Slashdot Mirror
Intel defocussed?
MPR's analysis of Katmai is that it offers no benefits
for business users or consumers not interested in 3D or video.
In fact, Willamette should be shipping by now (4 years since
the P6 was designed), but it hasn't even taped out... suggesting
the proliferation of P6 cores (Klamath, Deschutes, Mendocino, Dixon, and Katmai)
and the development of Merced is spreading Intel's engineers too thin.
It wasn't called Massive Marketing eXtension for nothing. These will just be faster ix86 machines, which should be fine for Intel. Smaller die size, right?
Let's face it: intel is putting a hell of a lot of resources into Merced, and the expect performance of Merced isn't even that impressive. And when you combine it with all the delays, and the fact that it would require special software support to be any good, intel's future doesn't look so great. With AMD capturing more and more of the market, and having upcoming processors like the K6-3 and K7 that will simply outperform anything intel has, I'd say that there is a damn good chance of AMD becoming the market leader. Especially when their Dresden plant is nearly done, and that should solve any production problems.
Tip: buy AMD stock. It's really low right now.
I am shocked, SHOCKED to see that editorial from Microprocessor Report. They are typically quite effusive in their praise of Intel (and equally damning of DEC/Compaq). Perhaps some objectivity is being added to the drinking water up in Sebastopol or wherever their editorial offices are located. Maybe now we'll start to see some more even-handed content. Maybe not.
I am an idiot.
Is it just me, or are these statements not quite in agreement? Sorry for the pedantry; I know "too many baskets for the eggs" doesn't have quite the same ring to it... :-)
And where are their opportunities? DEC was saying for the last 4 years that they wanted low cost cheap Alphas on desktops. Just when the machines got cheap $2-2.5k, the sub-$1k PC was born.
NT/PPC was dropped and while Apple is doing well right now they are the only mainstream user of the PPC. While Apple's recent success may change things, it was only a year ago that Motorola basically gave up on desktop PPC's to concentrate on the embedded market and I doubt we'll see too many embedded Alphas.
I see the opportunities of non-x86 chips closing very quickly. They'll continue to do well in the server and embedded space and continue to fade away on the desktop.
The only real benefit in buying a MMX Pentium (P55c) over a standard Pentium (P54c) was the MMX's larger L1 cache. The Pentium-III (Katmai) has no such advantage over the Pentium-II, so there is little or no reason to buy a P-III over a P-II. The expanded instruction set is really irrelevant to almost everyone. Add on video cards area vastly superior solution to adding 3D functionality to a system. The KNI are a solution in search of a problem.
Oh, and MMX = Meaningless Marketing eXcrement.
The nicest thing currently happening at Intel is the overclockability of the Celeron line....
They have already admitted the Merced will suck. Their leaders point to McKinley and say "THIS will knock your socks off!" HOWEVER, Hewlett-Packard is responsible for the design of McKinley NOT Intel.
The only way Intel can say itself is to buy a hot, new startup with real ingenuity.
The folks at Transmeta are gonna be gazillionaires.
hehe that's a good one
I can't wait for G4/AltiVec to come along and whoop the hell out of the P3s, bye bye Intel.
Sauna time. That's Finnish lingo for the steam bath Finns enjoy. I just came from "Windows Hordes descend on Linux" thread were the temp was 130C and someone just threw a pot of water on the rocks (Linus can verify the wimps head for the door).
Linux has a lot of bad programmers writing apps. Sloppy slow code, and they expect the hardware to keep up with their mistakes. Hey, fools can make more mistakes than wise men can cover over. It isn't Intel's fault your code is slower than molasses.
And after the sauna, a dunk in the ice cold river, a ring of mukkara, and a sauna beer. Gotta keep a sense of humor, too, eh?
ok, what about 3dnow? isn't KNI just 3dnow + a bit more? games are significantly increased in speed by 3dnow, and i suspect kni will at least do that much.
i've also heard about KNI, though, that the implementation of KNI on the pIII isn't as fast as it could be, and we should expect a better implementation on the next chip.
also, i don't see how a pIII with a 4-5 year old core can compete with a freshly designed k7. unless intel can get really high clockrates out of their processors and have 256k+ cache on their piii's, i don't think they stand a chance...
i don't know how many people tried to overclock their system bus when most people ( i still do ) had a 66mhz pII, but AMD's k7 will *start* at 200mhz system bus, going up to 400mhz later on.
wow. that is going to kick some serious ass.
on a side note, intel spends more on advertising than amd has in revenues. you'd think that they'd be able to hire 10x the engineers and pay them twice as much. i don't see how intel can be so far behind.
somehow i think intel has something up its sleeves -- i'm thinking that they'll have their 7th gen x86 cpu out sooner than we think.
jm3
The information leaking out of Intel for the best part of the year has all indicated that the Katmai's L1 cache was 32K (16K/16K), and still everyone has kept assuming that Katmai would have a 64K L1 cache. As we've all seen with the Celeron A (Mendocino) it's not the size of your cache that matters, it's the speed. And it doesn't take a rocket scientist to figure out that a L2 cache on the same die as the CPU at .25 micron is going to run the pants off an external part built at .5, .6 or some such.
Perhaps Intel did some simulation and figured out you didn't gain much from just making the L1 cache bigger and bigger. Actually it probably just gets more and more complex and the access timings get worse and worse as you add muxes and decoders. And perhaps they understand that if you can prevent worthless used-once crap from getting in the cache it won't just churn around. The ability to read and write without going through the cache, and adapt to reading arrays with a step that doesn't match the cache line size is also going to be important.
Intel has a very good track record for designing caches and combining L1 & L2 caches and their controllers. Even if AMD can produce a "faster" product they are likely to be destroyed in the market by Intel's ability to deliver a fast, clean, consistant and mature product.
I would agree that MMX was pretty much a kludge, but 3DNow! is also a big kludge. KNI represents what MMX should have been in the first place. I just wish Intel would have got a clue a shipped this last year. I have evidence that KNI was fully formulated back in Q4 97, and the only reason I can see to delay this long is to spoil AMD's efforts to compete. Katmai's 128-bit internal data paths are going to be hard to map into existing designs.
Intel is a big sucking company, and the P!!! and Merced are GOING to change how we do things, so we'd better start getting up to speed now.
Here's my sketchy view on what's going on: :)
In the good old days there was the pentium mmx and the p2 and nobody cared about lil' AMD and Cyrix. Intel wanted to kill socket 7 which AMD/Cyrix had and replace it with slot 1 which they didn't. Thus, the celeron. That first chip was a bust and AMD racked up all these pc makers: HP, Compaq, etc. Remember the purpose of the celeron was to replace the obsolete mmx but for some reason they decided to keep a separate celeron line instead of the strategy formerly embraced by intel (and still used by AMD) of using one line of chips. This just blurs the difference between the cheaper celeron and the more expensive p2. That's bad. Meanwhile AMD took over the sub $1k pc market hehe. So now intel wants the new kni for p3's and none for celerons which will now be socketed. Cuts costs and shows a difference but AMD chips at the same price-points will have 3dnow! which means more 3dnow! and less KNI. Also bad. Anyway, while intel is still confused, AMD is now working on notebook cpu's. The most popular laptop IIRC is a compaq that uses AMD and they just got Toshiba which makes things even more interesting. And the K7 is also going to be fun
The OS X G4 Alti0Vec should be quite good..
The G4 is Supposed to be good at multiple processors too.
BSD at its core, and maybe POSIX complient?
Especially good for photoshop and those 3d rendering types. And if they keep shipping Rage 128 standard, you have a nice game machine. Now only if they had a common reference board and Apple didn't have all that control (ie cloners)..
Alti-Vec might help emulation?
Pretty soon it will by Unix or Windows.
-A
The new DES cores use MMX intructions where available. I don't know how it works, but they're a damn site faster than the old core (for DES at least).
AC #18761231
Personally, I'm not too surprised that things are going this way.
Like everyone else, I've been amazed at the dramatic increase in microprocessor CPU power over the last 10 years or so. I'm an electronic design engineer and do lots of circuit simulations so I can use every drop of CPU power I can get.
But I'm far from a mainstream PC user - todays PCs have far more power than one needs for most mainstream business applications such as spreadsheets and word processors.
So, what really MAINSTREAM application is there to drive processor performance these days? You guessed it - games and video. 3D and video require alot of computing horsepower, and it's not surprising that that's where the optimization effort is going.
I don't mean the computers they sold but rather PowerPC chips for the desktop. Yes the G3 is here and the G4 will be out later but after that Motorola does not have a clear roadmap. Most of the PPC's they sell are for the embedded market and that is where they are concentrating. With Apple on the comeback they may resurrect some of the projects but I personally find it doubtful.
Examples? Too easy:
1) My mixed mode simulation on a subset of the IC I'm working on. 60,000 instances takes 21-36 hours depending on what I'm running.
2) Design Rule Check and Layout Vs. Schematic runs on that bugger. About 950,000 transistors on a 1/2" by 1/3" i.c. LVS (flat) takes about 21 hours. DRC, about 45 mins.
3) Synthesis and place and route for the 5 100K part FPGAs controlling this thing.
To name a few.
KNI is great for making MMX emulators.
How could you miss that?
java is slow
VB is slow
badly written C++ is somewhat slow
well written C++ kicks ass, it just takes more experience to write decent C++ than decent C
Today's software sucks for two reasons:
1. VB, java, etc are crap
(java is well designed for a beginners language, but is badly implemented - Sun's libraries have decent design but are implemented by people who think efficiency is a dirty word)
2. There are more idiot programmers around today. This is true for several reasons:
java is slow
VB is slow
badly written C++ is somewhat slow
well written C++ kicks ass, it just takes more experience to write decent C++ than decent C
Today's software sucks for two reasons:
1. VB, java, etc are crap
(java is well designed for a beginners language, but is badly implemented - Sun's libraries have decent design but are implemented by people who think efficiency is a dirty word)
2. There are more idiot programmers around today. This is true for several reasons:
There are more programmers (some will be idiots)
People are in it for the money, not because they are natural programmers
The barriers to entry are lower - when you needed to write assembly, you had to be pretty smart to program at all, now people write a few word macros and think they can program.
Sorry about previous post, my finger slipped.
All I have to say is what is going to sell more
processors: games or server apps? I mean what
sold more i386s than Wing Commander did?
I hardly can praise an objectivity of an article
which opens with the following sentence:
"Intel has always been a company of engineers,
run by engineers. "
"... Intel's dominance of the x86 processor
market is due in no small part to the company's
continued technical superiority over its
competitors..."
Excuse me??
--P
highly doubtful -- on both counts. M$ will survive this little "DoJ thing" and Intel
/. Intel has about as
sure as hell isn't going to go tits-up over the likes of AMD. Wake up people. AMD
is a bottom-feeder company with a tarnished execution record.
Oh, and BTW, "wintel" is the stupidest word floating around
much love for M$ as the DoJ right now (you can believe that or not).
Finally, someone has publicly stated the emperor has no clothes. The business community has been screaming that there hasn't been one single advance in computing in the last five years. Any increase in hardware speed has been soaked up by the bloated software resulting from the C++ compilers and object oriented programing.
Computer usability has been steadily regressing to the point where apps are no more productive or useful than they were in the late 80s.
A pox on all of you.
More like AMDinux, or LinMD... but by then, PC's won't have nearly as much impact on the market than smart single-purpose devices running eCos or something.
But I won't know until I test them head-to-head, which I plan to do.
I think KNI can be useful for games and 3D rendering (it should speed up matrix math something wonderful) but it really has nothing to do with normal business use, especially servers. How often does your file server need to raytrace?
--------- Webmaster, http://www.cpureview.com and
Posted by gruv:
You're "stuck" on Intel? I don't like to see that. Your're not "stuck" with Intel. NOt when you have AMD...
Avoiding cache churning would be a big advantage, but as far as I know, PIII doesn't do that.
What MIGHT be a good idea is something like an extension of MTRR to allow little used code (especially one shot init code) or data to be marked no cache to keep it out entirely. This perhaps with a persistant attribute for frequently accessed code or data. Ideally, compilers would automatically optimize by creating no-cache and persistant sections to minimize the number of memory zones affected by the new attributes.
The drawbacks to that include increased complexity, and the need for compiler support (or inline assembly) to see the benefits. Without compiler supported zone allocation, you'd probably run out of attribute registers and end up leaving some sections unoptimized. It would be really easy to do a bad job of optimization. Of course, KNI has the same problem.
In the meantime, that prices Xeon systems up in the lower stratosphere with faster RISC systems including Sun Enterprise servers.
A Pirate and a Puritan look the same on a balance sheet.
Do you wonder where these analysts get money from? They sell reports to the companies who buy these
processors. Nobody wants to hear what they just bought was junk, they want to hear it is the best.
And, until recently, everyone bought Intel. Now that more companies are designing-in non-Intel
processors, the coverage will probably reflect this.
You have to hand it to MDR, though. Almost everything they predict eventually happens (but
not usually in the same timeframe). But this is probably just a result them printing the press-releases
of Microsoft and Intel and post-hypnotic suggestion of the marketing guys who read their newsletters.
Er... my BBC Micro could do real time spellcheck back in 1987. That's a 6502, for all you suckers don't know. The fact that it's taken Microsoft so long to implement it is merely an indicator of their mind-boggling incompetence.
fish and pipes
Can anyone think of an application which is accelerated substantially by MMX or KNI, but which could not practically be accelerated by an add-on board?
The defects of add-on boards are that you can't put much RAM on them, and that the link to the CPU is slow. For graphics, the slow CPU link means you throw vertex data across, and get your textures by DMA, and use nifty things like DX6 texture compression to get around the lack of memory on the video card.
For sound, you don't need much RAM and you don't need tens of megabytes per second of bandwidth, so add-on boards are not a problem.
The obvious scientific applications really need more than 16 bits of precision - so you can't use MMX. Possibly KNI would be useful for data analysis - I think you could write an incredibly fast single-precision FFT using it.
For working with large numbers (RSA crypto, lots of mathematics), you really need double-precision FP - so KNI is useless - for the FFT approach to bignum arithmetic, and you want add-with-carry of packed 32-bit or 64-bit numbers - which KNI just might have - for the schoolboy approach. And, unless your numbers are incredibly large, they'll fit in the RAM of an add-on card; unless you're doing fairly unpleasantly complicated tasks, you can do all the work in the add-on card's RAM and not worry about transfer bandwidth at all.
OK. So what applications are MMX or KNI actually useful for? The only one I can think of, possibly, is clever video decompression where you work from tens of megabytes of stored history - but video decompression runs into the bandwidth-to-graphics-card bottleneck. So that's no good either.
I'm not sure hardware speed has been soaked up by bloated software. I think it's more a matter of hardware speed having gone beyond the capacity of software to bloat.
Sometimes you can get advantages out of hardware speed - but emacs can do real-time spellcheck and autocorrect on arbitrary small Linux boxes, and even Word can do them, with non-noticable delay, on a P133.
As far as we know, You can't use all of the performance of a P2/350 to wordprocess or to browse the Web - while I'm using this PC, and everything is running fast enough to keep up with my typing and update the screen smoothly, 85% of my cycles are going to Prime95.
If a P3/700 is three times faster than this box, all I'll notice is that 95% of my cycles are going to Prime95. It'll still keep up with my typing; it'll still update the screen smoothly.
!GIVE ME AN EXAMPLE OF AN APPLICATION (not a game, that's too easy - though even games aren't that bad now we've got ubiquitous 3d acceleration) THAT REQUIRES A P2/350!
There WAS NT on PPC. It died of neglect due to no native apps being available for it.
PPC/MacOS may not be such a bad thing by the end of the year: BSD based OS, end-user apps, newbie-friendly GUI...
And you can always run Linux on PPC....
One thing that's kind of funny about Intel is the price pressure they're putting on themselves. In an effort to milk the top end, they're in the situation where ~10:1 price difference buys ~2:1 performance difference.
I read that Intel's down to 75% of the x86 market, and dropping. I can only see that trend continuing. Merced's performance won't be compelling compared to x86 chips which will be available then. Maybe McKinley will turn things around for them??
I too was saddened by the demise of Mac clones. It was one way to keep PPC machines more competitive with x86 machines. Apple seems to have realized their previous pricing is not going to fly, and they are moving to be more competitive.
:-(
One sad thing about Moto is that they have PREP boards, they make the chips, they have (had) a subsidiary to make the boxes. They COULD build a PPC box with Linux on it. Evidently, they would rather focus on the embedded market
I thought this name was obvious. Kinda reminds me of Lisp...
I think the article is looking in the wrong direction for Intel's "spreading too thin" problem.
Their sudden and unpredicted need to churn out a series of Celeron parts must have screwed up all their other schedules. A year ago there were just realising the problem. They quickly introduced a crippled processor everyone laughed at. They had to do a better one real quick. Now they can't make any money from it, as the package costs a bomb (or is that too much of the BOM?), so they have to put more effort into Socket 7, er, Socket 370 (is that like an IBM 370?).
AMD's sales are small compared to Intel, but it seems that to a substantial extent AMD is now driving Intel's development plans.
Everyone called Intel the one product company. Now they are fighting this claim with a host of fragmented product lines.
When you compare Intel with a broad line supplier like AMD, the "spread too thin" argument doesn't hold water. From must lower total revenues AMD finances the development not only of their x86 series, but of a wide range of other products. Intel makes the CPUs, the supporting chip sets, and just a few other PC related parts. They keep pulling out of other activities, like their microcontrollers.
Example of using a fast PII fully under Windows 95, 98 or NT:
1. Start the CPU monitor
2. Put your mouse pointer over a large window's header bar
3. Press the left button, and keep is pressed.
4. Wiggle the mouse from side to side, and watch the CPU monitor.
That can keep a PII/400 fully occupied. It even keeps a high end Alpha running NT 4.0 about 70% occupied. Even with the latest video accelerator cards Windows still requires huge compute power for crisp display update. NT3.51 required only a fraction of that power. Sure a slower processor works, but the feel of the machine is much better with the faster processor. Everything is so jerky with a more mundance CPU.
the end is nigh for the two
traditional computing superpowers...
bye bye, wintel machines.