Posted on 12/23/2004 8:21:56 PM PST by antiRepublicrat
In 2005, IBM plans to bring a significant feature from higher-end servers to the next generation of its PowerPC 970 processor line used in Apple Computer machines and Big Blue's own blade servers.
The next-generation chip will have technology that lets it run multiple operating systems simultaneously, said Karl Freund, vice president of IBM eServer pSeries. Doing so allows a computer to handle more jobs at the same time and to be used more efficiently.
The technology, called partitioning, relies on a concept called virtualization that breaks the hard link between an operating system and the underlying hardware. Partitioning is available today only on servers using IBM's higher-end Power4 and Power5 processors and in competing server designs from Sun Microsystems, Hewlett-Packard and Intel.
"The goal is to make virtualization capability ubiquitous across the Power line," Freund said in a Tuesday interview. "We want to drive it down to lower price points and make it available on products like BladeCenter as well."
(Excerpt) Read more at news.com.com ...
Next technology for Macintosh PING!
If you want to be 0 or 1 on the digital Mac Ping list, Freepmail me.
Actually they have been with the better chip company from the get go. They started out with the Motorola 68/6900 series. That was a screamer for it's time. IIRC, they've been RISC ever since. Would have been good if Apple worked with DEC when it first introduced the Alpha, but Bob Palmer CEO and chief fool at DEC wanted to get a 64 bit version of windows server. VMS was converted in a just a few months, but BP kept waiting for Gates to get the 64 bit version of Windows (NT5.0?) for 2+years. Gates put Palmer and DEC out of business.
I am watching these chips being tested right now.
"Gates put Palmer and DEC out of business."
Gosh, I'm sure he didn't mean to. [laughter]
"technology that lets it run multiple operating systems simultaneously"
Obviously something that will come in handy around the house here... ;')
At the plant where we made the Alpha chip (where palmer was plant manager before being CEO) we kept hearing NT5.0 was just a few months away. Sadly palmer knew squat about software and he kept believing the BS. As he had been plant manager and I was a principle software engineer there, I can say, he was totally clueless about software. Hell the CIM manager was clueless.
The freaky thing is LPAR on the desktop. It's something I never thought I'd see outside of big iron.
LPAR or dual-core? I'd love to ask a lot, but I'd bet you can't say anything.
I had a 68K in an Atari ST, very nice chip. But as the turn of the millennium rolled around we found that Motorola was no longer able or willing to supply decent chips for Apple. However, since IBM's fortunes rest in the higher-end chip business (not just embedded), Apple should have a supply of great cutting-edge chips for years to come.
Dual core.
One thing I can say is that at $97 IBM is going to look like a bargain a year from now.
Less shiny wrapping and more meat please.
Funny! And true! And pathetic.
NT, which was billed as highly portable, was, in fact, portable only in thoery because it costs Microsoft $20MILLION to QA a release candidate for Windows.
From the interviews I've read with POWER and PPC970 engineers, it's clear that the current PPC was rushed to market, leaving off certain features -- they didn't even manage to get the most modern vector processing unit in the chip in time. And the 970FX was just a power-saving version of the original 970.
So I don't think they'd bring an LPAR-capable PPC970 to market, adding that much technology to the chip, without bringing the memory controller back on board, which was technology in the original chip it was derived from in the first place. It'll be a really stupid move if they do.
Is that why the first Opteron Supercomputers on the 500 Fastest list are in 17th and 18th place with 2560 processors and the Apple PowerPC is in 7th place with 2200??? 2/3rds the speed with 1/6th more processors...
Or how about the new 3132 Apple cluster at COLSA Corp. that would come in fourth if it had been running when the tests was made in November?
The 28th position supercomputer is another Opteron that does have the same number of processors as the Virginia Tech installation and it runs less than 50% of the speed...
That doesn't sound like "mopping the floor" to me.
Obviously you do not actually DO any supercomputing or run applications that use it. The Top500 list is considered a joke in the supercomputing community -- it is a very narrow benchmark that basically measures how DSP-like the instruction set is. The PPC ISA literally has DSP instructions built into it; not much good for most real supercomputing apps, but great if you are running LINPACK.
The accepted general purpose supercomputing benchmark used by the supercomputing community is probably STREAM, which maps very well to general real-world performance (LINPACK has almost no mapping to reality outside of a couple narrow apps like fluid dynamics). For STREAM and most other benchmarks, the Opteron soundly outperforms the PPC970 by any metric you care to use. It has 50-100% more bandwidth, fully half the latency, slightly faster CPU performance clock-for-clock, etc. Yes, the PPC does moderately better at the outlier case of CPU-bound DSP codes, but it is mediocre everywhere else. If your business was to know these things, you would know that the AMD64 system architecture is preferred by a wide margin over PPC for supercomputing apps.
I use PPC all the time, but it is also my job to understand its limitations. It excels at DSP, and is mediocre everywhere else. Like most people (and most people that run supercomputer apps), my codes are not at all DSP-like. Clock-for-clock (which roughly means dollar-for-dollar), an Opteron will outperform PPC970 on our supercomputing codes by 50-100%. I like the Macs, but they aren't that hot performance wise in the areas that matter for supercomputing e.g. memory latency. All our heavy lifting is done on Opterons, which simply cannot be beat for performance on a very large number of codes.
You are drinking the same marketing Kool-Aid the Apple was serving when they used that ridiculous Photoshop "benchmark" to show that their older systems were "faster" when everyone knew they weren't. They use an extremely narrow strength that doesn't apply to 95% of everyone, and then imply that it is a general indication of performance. Same story here. The system architecture of the PPC970 is mediocre and the Opteron does school it -- I see it every day because we use both -- but Apple dumps scads of marketing dollars on LINPACK to give it supercomputing "credibility" with the masses (not with the actual supercomputing industry) to imply that the system is more capable than it actually is.
Give me a break. Stop drinking the Kool-Aid. AMD pulled a real rabbit out of their hat and everyone in the supercomputing industry knows it. I would point out that there is a good reason that none of the supercomputing companies have selected the PPC970 for their proprietary systems, but almost all of them have opted for the AMD64 platform in their upcoming systems. I use high-end Macs every day (and right now in fact), but I'm honest enough to admit their limitations.
Show me a real benchmark that is something other than a publicity stunt. Nobody in supercomputing actually uses LINPACK for a number of reasons. Take your pick: STREAM, SPEC, lmbench, etc. and compare the PPC970 numbers to an equivalently clocked Opteron. If you are a PPC zealot, you will not like what you see. This has been hashed out many, many times on the supercomputing mailing lists and forums. The information is out there, if you care to look.
I agree with the rushed to market part, but IBM will have to decide how they are going to manage the differentiation of their Power series and PPC series processors.
Whether people like to admit it or not, the current AMD64 core architecture is very good. It is both remarkably efficient clock-for-clock and has memory performance that no one can touch without prices getting into the stratosphere -- a real class act. And by all accounts, the core was designed to scale very well.
AMD is in an interesting position because they have access to all of IBM's fab process technology (by longstanding agreement), and a core that is every bit as good as IBM's. It is worth pointing out that the AMD64 core was not built internally from whole cloth; there is a lot of next-generation Cray and DEC Alpha inside AMD64. They took what was fundamentally an excellent core design (acquired from NexGen way, way back), purchased next-generation interconnect and processor technology from Cray and the Alpha processors, and built a new architecture from the ground up, integrating a bunch of extremely high-end features into an already very efficient basic core.
Intel is the one screwed here, actually. IBM and Apple have licensed many of the CPU design technologies AMD uses, probably in exchange for access to IBM's state-of-the-art fab technology. Intel is left a little bit out in the cold here, technology-wise, and while they are huge they are also having a hard time keeping up and market share is slipping. But for IBM to remain competitive with the aggressive roadmap of AMD, they really need to drop the pretense of having a separate high-end processor line with all the feature goodies that AMD will sell you for a couple hundred bucks in their mass market product lines. Hobbling the PPC line will eventually catch up to them in the market, because there is no price differentiation on the low-end.
Thank you for the reply.
No, I don't do any supercomputing. I appreciate your explanation and information.
The Kool-Aid remarks are not really necessary. Just explain it as you did.
I am not a PPC fanatic... I do appreciate OSX. If Apple were to pick up the AMD and provide an OSX version for it, or if IBM were to address the issues you and antiRepublicrat have brought up, I would be very happy.
I've been using exclusively AMD at home since I swapped my first 486/66 for an AMD 100MHz board. Then I went K6-2, , Athlon, Athlon XP, and so on. I've used Pentiums at work and have always been more impressed with the Athlons, both for their speed and better internal architecture. The AMD-64 architecture is already pretty mature, so I expect the speeds we're seeing.
The PPC970 artichecture fascinates me because of its potential and radically different design from other desktop chips (having been derived from a server chip). Being that the current generation is the first one, I'm expecting a lot from the next two. I'll consider it a failure if it's not thoroughly competitive with the best from AMD by version 3.
Hobbling the PPC line will eventually catch up to them in the market, because there is no price differentiation on the low-end.
There will always be differentiation between the POWER and PPC, although I think it's a good idea to develop the cool technologies for the high-end chip and let them trickle down to the desktop chip.
The POWER will always have enough to differentiate it from the PPC. There's no way we'll ever see a PPC in an 8-way, 4-chip module with 144MB of shared L3 cache. There's also no reason to blow chip space on the high-end monitoring features of the POWER5 (it internally monitors hundreds of performance and reliability events and conditions, storing them in registers around the chip).
But Apple had better get in gear for a couple of things. First, they'd better bet their MB ready for the better memory management, and they'd better get their OS ready to handle POWER-style SMT. If not, both of those features are useless.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.