needed.

A deep and complete understanding of FPGA is irrelevant, which is why I'm ignoring the discussion about them. I know what FPGAs are but you're choosing to look at things in a different perspective. I used FPGA as an example of a piece of hardware that is relatively simple compared to MIPS. Sure, you can make it emulate another architecture, but that tends to defeat the purpose of using them. When you have an FPGA act like an x86 CPU, it's probably not going to do a very good job or else everyone would be buying them. So pointing that out that capability is irrelevant.
Ironic in a way, how you're doing the exact opposite of me. Obviously, I'm going to focus on the goalposts, because that's all that matters - the reason you replied in the first place is because you don't like where I put them. But you would rather focus on the little side things I said along the way when they have no impact on the big picture. What you can do with FPGA has nothing to do with this, but you would rather use that against me simply because you can.

Give me 3 real-world examples where you gave an antagonistic reply and someone willingly admitted they were wrong. Keep in mind, I'm not questioning whether your phrasing was warranted or not (I don't blame you for the way your first response turned out).
You are your own problem. Most people by now would see me as a troll and ignore me. It is usually in the interest of article publishers to not intervene in comments unless to point out a correction against a proposed fact. I made an opinion, albeit, a needlessly negative one without the foresight to be more specific about "general purpose CPUs". I will also admit after your first reply that I misread one key fact about the article, which was that it was exclusive to European designers only.

You should ask yourself what you personally gain when you get involved in stuff like this.

 

If it can't ever run just any calculation you want, it doesn't have enough of a wide-spread use to be of interest to a wide server market. But for argument's sake if you could make it run any program you wanted, in some cases its performance wouldn't be all that impressive. I get the impression the EC wants something that is more than a glorified ASIC. Maybe I'm wrong, in which case, discussing this doesn't matter.

Doesn't matter if you can buy them yet. There are already results of it's performance, so it's already a proven product. So, why haven't they attracted enough interest for potential buyers? Maybe they have, but since you like proving me wrong so much, I'll let you do the research.

Alright, that I will "happily admit" I badly misinterpreted. Feel proud yet?

I don't have the time or care to thoroughly study these things; that's actually your job. However, in my own research, I have yet to find any GUI tools that run on RISC-V. As for web browsers, I'm sure it can run Lynx, but that hardly qualifies as one.
Again, not denying that RISC-V can run a GUI.

That doesn't disagree with anything I said. That doesn't prove I know nothing about FPGAs. I'm aware of everything you said. But as usual, you keep focusing in a different direction. I don't know how many times I need to explain that I'm not talking about what an FPGA can do, I'm strictly referring to it on a hardware level. Yes, of course some FPGAs are meant to behave like another CPU, but generally they aren't meant to be production-level and mass-produced as that CPU. Otherwise, companies like Intel would only produce FPGAs that are pre-programmed with an x86 design.
This is why discussing this isn't relevant - it's completely besides the point.

Based on your perspective of my initial response, I absolutely was wrong. I failed to be more specific. So, I specified, and here we are now.

Not going to happen. Admitting I'm wrong about that would mean you are also wrong, because I don't disagree.

lol yeah ok, keep telling yourself that. Your next sentence shows frustration. Your previous posts showed frustration. You've dropped your professional attitude and are risking your public integrity to put up a petty argument with someone who you have admitted is repeatedly stubborn. If that sounds like more fun than going to a pub, you've got some bad priorities, and frankly, is sad. I know I'll be enjoying my time away from the computer when going out tonight.

My phrasing (where it matters) does not dictate fact. For someone as picky about wording as you (not an insult, just an observation), you should realize this. In the places where I am stating an incorrect "fact", I have already admitted being wrong.

Mind explaining how I would consider digging into a pit of ignorance a gain?
Anyway, I'm not gaining anything - I'm not trying to and never implied I was. This just helps pass the time while I'm bored at work. Meanwhile, your wife is taking care of your kid for you while you're keeping me company. I might not be gaining anything, but seems to me you're more at a loss. But if you insist this is fun, I unfortunately have to break it to you that in a couple hours I'll be putting this on hold once I get out of work.

 

OpenCL and CUDA make for relatively easy programming. Sure, GPUs might not run anything you want, but they'll run anything that they can actually take advantage of.

What you wrote doesn't address what I said.

I never said a GUI was important to a server, I was just using it as an example of something it currently doesn't do. I still haven't found any evidence of an environment like XFCE or openbox running on RISC-V. Just because you can install it, that doesn't mean it's functional.

I think that says more about your intelligence than mine.

Yes... and? That doesn't conclude anything.

Mhm ok. Whatever you say, Mr. Ace.
I never said I disagreed with you (regarding FPGAs). You're just ranting about something I wasn't talking about. It feels like I'm just standing here with you yelling "water is wet! BTW, rain falls from the sky!" meanwhile, I'm talking about water at the molecular level.

Alright, if you insist. Am I supposed to care?

Yeah... I don't think that's how it works when you post using the same account. An employer might not care what you do on your own time, but if you clock out and stay at your desk while watching porn, I'm sure said employer isn't going to be too happy. Sure, that's an extreme case, but the point is it's an employee being a poor representative of his organization.

Doesn't make it any less sad, you're just diverting the attention away from the elephant in the room.

Well, you should have first realized it was kind of a rhetorical question. Do you honestly think I care about the answer? If you were enjoying yourself as much as you claim, you wouldn't be so determined to prove me wrong. The fun of arguing with someone is watching them squirm and break their character, which seemed to happen to you pretty early on.
What I personally enjoy is the fact that I know you really dislike me, but I don't dislike you. I'm feeling a little disappointed you had to show your true colors like this, but I do respect you.

Don't worry you've got plenty of time to work on that, it's just going to be delayed for a little over a day. I'll be back eventually!

 

*DING* 1st round bell.

 

This thread -

 

Especially when you're having fun.