Slashdot Mirror
The Amazing Integrated Microprocessor
An anonymous reader wrote in to say that "SiByte Inc.
has
announced the
Mercurian
processor SB-1250, a single chip
containing two MIPS64 CPU
capable of running up to 1GHz, plus a 512kByte L2 Cache,
PCI, gigabit ethernet (3) and serial (2) interfaces.
The whole chip consumes
a mere 10 watts with both processors and all interfaces
running at full speed. While it is targeted at networking and communications systems,
wouldn't you like to have a notebook allowing to carry with you
that much processing
power? The chip will be able in volume mid next year, and the data
sheet states, they will provide support for porting NetBSD
and Linux to it." Sure it's vapor, but it sounds pretty impressive.
Despite the availablility of multiprocessor Mips32 and Mips64 systems, there is currently no support for SMP in the Mips kernel tree.
This is becase where the ia32 archetecture has a common and well-documented SMP method, APIC, Mips does not. Each vendor seems to do their own proprietary thing.
This would definately be cool but I'll believe it when i see it.
This is just like television, only you can see much further.
My 175MHz R10000 (1MB L2 cache) manages to crunch a SETI@Home packet in only 9 hours.
My celeron 450 will crunch a seti packet in about 8 1/2 hrs, so they are roughly equivalent with this test even though the mips is running at less than 1/2 the clock speed. I would certainly like to see SGIs with 1GHz chips. I certainly couldn't afford to buy one, but I can always dream.
the good ground has been paved over by suicidal maniacs
Yep, these "targets" have been making me chuckle for years. The 386 was targeted for servers. All I could think of at the time was that I wanted one to replace the 8086 on my desk at work. Fuck servers! I wanna compile/link faster!
Is there any reason someone wouldn't want to use this new processor for other miscellaneous purposes? Typically, only downside to these "fringe" processors is that they don't run legacy x86 stuff, but if you run an OS like Linux or *BSD where you have the source to almost everything anyway, then binary compatability doesn't matter nearly as much, as long as you have a compiler. Once you get away from the need for binary compatability, you have the luxury of getting to focus on performance issues, so the marketroids' claims about the "target" may be completely irrelevant.
---
As copyright owner of this comment, I authorize everyone to defeat any technological measure which limits access to it.
When IBM introduced the IBM PC/AT in 1984 they were very clear in their press materials that a computer that powerful was meant for server use and that nobody would need the power of the mighty 80286 on the desktop.
Maybe it's time to start porting OpenGL to IOS :]
www.zflinux.com describes a 386 class system on a chip, including all interfaces
But, in my opinion, the company is poorly run. All my inquiries about their products have been completely unanswered. I would definitely not purchase from them unless they made an extreme turnaround.
If tits were wings it'd be flying around.
heheheheh...
? shoot NORTH
you missed
? shoot NE
hit!
? use BFG
you are now using the Big F-ing Gun
? shoot NE
you have killed the llamas.
Indeed, and this is due to a funky little trick kind of pulled from VMS and OS/400 (neither of which ever really got ported) called the HAL or "Hardware Abstraction Layer".
:-)
The theory is that the OS does not handle any calls to hardware services itself at - it asks the HAL. The HAL in NT4 was only a few Mb of compiled code that picks up this call, interprets it for the "local" architecture, and then passes it on down. This means that NT can be ported by re-wrting just the HAL itself. Or at least in theory. I'm not an MS code-monkey, but my friends who are tell me there are other "areas" that need to b re-written for the OS to be ported, but even then it shouldn't be more than about 20Mb of compiled code.
In actual fact, it's the one aspect of NT that I really like. There are times I wish a free OS would implement such a system (I think NetBSD uses a similar principle in that the amount of core code to port is reduced), because then we could basically get a bunch of free OSes running on loads of hardware. FreeBSD on a Psion 5 would rock my world to say the least. OpenBSD on a Palm V. Linux on a Sun Enterprise 10000. NetBSD on an AS/400. You get the idea.
Anyway, as I was saying. I always thought that the HAL would slow things down massively, but apparently the performance hit is minimal. I'm sure somebody else will have more info on this.
Guess what will be next cisco aquisition.
Cisco IOS has run on MIPS32 for ages, I see no problems for them making it run on MIPS64.
This just looks like the next CPU for a cisco line card. If foundry, intel, juniper or any of the other sharks in the pond will not eat it first.
That is if this is not vapourware.
Baker's Law: Misery no longer loves company. Nowadays it insists on it
http://www.sigsegv.cx/
Reminds me of endless blurbs in PC Magazine about what things are "targetted for..."
----
Wouldn't that mean 'single point of failure' too?
If the processor goes, everything goes.
Bus speed runs at half of the speed of the processor, so it's 500 Mhz. (The key to Mercurian's unique architecture is its intelligent, high-performance MP design built around a fast, on-chip internal bus called the ZBbusTM. The ZBbus, which runs at half the CPU core clock with a data width of 256 bits (one cache line), connects all the major blocks of the processor including the CPU cores, cache memory, and I/O.)
The chips work in conjunction with each other, not against each other or in a master/slave relationship. That's got to help matters as well.
The only thing I can squawk about is the L2 cache being shared by both processors. I suppose this means you can claim 512 L2 cache for each processor, but what if they are both working in completely different memory spaces; would that effectively drop to 256k cache? Does it matter?
And the PCI bus looks to be nice; claiming 400 Mhz clock with 6.4Gb/second; looks to be utilizing a trick similar to AMDs 200 Mhz FSB.
This chip just goes to show you what you can or could do by leaving the x86 architecture behind. (Backward compatibility, backward compatibility, backwar Shuddup!)
Please pass the FUD.. :-)
-- Talonius
Off-topic: What would the world be like without a past and legacies? Where would we be? What dreams and myths would we come up with, in today's day and age? Something fun to think on.
My reality check bounced.
Check out http://www.eet.com/story/OEG20001009S0026
"German startup Pact GmbH will attempt to leapfrog the growing field of highly parallel processors targeting communications when it rolls out a complex 30-million-transistor CPU that integrates 128 thirty-two-bit processors at this week's Microprocessor Forum."
Tastes Like Chicken
________________________________________
Napster-to-go says "Fill and refill your compatible MP3 player", which is a lie. It's not MP3. It's WMA with DRM.
not sure this is the same company, but someone in the silicon valley had a ultra low power MIPS running at 1GHZ in their lab some months ago, designed specifically for switches/routers.
i guess it's okay to dream about logging onto procewatch and seeing quad consumer grade cpu ATX mobos for this beast someday, even if the possibility is remote.
Treatment, not tyranny. End the drug war and free our American POWs.
See my user info for links.
Of course this assumes that everything goes well. High frequency/low power stuff at Very Deep Sub Micron (VDSM) can run into a lot of problems during tapeout (crosstalk, timing, DRC violations, buggs in the back end tools). Still, that core is sweet.
Is not the most insightful comment I've ever seen. Vapourware always looks impressive.
"Encyclopedia" is to "Wikipedia" what "Library" is to "Some people at a bus stop"
Then multi-threaded apps actually have less overhead. The processor caches are already synched up.
If course whether this is good or bad depends on the problem.
- Intel IA32 (x86)
- Intel IA64 (Itanium)
- Clipper
- MIPS
- Alpha
- PowerPC
And possibly others in the lab that weren't even as commercially viable as the Clipper. Sounds portable to me. Of course, nobody seems to have wanted to actually buy the non-Intel architectures but that's another issue.