Slashdot Mirror
Oracle Won't Abandon SPARC, Says Ellison
fm6 writes "When the Oracle acquisition of Sun Microsystems was announced, it was widely assumed that Oracle was interested only in Sun's software technology, and would sell or discontinue all its hardware businesses. Larry Ellison, in an interview just posted on the Oracle web site, says that's not what's going to happen. In particular, SPARC isn't going anywhere (PDF): 'Once we own Sun we're going to increase the investment in SPARC. We think designing our own chips is very, very important. Even Apple is designing its own chips these days.'"
"Even Apple is designing its own chips these days."
Unlike Oracle, I think Apple is traditionally a hardware company.
I wish them the best carrying on the Sun baton.
Well, of course he's going to say that - he's not just going to say "well, we're planning on axing 20,000 jobs and kissing bye-bye to the SPARC line". He has to at least maintain the *illusion* that they're going to keep producing SPARC chips.
I love the line about "even Apple" is designing its own chips. One could say "even Sun" sells Intel.
Cemil.
For many years, there were a multitude of different architectures, and all of them were supported by major software developers. Over time the number has gotten smaller and smaller, the only one used in typical desktop computers anymore is the x86 (mainly thanks to Intel investing mountains of money into the manufacturing process). Unfortunately for Intel, manufacturing isn't the advantage it once was: AMD is still able to compete with them moderately well even when they've been a generation behind in manufacturing. Other things are coming into play besides raw processing power, things like power consumption and battery life.
Intel is going to have trouble competing on battery life with ARM, or even PowerPC. Going into the future, we are going to see more ARM based netbooks (and they are going to be more usable), and the already common ARM handheld device is going to become more powerful. Suddenly there is going to be a need for software that runs on more than one architecture again. This is a good thing, in my opinion: it means x86 will not necessarily be the dominant processor forever into the future.
Qxe4
While Oracle is big, I kind of doubt that they could ever keep up with Intel. Even in turn-key appliance servers (sort of an iMac of databases, pre-configured computer), Intel/AMD will outstrip them in performance and they won't be able to stay up to date.
The only place I can think that this would be useful is routers. In a turn-key appliance like that that does a very specialized job (especially one that requires custom silicon to do the routing fast enough), SPARC could make sense. It would make it harder to steal their software (because you'd have to run on SPARC). It would give them total control (no need to source processors from external companies). They could even build the SPARC cores into the same chips that hold all the magic high-speed routing magic.
SPARC could be useful, but I doubt they'll try and compete in the general market.
This is just off the top of my head. Is there something special about SPARC that would make it remarkably good at some specific application that Oracle uses?
Comment forecast: Bits of genius surrounded by a sea of mediocrity.
People have been saying for years that we're about to reach the end of the line in terms of Moore's law. So far they've all been proven wrong, and scaling continues unabated.
Dumping processors in a box is "easy", but multicore programming is not easy. The software tools are not there yet. Not to mention, you need deep pockets to roll your own multicore IC and build up the requisite software ecosystem. Just look at how much trouble Sony had with Cell. Everybody is watching to see if Intel will succeed with Larrabee.
Now Oracle may have good reason to be interested in Sun's Niagara. Database applicances, perhaps.
And where does Apple come into this, exactly? PA Semi's focus is on a totally different market segment.
I'm trying to figure out if that was a insightful speculation or a bunch of words thrown together randomly.
This is just a minor nitpick, but knowing what your customers want is part of the marketing. Marketing is not just advertising, though many seem to forget that.
People have been saying for years that we're about to reach the end of the line in terms of Moore's law. So far they've all been proven wrong, and scaling continues unabated.
Unless you know something I don't, you can't make a silicon wire smaller than the width of a single atom, so there is definitely a physical limit that we aren't that far away from. I've read that practically, the limit is 4nm for silicon nanowires. That means that if we're at 45nm today (Intel's 32nm chips are slated for 2009), and we're assuming size shrinks 50% every 18 months, in less than 72 months we'll have reached the practical lower limit for silicon features. Even assuming that you can make silicon chips with wires the width of a single atom, given that the atomic radius of Silicon is 110 pm, that only gives 144 months.
In addition to that, at 3.2GHz, light in a vacuum can only travel about 9.36 centimeters per cycle. Given a dialetric constant for the Si02 used in chip manufacturing of 3.9, you can calculate the velocity of propagation of the electromagnetic waves through the Silicon as about 50.6% of C. Therefore, at 3.2 GHz, the electromagnetic waves inside the chip can only propagate about 4.7 centimeters per cycle. You also can lose a bit depending on the switching speed of the transistors, but they actually become faster the smaller they are, so the real limiter is the propagation speed.
You've probably noticed that we haven't had any really major jumps in the clock speeds of consumer processors since about 2002. Intel originally thought they'd be able to scale the Pentium 4 Netburst architecture to about 10GHz, bu they ran into a frequency ceiling at about 4GHz.
In short, unless there is a major materials breakthrough, or materials change, I would expect Moore's law to hold for the next five years or so, but not much longer after that. We're rapidly approaching the physical limits.
You don't make the poor richer by making the rich poorer. - Winston Churchill
"But the appeal of the Mac and the Lisa was as much or more fashion and style as it was practical."
That's an interesting statement, and it betrays more about you than about the topic we're discussing. I remember back when I went to school and the schoolwork our teachers handed out suddenly changed from photocopied hand-written stuff to neatly layouted, professionally looking stuff. That was when the Mac came out and normal people were suddenly able to use computers in a meaningful way.
You're a geek. You don't care about normal people, because you were perfectly happy with DOS or whatever you were using. To you, all that stuff that made computers usable for everyone else was just "fashion".
You were as wrong then as you are now.
To you, the iPod is a fashion statement because you were happy with the MP3 players that came before the iPod. To most people, those were unusable, bulky pieces of crap. You were happy with cell phones before the iPhone came out. Most people hated their cell phones and used them only for the most basic things.
Perhaps creating things normal people can actually use seems like "fashion" to you, but most people don't use these devices for their own sake; they don't enjoy learning complex stuff just to learn complex stuff. They want to get stuff done, and all of those things that you like, all those ways you can tinker with your toys actually only get in their way.
Apple's success is not about fashion and style, it is about normal people getting stuff done.
There is potential, especially in data mining. The equivalent of "run a task 2 million times" is "search through 200 million rows in a table." A speed-up of two or three orders of magnitude is straightforward (SSDs and n-way parallel processing, n>32), but getting the next two is not quite so easy. Specialised hardware might help (4096-bit data paths, anyone?)
Traditionally, the big issue in database is disk management. It seems to me Sun has quite a good track record in this area, and system administration generally. Sun's recent experience with ZFS might be of use to Oracle too - not so much ZFS the product, but the experience and insight gained by the engineers working on it.
In all, I think there are quite a few possibilities for Oracle. IBM may be kicking themselves in a few years for missing the opportunity of keeping Sun out of Oracle's hands.
People have been saying for years that we're about to reach the end of the line in terms of Moore's law. So far they've all been proven wrong, and scaling continues unabated.
That means that if we're at 45nm today (Intel's 32nm chips are slated for 2009), and we're assuming size shrinks 50% every 18 months, in less than 72 months we'll have reached the practical lower limit for silicon features.
I don't know if you realize it, but you are really just confirming the OP's point -- you are just another person predicting the end of Moore's law based on the technical obstacle du jour.
Moore's law is solely about the number of transistors on a single IC for a constant cost. Feature size may appear to be a limiting factor, but that doesn't mean it will be one when we get to that point. Just like leakage for features sizes below roughly 100nm was once thought to be an insurmountable obstacle to Moore's law, and then some smart people figured out how to handle it, or how lithography processes were also considered a limiting factor below roughly 60nm -- until they weren't any more.
So maybe 4nm really is a hard limit, somebody will come up with something to get around that obstacle - like say 3D ICs - adding a couple of layers and you've easily doubled the number of transistors on the same size chip.
In short, unless there is a major materials breakthrough, or materials change, I would expect Moore's law to hold for the next five years or so, but not much longer after that.
The smart money is on the breakthrough, we've had plenty of them before and there is no reason to believe they are going to stop coming.
When information is power, privacy is freedom.
Biggest fight between Apple and IBM came from IBM not keeping the promise of shipping a 3 Ghz G5 for Apple right? Also they didn't make something for portable which is the future. Months later, they shipped an architecture which can scale to 6 Ghz (by some exclusive tech) and shipped a real working 4.7 Ghz enterprise CPU (POWER6) which they keep selling. So, IBM isn't just capable of 3Ghz, they have such a technology in hand making competitors Mhz look so funny. It is almost like ultimate justice for years of Mhz myth by X86 vendors.
Apple didn't design the entire G5. It is actually scaled down POWER4+Apple design choices+Altivec (which almost shouts like "I come from Apple").
IBM wants to stay away from "end user" and they want to sell CPUs to companies who makes consoles/very high end TV/BluRay etc. XBox 360, Sony PS3, Nintendo Wii are all IBM CPUs designed with the respective partners. XBox 360 is almost designed for MS engineers needs, that is how it does serve them great. I was visiting a friend at IBM one day, one line had a 10.000 client network having some speed issues and other line was a teenage bitching about his FPS performance... It was in 1990s and when IBM sold their PC division to Chinese, I wasn't surprised a bit. Enterprise and end user really doesn't go together.
Apple also wanted this situation: Consumers should be able to run x86 software and even can run Windows as exclusive OS (if needed). Don't let the comments/rants fool you, there are some amazing numbers of virtualisation software/ boot camp updates downloads from sites like versiontracker, macupdate etc. It is only x86 which can do it, you won't be emulating a same generation CPU with something completely different down to endianness. I actually run MS Virtual PC 7 (with their exclusive info,undocumented access) on Quad G5 2500 Mhz. Trust me, x86 isn't easy to emulate even if you are Microsoft itself. For year, before iPod, people had question "What happens if Apple dies?". If you ship them something that can run Windows even better than generic PCs, you won't have that question asked at all.
Basically both companies wanted to end partnership. Steve Jobs likes to have "No 3Ghz for me, damn you IBM" and IBM likes to exit end user chaos, both are happy and interestingly, consumers are also happy. People actually hoping for CPU arch competition aren't happy, that is it Intel also lost a good reason to push SSE etc. achievements, who will ship something like Altivec now? AMD?
No, I think you're missing the main reason Apple dumped IBM. Apple saw the market moving towards laptops, and IBM couldn't bring the operating temperature of the G5 down. Apple never built a G5 laptop, and it was killing them. Meanwhile, Intel was building fast, low power CPUs and chipsets, and in the quantities Apple wanted. Apple could build more powerful portables, and smaller, lighter, more compact desktops like the iMac and Mac Mini, as a side effect.
The virtualization was just a nice bonus. It's actually easier to emulate an x86 on a RISC chip than the other way around. The Rosetta guys did some amazing things to get PowerPC code running on Intel, and even then it was just a stopgap measure.
www.lucernesys.comHorizon: Calendar-based personal finance
With all the talk of container and "lego" data centers, Oracle wants to become fully vertically integrated so that you can go to Oracle and say: "I've got $10 million -- sell my data center blocks".
Sun's already been developing their own data-center-in-a-shipping-container, and Oracle now has all the bits and pieces:
Also, having a horde of hardware engineers is Ellison's wet dream. As I said before, Larry Ellison wakes up every morning and asks himself, "How can I [fsck] Microsoft today?" Larry has stated in the past he wouldn't mind moving beyond databases, and with Sun's hardware and Java, he's poised to do pretty much anything he wants. So he might entertain delusions of mobile, return of the net appliances, home multimedia, etc. In the short term, though, I think he's hoping he can create custom hardware to make Oracle and Java run much faster. Will he succeed? Dunno, but Larry Ellison has a ferocious desire to succeed, and often, that's all you need.
The bitter lessons of a veteran coder: http://bitterprogrammer.blogspot.com
They also built things no one wanted. In fact, they had a really hard time figuring out what people wanted, this was their weakness.
That was supposed to be the job of their ambassadors and maybe the sales/marketing people - to get feedback from potential customers as to what they wanted to see in future products. Problem is, they mostly wanted a solid reliable OS that that they wouldn't have to wait for the first service pack before upgrading an entire department as well as having a competitive price/performance ratio.
For Sparc processors like Niagara II, the server group would want more cache and hardware support for encryption, but the workstation group would want more floating-point processors. In the end they both get what they want with multi-core chips.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
Months later, they shipped an architecture which can scale to 6 Ghz (by some exclusive tech) and shipped a real working 4.7 Ghz enterprise CPU
IBM could make a chip which ran up to 4.7GHz, but did you see the cost of it and the power consumption? IBM didn't have anything that could go in a laptop, and with the PowerBook the best-selling Mac stuck shipping a 1.67GHz G4 while the competition was shipping 2GHz+ chips with two cores using less power for the same or better performance. The G5, even at 2.7GHz, needed massively engineered cooling.
I actually run MS Virtual PC 7 (with their exclusive info,undocumented access) on Quad G5 2500 Mhz. Trust me, x86 isn't easy to emulate even if you are Microsoft itself
What undocumented access? VirtualPC 7 is an incremental improvement on VirtualPC 6, which Microsoft bought from Connectix. It's a fairly good x86 emulator, but it's based on old technology. Microsoft have no incentive to improve a product that makes it easy for you to migrate to a competitor's product.
I am TheRaven on Soylent News
It's actually easier to emulate an x86 on a RISC chip than the other way around.
The distinction between RISC and CISC are largely meaningless when talking about an x86(-64) CPU. If you can decode the instruction set, you can emulate. The hard part is emulating the attendant chipsets and their interactions with the emulated and real system.
Oracle wanted the hardware, so they could become the kind of top-to-bottom solution that IBM used to be in the Mainframe days. IBM failed to prevent it, so now they're loudly saying "sour grapes! sour grapes!"
I suspect the commentators who missed why IBM and Oracle wanted Sun were the same ones who said IBM and Sun were doomed technologies, and that the future was NT 4 on Intel x86-32.
And to answer the question literally, you put your marketers on marketing the company while you put your lawyers on working on the merger. I assume they're different people (;-))
--dave
davecb@spamcop.net
Afaict the hard part in emulation is doing it fast.
Simple interpretive emulation is pretty easy and if you only want to run apps from another CPU (rather than a whole OS) you don't need to emulate much in the way of hardware since you only have to emulate the userland environment.
If you want good performance from your emulation you have to use "dynamic recompilation", basically converting the machine code from one CPU to another in blocks and then emulating it.
x86 (remember the original intel macs were NOT x64, that came later) is widely known as a register starved architecture. PPC OTOH has plenty of registers.
I would imagine translation of code from a register poor architecture to run on a register rich architecture would be much simpler than translation of code from a register rich architecture to run on a register poor architecture.
note: i'm known as plugwash most places but i screwd up registering that here somehow in the past and now can't register