Slashdot Mirror
Intel Releasing 700Mhz P3s
NoWhere Man writes "Intel plans to release Coppermine, its 700Mhz+ series of Pentium III chips, on Monday. The new chip will, not only be available for desktop machines, but notebooks aswell; thanks to a new design which makes them cooler. This release will allow Intel to say, once again, that it offers the fastest chip on the market. "
How many programs will require more then 1Ghz?
I am sure gamers will want these. Ditto anyone using software like Mathematica.
Which are two of my favorites.
This topic deserves some intelligent conversation. I can't find any benchmarks on spec.org for PowerMacs (which is silly), but I'm going to make some educated guesses. I think these are fairly reasonable assessments, quite unlike all the "MAC SUX" stuff so far.
o A regular integer-centric application (Word, Excel, etc.) is likely to be faster on a Pentium 700 than it is on a PowerPC 450. It's generally accepted that a 450mhz G4 is comparable to a 550mhz, possibly 600mhz Pentium.
o An application optimized for the G4's Velocity Engine (AltiVec) is likely noticeably faster than an application optimized for a 700mhz Pentium. These would include Photoshop, Media 100, Media Cleaner Pro, and possibly SETI@Home.
o An unoptimized (on either side) floating point-centric application may be closer to a toss up.
o A pentium 700mhz likely runs significantly hotter and sucks down more enegry than a G4. While G4s consume a bit too much more at this point to be great for laptops, there's much better chance at them going mobile that the high-end pentiums anytime soon.
o Quake 3 will probably be quite a bit faster on a P3 700 than a G4/450 (based on Carmack's recent observations on the topic).
o Somebody who stares at xterms or MS Office all day probably isn't going to notice much difference one way or the other.
I don't know enough about Athlon, so I'll stay out of that.
- Scott
------
Scott Stevenson
Scott Stevenson
Tree House Ideas
this is like when you write a recursive function and forget to check for an end state and return;
To buy a chip:
wait for the faster one
then a buy a chip
and of course, to buy a chip, you recur
wait for the faster one,
then buy a chip
and so on.
eventually you run out of stack space and are forced to go back to drooling in a padded corner with an etch-a-sketch
-I go to Rice, so figure out my email address
I'm gonna lay my thoughts on the line and let people who *really* know about cpu architecture let me know the truth :)
:))
;))
As far as a know/can tell the CuMine is simply a bog standard p3 core (maybe with a few tweaks, but nothing major) with a 256K L2 Cache running at core speed, all running on a 133 frontside bus.
From the benchmarks i've seen so far, (Anandtech being one of them) the new CuMine gets close to/beats Athlon at the same/simlilar clock speed for some tests. So it kinda puts Intel back into the game, although they won't be able to stop fighting yet.
Now from what i've read Athlons can support upto 8M of L2 cache running at core speed. And they are planning on upping the EV6 bus to 266(133) in the Ultra chips i believe.
If AMD wanna do an Intel on Intel then all they need to do is jack up the speed of the L2 cache to, or close to, core speed. I'm sure that because of the better architectire of the Athlon the core will be able to make better use of that high speed L2 cache than a p3 core can. The p3 core is really the bottleneck in intels current designs. Adding high speed cache will help, but when it comes down to it, i have a feeling that an Athlon would get more out of it due to it's better design.
Now that all relies on AMD being able to produce memory able to run at that speed, and produce it in large quatities. This could be AMD one failing.
Fab30 *may* change that though, we will have to wait and see.
As for me, i'm gonna wait until the 0.18 Athlons are out and have settled down, by which time *hopefully* someone will have produced a quality SMP capable board.
If anyone who knows there stuff would like to correct me on anything, please go ahead. Knowledge is power after all
Acctually... :)
The Alpha processor is probably the worlds fastest . Those 21[123]64 chips are pretty amazing.
They're still king of the hill..
(-1,offtopic)
Well, they just announced a 733 MHz processor today, so I guess I was right =)
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
There have been reports about these before; it seems Intel had released some of the chips without as many NDAs as before, and people had been benchmarking and overclocking them :-).
The higest speed they released was 766; this doesn't mean that the official release won't have higher speeds, of course. And someone overclocked one a little, nothing really impressive.
The Athlon 700 matched a P766 on a 820motherboard in most tests. Of course, the 820mobo is really nice, with 4xAGP and such -- but it's _completely_ unavailable. So expecting a P766 to match an A700 on available mobos is unrealistic.
And, of course, AMD supports 'overclockers' officially, so you can buy K7-900 systems with warrenty intact. I found another place which sells overclocked A750 for the cost of an A600 elsewhere with a lifetime warantee -- and no cooling aside from the ubiquitous fans (they take the cartrige apart and mount a heat sink directly on the chip).
-Billy
Hrm, this story was posted prematurely. There
are only a few benchmarks that compare a CuMine
700+ against an Athlon 700, with the Athlon
coming out 6% faster in some obscure business
test so far (see http://www.jc-news.com/pc/).
You will go nuts if you try to follow all the
lastest developments early in this CuMine vs.
Athlon game. My advice: don't. Wait for some
conclusive benchmarks.
Pro Intel:
* good compatibility with CPUs
* availability
* good o/c ability
Con Intel:
* multiplier locked
* core maxed out earlier @.22u than AMD's
* price/performance usually only average
Pro AMD:
* excellent CPU core, fast FPU
* EV6 better than GTL+
* excellent o/c ability, multiplier not locked
Con AMD:
* availability
* bad bad chipset and mobo situation
* bugs in chipset, slower than need be,
high current probs w/ TNT2U etc.
* s/w compatibility problems with chipset
* did I mention chipsets already?
Enough to drive everybody but the die-hard
hardware freak nuts...
WAIT and RELAX.
Everyone posting messages along these lines is forgetting something. The Athlon is faster than the regular P3, and a coppermine P3 is modified. No, it's not just cooler, the cache speed has been raised to full core speed (like a celeron). Unfortunately, the cache size has been halved (512K to 256K). The faster speed of the cache should more than compensate for the shrinkage though. Why do I think this? Well, celerons perform within 1-2% of P2's in pretty much any real world situation, and they have 1/4 of the cache running at full speed. So to compare the coppermine p3s to the athlons assuming they are the same as the regular p3s is just wrong. I'm not saying which would be faster (I can't say that for sure *yet*, and neither can you). One thing I have seen though are the benchmarks circulating on the web of these chips. They generally show a screenshot of a CPUID program and the chips being tested always report *no* cache at all. I don't think that these chips being benchmarked have the cache enables, and if they don't, well, you can't compare with those either.
Pipelined internal caches have issue times at least 8 time as fast as the sort of external caches present on the K7 and pII/pIII. Internal L1 caches tend to have latencies either 2 (K6)to 3 (K7 & Pii/Piii) core cycles long. Internal L2 caches tend to have latencies from 6 to 12 (core) cycles long. External L2 caches tend to have latencies from 8 to 30 core cyles long (about 20 on both the K7 and Pii/Piii, although it will be slightly worse on the 750 mhrz k7).
The CuMines cache should be about the same overall speed as the K7s cache. It's FPU will still be a ways behind though. The CuMine is a lot cheaper to manufacture.
Very informative post, thank you.
One impression I always have is that intel chips are
constantly underrated. Yes, the architecture is old, yes there
are some chips like alpha and g4 which are faster under some
circumstances.
But theres one thing which some friends of mine, who do
numbercrunching on clusters allways tell me, that is the
compiler aspect.
For instance they say the alpha looks in fact very good
theoretically, but they also say that the compilers job
in optimizing for the alpha is harder than for ia32.
The real-world performance of ia32 seems to be nearer
at the theoretical (benchmark) performance. I have seen
some tests in magazines which tell me the same.
And even on the intel plattform theres a huge difference
depending on what compiler one uses, i.e. ms VC on nt was
up to 30 % faster than gcc on linux. That was an older
benchmark, newer ones I saw seem to indicate gcc has
closed the gap or is even faster than ms compiler.
IMO this numbers indicate cleary that comparing prozessor
architecures on the basis of theoretical or artifical (benchmarks)
aspects is a bit naive, because compiler builders also might
want to optimize their product to handle benchmarks well, and
that doesnt describe the real world.
Urg. Sorry, but I'm just gonna repost the comment I made about a week ago when some other wierdo spouted the same ignorant blather.
Oh wait: it wasn't just any wierdo. Believe it or not, it was you, Millennium. Perhaps you could read what people respond to your comments. After all, saying something incorrect once just makes you look uninformed; saying the same thing after you've been publicly corrected makes you look willfully stupid...
--begin repost--
They're already at price/performance parity, more or less. The G4 is roughly triple the speed of a P3, and this is using Intel's own benchmarks, mind you. We're not talking Bytemarks here, boys and girls, we're talking benchmarks no one dares discredit.
Oh lord. Another one falls for the Apple FUD.
No, we're not talking Bytemarks here; this one, if you can believe it, is even worse. You see, at least Bytemarks is a benchmark. It's about 10 years old and has absolutely no bearing whatsoever on the performance of a modern CPU, but at least when they came up with it, someone was trying to get an idea of how fast a chip would run.
These 6 tests are not benchmarks, in any normal sense of the word. Benchmarks measure how long it takes for a computer to perform a real-world task. These tests (Apple's got 'em posted here; scroll to the bottom) measure the speed of individual ops.
That's right: the G4 performs 6 specific operations an average of 3 times faster than a P3. We're talking things with names like "1024 dim. DotProd" and "256 Pt. Complex FFT". The G4 can take a dot product 3.68 times faster than a P3. Oh wait--not even that; a dot product in a specific dimension. Whoopdee. A 128-bit unit can do operations on very large numbers faster than a 32-bit one. Wow. This is like posting the fact that a 64-bit CPU can add two 64-bit numbers faster than a 32-bit one. Who would have thought.
And yes, these benchmarks were "published on Intel's own website." Of course they were. In the technical specs on the SSE core. Deep in the technical specs on the SSE core. Where information that is completely useless to anyone not planning on optimizing a compiler belongs.
Essentially, this benchmark is as misleading as quoting MFLOPS (oh yeah: Apple stooped to that one too...). Except that usually when you quote MFLOPS you at least generally need to average over the entire set of floating point ops. Not here folks. They picked out their favorite 6.
Oh wait--here's another difference: when you quote MFLOPS, you actually need to, uh, benchmark the thing. These numbers are all theoretical--just compare the number of clock cycles it takes to do an operation, and multiply by MHz. Now, it turns out they'd probably be even more in the G4's favor in practice--if I remember correctly, the AltiVec unit has a much better designed pipeline than the P3's SSE unit. But still, these numbers are absolutely, completely, worthless.
I don't have the URL offhand, but I've seen the Intel page they copied these tests from, and there were literally hundreds for them to choose from.
The point is, you can always find an operation that is carried out in less clock cycles on one particular archicture as compared to another. Always. Now it turns out that, in this case, the AltiVec apparently really is vastly superior for the sorts of things it does when compared to Intel's SSE or AMD's 3DNow. (Of course, it also takes up half the chip. Any guesses as to why they can't fab any 500's??)
However, the fact is that except for very specific applications (SETI@home in particular, and some signal processing stuff, IIRC), it doesn't make all too much of a difference. A 700 MHz Athlon will smoke a G4 450 or 500 or whatever on your basic integer stuff, and a 600 MHz P3'll be right up there with it. For the stuff that can be done with AltiVec, the G4'll certainly come out ahead, but for general floating point work, again, they're about equal. It goes without saying that, at this point, nothing crunches graphics like a year-old PC with an NVIDIA GeForce in it (except maybe something from sgi)--which, of course, is about the only thing the average user needs good float performance for anyways.
In the end, the G4 is just a decent chip with a neat vector processor that's proving hard to fab. Is it damn fast? Yes. Is your new G4 450 going to touch the Coppermine P3 733 that's shipping by the time yours actually ships? Nope. Is it "two or three years ahead of its time" like Stevie says? No way.
-Dave
P.S. And yes, you can sell them to China as well. As much as I want to like Apple these days (a simplified vertically integrated product line is a very good idea in many cases; OS X just might be incredible; and geez--did you check out the new iMac subwoofer??), the fact that every single word out of their marketing department/CEO's lips is a baldfaced lie...gives me pause.
You are correct about the statement that a higher frequency means higher current onsumption (the eternal problem of handhelds and laptops: how to have better performance but less consumption) but I argue that upping the voltage would really solve the voltage-drop problem. A real solution would be to decrease the internal impendance of the power source. If you studied electrical engineering, you would remember Tevenin's theorem, which would remind you that a power source can be represented as a serial connection (ok, I don't know all this stuff in english language) of an ideal voltage generator, and an impendance (called "internal" impendance of the voltage source). In practical life, a better power source would be equivalent to an ideal voltage generator with a smaller impendance. It's technically very simple to have better power sourcces, you effectively have to inrease the power of the high frequency MOSFETs of the switching-mode power generator and/or the output linear circuitry of the on-board voltage regulator. I believe the guys at kryotech have modified the motherboard accordingly.
Sigged!
Washington(AP): In a move that suprised people everywhere, Intel (Nasdaq: INTC) has made an agreement with the National Air and Space Administration to fund the next 5 years of space programs. The agreement, valued somewhere between $500,000,000 and $750,000,000 calls for renovations to the space shuttles and the space program itself. Each of the Space Shuttle's control computers will be refitted with Intel Pentium III Xeon processors. In order to pay for this, NASA has let Intel paint "Intel Inside: Pentium III Xeon" in large letters on the Space Shuttles. Additionally, a new design of a sattelite, being co-designed by Intel and Nasa at Kennedy Space Center, will help to illuminate the night sky during new moons. Instead of a moon, there will be a square banner that reads "Planet Earth: Intel Inside".
At the time of press, neither NASA nor Intel could be reached for comment.
(Note: the above paragraph contains many mechanical and scientific inaccuracies. For example, no computer really needs that much power.)
Gates' Law: Every 18 months, the speed of software halves.
Analysts said the launch of the new processors is important because it allows Intel to again say that it has the fastest PC chips.
Um... No it does not, It mearly means that intel Chips can at least run as fast as AMDs The program speed will still be less however, since Athlons are still faster clock/clock then p6 cores.
I'd be willing to bet that these 'analysts' actualy work for intel, Its sad that wired news would print things like that as fact. It really makes you wonder...
"Subtle mind control? Why do all these HTML buttons say 'Submit' ?"
ReadThe ReflectionEngine, a cyberpunk style n
line of processors is arguably one of the most successful computing architectures ever. Many people bitch (especially here) that its dying and that PPC and IA-64 and blah blah blah will overrun it. Coppermines and Athlons both are an example of how strong the x86 architecture is right now. Think about it, look back five years ago and ask yourself what companies were running their mission critical systems on. It wasnt 486es or Pentiums (not until Pentium Pro was released), comapnies ran Sun, SGI, Alpha, but not many Pentiums. Now look anywhere, your favorite flavour Pentium is probably on some server in your office (but then again maybe it isn't). The x86 line in many ways has allowed smaller companies to have fast servers and workstations and gives them a more competitive edge in the marketplace of the web. Not everyone can afford Big Iron but they can put up some cash for a comparitively cheap Pentium based system which suits their needs. Pentium will eventually die out and I'll be ok with that, I don't worship Intel or anyone else but I think they do deserve some credit on what they've done in the past ten years.
I'm a loner Dottie, a Rebel.
Go read the documentation on the L2 caches. The Coppermine's L2 cache will run at 1/1, ie 700 Mhz. The Athalon doesn't scale 1/1 up to those speeds. At 700Mhz the Athalon's L2 cache is only running at 700/2=350 Mhz. (moving on up it will move to 1/3, and 1/4) So real-world peformance the coppermine will start catching up quick.
Looky what I found...
cascade
Pronunciation: (")kas-'kAd
Function: noun
Etymology: French, from Italian cascata, from cascare to fall, from (assumed) Vulgar Latin casicare, from Latin casus fall
1 : a steep usually small fall of water; especially : one of a series
2 a : something arranged or occurring in a series or in a succession of stages so that each stage derives from or acts upon the product of the preceding [blood clotting involves a biochemical cascade]
2 b : a fall of material (as lace) that hangs in a zigzag line
3 : something falling or rushing forth in quantity [a cascade of sound] [a cascade of events]
Intolerant people should be shot.
Um, no. First of all, a the voltage regulator will maintain a constant voltage to the chip regardless of current consumption ( within some bounds). Hence the name "voltage regulator." The max clock speed for a chip is derived from the time it takes to swich the state of one of its transistors. This switching time arises from the charging and discharging of the inherently capacitive traces and mosfet gates. Slapping more voltage across the chip charges these caps faster, reducing the switching time. I refer you to _Principles_of_CMOS_VLSI_Design_ by Neil H. E. Weste and Kamran Eshraghian for further details.
Fastest? Not quite Bubba.
I am typing this on a commercially released Athlon 700MHz. It has been on the market for a month.
And, at the same clockspeeds the Athlon has already demonstrated significantly faster integer performance, and massively better floating point speeds.
Intel will have to hit at least 800MHz to best the Athlon at Integer, and an entirely new design to get back in the floating point game.
Maurice W. Hilarius Voice: (778) 347-9907
Slightly off-topic, the new PIII Xeon are codenamed "Cascades." Is this pronounced kass-kades or kass-kad-ees or something entirely different?