48 cores

Linux and Unix has that, accessible to the general user.

?-)

there have been several,

Tyrrell P34, double fronts, Williams FW08B,Williams FW07D,March 2-4-0, double rears

-Lasse

-Lasse

Tyrrell P34 and Williams FW08B, perhaps even more.

If the application does not need multi-core, don't use it !

A few real CPUs or even hyperthreading is nice to lock a badly behaving program (such as Firefox/Adobe Flash) into a single processor (Windows Task Manager/Set Affinity) and let the other processor(s) do some useful work.

Why limit to 2 cores ?

For MPEG a 16x16 macroblock could be handled with a single processor. Doing motion compensation would need more processing power to detect from where the pixel group came from.

I was thinking about fast, direct, program-code-level ability to read and write realtime data registers. That was easy on a PC/XT or an Apple II. It was nasty but do-able over the PCI bus. It's really nasty over PCIe. The PC is now pretty much inaccessable for folks to make their own interfaces. Even making a PCIe card with an FPGA is a major project, and latency will suck.

--

John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    

Precision electronic instrumentation

The original Unix system had just the timesharing "nice" feature setting the quantum size.

Linux also has a usable real time priority settings.

Windows also has a 32 level priority system, of which priority levels

16-31 (real time priority class) are strictly fixed priorities (a task runs to completition before waiting for some external event).

Of course, anyone working with strictly priority based systems must be extremely careful not to spend too much time at higher priority levels (even due to error conditions, such as the partner does not respond).

The real issue with hard priority based systems is which task priority you can _lower_ without affecting the system total throughput, not which task priority should be increased :-)

Except for the SSE instructions (used less than 2% of the time) the X86 is purely SISD.

Multi-core is only that, multiple SISD machines.

?-)

I don't know why you are even talking about using the x86 line for embedded stuff that requires the sort of bit banging I/O that you seem to be talking about. That would be like driving a bus in the Le Mans auto race. If you need bit banging I/O you would use an embedded controller, not an Intel x86.

Again, they have their niche just as every processor does these days. Why use an ARM11 when a CM0 will do the job. You make it sound like MCUs are one size fits all. You only know about your needs and seem to think that is what the world revolves around.

The past, present and future of embedded processors is the low end where the chips are disposable and literally billions are produced each year. Intel and Xilinx have a small piece of that market.

As an analogy, think biomass. There is Intel, etc roaming the earth as mammals, the ARM11s, MIPs and similar devices as the much larger insect population and all of that is outweighed by the tiny, invisible, micro power micros acting as the bacteria in the embedded world.

--

Rick

But, once you do the indexing, the fastest way to do a BIG text search will be bottlenecked at the disk interface(s); autonomous disks with search instructions in their firmware DO exist.

On a sunny day (Sat, 7 Jun 2014 13:36:45 -0700 (PDT)) it happened Lasse Langwadt Christensen wrote in :

You too :-)

Thanks.

On a sunny day (Sat, 07 Jun 2014 23:57:55 +0300) it happened snipped-for-privacy@downunder.com wrote in :

That is true, although 'I have bad software' is not a great argument to buy an X core. Better fix the soft (or get rid of it if it is not open source).

That reminds me of a discussion here I overheard and I told about long ago: I was in the computah shop, guy comes in, complains that his PC is too slow for email, sales druid tries to sell him the latest GHz model. Probably the poor guys PC was sending spam all the time or in other ways infected.

Creating bloat by for example companies like Microsoft definitely helps companies like Intel and Amd sell more hardware. The secret that all that same stuff can be (and has been) done with a Z80 at 4 MHz is kept from the public.

Maybe Raspberry broke some ground there, doing HD video.

Yes, true, but that can be done with dedicated hardware too. Often no need for playing whatshisname Turing.

That remains me about the Catt spiral patented by Ivor Catt in 1972, in which the whole wafer was used to hold a large number of cells containing a lot of memory and a primitive processor in each. The patent described ways to circumvent bad areas on the wafer. This WSI was apparently intended to be a big semiconductor memory with some local processing capacity.

With 48 processors with 16 MiB cache each, we are nearly in the same situation.

I'm not, because Intel has made it impossible.

That would be like driving a bus in the Le Mans

Something like an Atom-based miniITX would be a good platform for mid-scale apps, but there's that interface problem.

Don't be obnoxious. We use chip processors from AVRs to LPC1768s to LPC3250s to dual-core-ARM XYNQs. We've done several embedded things with purchased motherboards. We use multicore PPCs and x86 in cPCI crates and x86 SBCs in VME crates. We recently did two VME crate controllers, Ethernet and USB and cabled PCIe interfaces, with LPC3250s inside for supervision.

It's just that the "PC" is now a closed system with very complex, high-latency interfaces that need complex drivers. Intel has gone into and out of the embedded market many times, and now seemed out for good, just as the Internet of Things is hot.

Lately I like the microZed as a board-level processor plugin.

I think of Intel as a dinosaur, threated with extinction by small mammals that eat its eggs.

--

John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    

Precision electronic instrumentation

Yet they're still trying to sell their stuff to the embedded market. ...complete with Intel margins thrown in.

No so. They're still trying. The last time they were in, it was hard keeping a straight face when they gave the price. Well, I did tell them that they were talking about taking all of our margins, too. Good lunch, though.

Not very exciting anymore but I'd take their bank account. It's not clear that mammals will evolve anymore, either, but it's a lot better being at the top of the food chain.

...snip...

Anyone who would compare a PC to the "Internet of things" clearly does not have a clue about the issues. Your bias for the highly complex and expensive Zync is also misdirected for IoT. But when you only produce a handful of products a year I guess you can afford to be inefficient.

Yes, I am sure you can shoehorn that into any application you want.

Lol. However there are still plenty of market for Intel boxes and all those servers which is exactly what Intel is optimized for. I guess what you can't seem to understand is that the markets are specialized. There is *no* one size fits all solution or you wouldn't keep going on about how Intel doesn't do embedded anymore. DUH!

--

Rick

Looks like your desktop has already given up on 8750 processes.

Jamie

We try to optimize. It makes sense to spend $200 on a plugin board that has DRAM, Flash, USB, Ethernet, SDcard, power supplies, Linux, and a tightly coupled FPGA, for a product that sells in modest volume for $6K each. We'd be nuts to make all that from parts. But Intel doesn't make chips with real busses any more. Their Arduino has to re-create parallel port bits from PCIe or something.

When it makes sense. We use $4 NXP ARMs when they make sense.

Aren't there TV sets with FPGAs inside? Makes sense.

Multicore ARMs may well eat the Intel server market, and tablets are killing laptops and big-box PCs. Intel is still, basically, running the 8008 instruction set, with AMDs extentions, and everything else that they tried has crashed hard; Iapx32, their RISC thing, their dabbling in ARM, and the gigabuck Itanic fiasco.

Apparently Intel is trying to do the ARM thing, licensing x86 cores to go into other peoples' custom SoCs. Sounds crazy to me.

--

John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    

Precision electronic instrumentation

What? You often make no sense to anyone but yourself. Why would they put an FPGA into a TV? There is certainly little reason to include a Zync chip. Much cheaper to add an ARM CPU and a memory chip along with the tuner and video chips.

Lol. Yes tablets are the up and coming thing, but laptops are hardly dying. ARMs in servers are a long way out still. Intel is not as stupid as other, smaller companies. They don't need to work in niche markets.

Yes, I'm sure you could do very well advising the rest of the world how to make money.

--

Rick

Much cheaper to add an ARM CPU and a memory chip along with

Your hostility is blocking your ability to think. That's bad engineering.

Ditto:

--

John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    

Precision electronic instrumentation

The very first words on the page are, "High-end digital televisions at the forefront of the display industry". So they are talking about the Lamborgini's of TVs, not the ones you see in Walmart and Costco... in other words, irrelevant.

Sorry if I am coming across as hostile. I just find your comments to be very silly and worse are the ways you keep trying to prove the points.

Every one of these support my statement.

"Canonical and Applied Micro demonstrated IceHouse Production OpenStack deployment in preparation for Computex next week". They made it to a trade show, clearly everything we see at a trade show is big time!

"ARM server debuts", that is not mainstream, that is getting your toe in the door.

"Microsoft has joined a new project to accelerate the development of ARM-based servers, suggesting ARM versions of products like Windows Server and Hyper-V could be in the works." How many years before this reaches the market?

Clearly ARMs in servers are a way off. Yeah, there might be a few out there, but I bet there are still more Itaniums, lol.

--

Rick