Slashdot Mirror
Merced Design Completed
NoWhere Man writes "Merced's design
is complete and are due to go into production mid-2000, but it is expected that McKinley, Merced's successor (due late 2000), will likely be the most popular in Intel's 64-bit chips.
"
← Back to Stories (view on slashdot.org)
As anyone in the hardware business knows, there is a huge gulf between what Intel's just announced (preparing the design to be fabricated into physical chips for the first time, traditionally called "tape-out" becasue the design data used to be shipped to the people making the masks on spools of tape) and actually being ready to ship product to customers (i.e. OEMs).
Keep in mind that this means they don't have a sungle physical Merced chip yet, which means they've only been simulated, never really tested. Although I don't know anything about Intel's internal simulation tools and methodologies, I'd be surprised if they've even been able to boot an operating system on their simulated design. I'd be amazed if they've simulated significant real-world applications. In other words, at this point Merced has had very little real-world testing. That's what the next step ("first silicon") is all about: testing out whether the design actually works in practice (and systems) rather than just in theory (under pre-silicon simulation).
That's not to say that Intel hasn't expended a huge effort already in working to eliminate as many bugs as possible before taping out (in fact, I'm sure they have), but there are always more bugs when the silicon arrives. In fact, many of the problems encountered in first silicon are effects which simulation can't (or at least doesn't) effectively capture (race conditions that earlier analysis missed, unanticipated electrical and electromagnetic effects, yield problems, etc.). These problems are worse when you're talking about a new from-scratch design. There are also bound to be other problems when you're talking about a totally new processor architecture (such as IA-64).
Usually the process between tape-out and shipping to clients goes like this:
To me, it seems that the estimate of one quarter between tape-out and shipping samples to OEMs seems extremely optimistic. Of course, I suppose Intel could be planning on using the OEMs for debugging their first or second pass silicon (before they've made it really robust), but somehow I doubt that. I would expect that it might be more like two quarters before anybody outside Intel actually sees a physical Merced chip.
But then again, hey, what do I know.
Kenneth C. Schalk < kenneth.schalk@compaq.com>Software Engineer
CAD & Test Group
Alpha Development Group
Compaq
I just had to reply to all these random comments about Merced to answer a few questions. I don't know everything because I didn't design it but I've heard bits and pieces. These first concern was when is Merced going to be out. I noticed there was on reply that explained that pretty well. They just finished the DESIGN of the chip. You are going to have to wait until the processor is taped out, and silicon is ready for system validation. After the initial chip is produced which takes a few weeks, it will be hammered for bugs. I also wanted to comment about it being buggy. Intel learned from the first pentium that had problems. You don't make a mistake like that twice, but there is no such thing as a bugless chip. Some bugs will never be found. Sometimes you have to run a single test 100,000 times to find one bug. But you better bet this chip isn't going to be the Windows 95 of microprocessors. I also wouldn't get your hopes up to actually get one of these in your hands. They are going to start out like the Xeon. Really expenisve for businesses. There is no way everything is going to jump from 32 to 64 bit architecture over night. So keep your pants on, it will be a while. Anyway, I can't wait until they are affordable but it will be a long time. Faster games are always good, but they don't rely on CPU's as much anymore. Processor speed isn't the bottle neck anymore. Think about it. I have a Pentium 400 at home, but the bus speed is only 100. Anyone see the problem? Oh yeah before I forget. I don't know what you are all talking about the Alpha chip for. If you mean the really fast one, that Digital designed... It doesn't exist anymore. Digital is dead. It was bought by Intel and Compaq. That is what I do. Intel got the StrongARM chip from Digital. It will be used in top set boxes. Really useful when digital television becomes a standard. expect to be able to surf the web on your tv pretty soon. I think some of the technology that went into the StrongARM cames from the Alpha chip. Its a risc processor, the 2nd chip due out should give 600 MHz at half a watt. Check it out on ARMs homepage. :)
Linux will not be a 32-bit OS on Merced. If it can be made to run natively at all (and Linus has been quoted as saying that this is a "done deal"), it'll be full 64-bit, just like it is on Alpha and UltraSparc.
>Look at Digitals compilers under Digital Unix.
.25G.
>Produces much faster code than \1. But does
>that matter one bit for the \1 community? No.
>Didn't think so.
Speak for yourself. I work daily from my FreeBSD box to my boss's linux box running a commercial fortran compiler (g77, etc., don't even play in the same league). We have absoft fortan, but it would have been digital if it were available.
We were even willing to pay the extra cost for the alpha box, but the costs of DU itself, both for purchase and the risk of getting sucked into the university system and fee'd to death there, mean the x86/linux/absoft solution.
A year an a half later, I've hit the price. I never thought I'd see the day I *needed* a 64 bit operating system, but now I do: I need an array with more thatn 2^32 bits, and more than 2^32 bytes would be nice, too. Absoft uses Cray code that bit addresses, leaving the size limit on an array of derived type at about
There's also something bigger, better and eventually cheaper on the horizon. If a system meets your needs now then get it.
Does it also mean that it is a bad idea to buy an "old" PentiumIII processor ?
Software written for the Merced may not run on a PIII, but software written for older x86s will still run on the Merced. Just as current x86 chips can emulate "real mode" for older apps, so the Merced will be able to use and/or emulate the processing modes used in current x86s.
A dedicated x86 clone - like the K7 - will be able to run these applications faster. However, _if_ they did a good job on the Merced's core, applications written natively for the Merced will run faster than applications written natively for the K7, as the K7 will still be hampered by the x86 instruction set and register structure.
_If_ Intel did a good job on the Merced core, it will be fast but still 1.5x as expensive as other RISC solutions due to the extra silicon needed to support x86 legacy features.
However, I gather that they may not have done such a good job on the core. We'll see when prototypes are benchmarked.
Please click on "user info" above and see my previous response in this thread. It is a reply to another poster who presented almost identical arguments.
Again Linux is the exception....
Correction: Windows is the exception. See the post that I referred to.
Re. games, I realize that most game software isn't perfectly tuned, but all of the "boost FPS by 50%" optimizations will already have been made, because it is in the game company's financial interest to do so - as a result of this optimization, they can either lower the system requirements or keep the frame rate and jack up the graphics detail. Both correlate directly to better sales.
From where do you get the impression that games are horribly written?
Um, no. New games require more hardware because they have fancier special effects and more detailed models. This is not really related to code complexity. It makes the _data_files_ larger, naturally, but that's about it.
Granted, there are some game writers who consider special effects a reasonable substitute for gameplay and plotting. These writers' games will sink, however, because consumers do want games that are actually fun to play.
Re. hardware vs. games, game hardware requirements will plateau when cheap hardware exists that can handle just about all of the special effects in the OpenGL feature set for photorealistic models at high resolution in real-time. Beyond that, there isn't anything left to add hardware load on the graphics side of things.
Things like AI and physics may continue to develop after that, but physics at least won't add much more load if you have hardware that powerful.
Added to that, processor design has become more bloated, moving deeper into a large, complex instruction set. Simpler processors, such as the ARM, outpaced the Intel chips even at a fraction of the clockspeed, because they were better designed.
Um, no. Look at just about any non-Intel processor. Intel chips are bloated because Intel continues to support and extend an instruction set that wasn't designed to be extensible. They're about the only major microprocessor manufacturer that made this mistake.
Also, didn't ARM not _have_ a floating-point unit? With more silicon to devote, of course they'll be faster at integer operations.
Finally, throw in that most modern OS' are bloated and top-heavy, Linux being one notable exception
And *BSD and BeOS and...
Microsoft is the primary culprit for slow OSs. This is because Microsoft is purely market-driven, and the market that they cater to would rather buy a new version of the OS with more features than a new version of the OS that works more efficiently.
OSs and chips can be designed cleanly - and _are_, with only a few exceptions. Take a look around at what's available, and you may be pleasantly surprised.
It must be nice to have a major website quote you without any challenge whatsoever. Where's the skepticism? If Intel has missed all of its other targets, why does anyone think Merced by middle 200 will be any different? I'm not knocking what they are trying to do, but they are still along way from a product anyone can use, and there's no guarantee that this will be a success.
/. but you're young, you can afford to slum a little.
This article from Byte goes into some of the problems Intel has from this stage forward. A little low-tech for
First off, i have some pics of the Athlon, Merced and G4 here. TheRegister has a lot of info on whats going on in the CPU market.
Basically: Merced has been on the brink of failure for quite a while now. The performance of the ones made so far are considerably less than those of the PII at lesser Mhz's.
The development of the Althon (aka K7 by AMD) has been quite secret. It is actually a super powerful chip and is using something like 256k cache to bring down price and still will whoop the PIII at equal Mhz (and 512 cache, in FPU benchmarks too!). Rumor has it that they will be releasing a 512k and 1mb cache intel killer Athlon shortly after debuting it at 256k. The Athlon will be using a slot A which makes sense as they have been in bed with Alpha Processors Inc., Samsung's processor company. (Those of you who still think that Digital owns API, you're mistaken as they made a deal to sell off their majority in the company to samsung). So as we see 1ghz Alphas debut without cooling you know that they are sharing that technology with AMD so the future is really bright for AMD. One nifty thing is, on the register, i saw an 8 proc. motherboard being made for the Athlon. Hello low-cost supercomputing.
Motorola, makers of the PowerPC processor line will be introducing the G4. Rumor from the mac side is that due to a dispute between Motorola and IBM, who share in the production and design of the PowerPC processor line, there will be two different versions of the G4 coming out. One, which will be made by IBM will include a special instruction set that Mac OS X can/will be optimized for that will increase 3d rendering (sorta like 3dfx from what i understand of it) wheras Motorola will make non-optimized G4s that cost much less than the IBM manufactured ones. This probably means that lower cost macs, such as the iMac will use the Motorola G4 and upper end Macs like the PowerMac will use the IBM one.
-Z
I'm afraid. I'm afraid, Dave. Dave, my mind is going. I can feel it. I can feel it. My mind is going.
Intel themselves have said in press releases that IA-64 is intended to be a server architecture. It may happen somewhere down the line that they slowly migrate into the consumer realm, but not even Intel is pushing that concept. If it's any indication of their plans, the "Willamette" architecture has been handed the "P7" designation, not Merced. Maybe they'll call it Pentium !V or something like that.
Merced family is heavily dependent for performance on paralellizing compilers. I suspect that making the silicon will be the easy part (I'm a software guy, hardware guys may disagree with that), but making good compilers to take advantage of the chip will be a bitch.
It seems that we are entering an era when the performance of your application is going to depend on the quality of your compiler/interpreter as much as on the actual hardware inside the machine. This is both good and scary. Good if the free compilers (like egcs) will be able to compete with and outperform commercial compilers -- that will be a great boost to free software. But there is also the scary part: if the free compilers fail to keep pace with commercial offerings, they will die. Think about it: if a kernel compiled under, say, Sun compiler will run twice as fast as one compiled under gcc, what will happen to gcc?
Kaa
Kaa
Kaa's Law: In any sufficiently large group of people most are idiots.
What reason will there be to buy an Intel IA-64 chip when the major reason business' buy Intel is backwards compatability. I will be able to get a 64-bit Alpha chip that not only outperforms but is cheaper than a 64-bit Intel. The reason people kept buying Intel is backwards compatability, and while this still maybe true it is irrelivant. The people buying 64-bit processors are usually corporate customers with large databases in mind. So unless, maybe you just want a kick ass game box, why would one buy it when there are far better solutions through other vendors.
---- aut viam inveniam aut faciam