Slashdot Mirror
Pentium IV study
InQuest has published a study of the Pentium IV. It
found that because of the P4's cache design, it uses 4 times the memory
bandwith a PIII does on random access data benchmarks. Even on benchmarks which
benefit from the larger cache-line size, the P4 shows no benefit due to its
higher clock rate. Furthermore, the 1.5GHz P4's thermal diode throttles the
part to effectively 750Mhz as soon as power consumption exceeds 54.7W. Without
this limitation, a P4's maximum consumption would be 72.9 watts, similar to a
1.33 GHz Athlon's 73 watts.
- A.P.
--
Forget Napster. Why not really break the law?
"Remember when the U.S. had a drug problem, and then we declared a War On Drugs, and now you can't buy drugs anymore?"
Oh, but I do appreciate thermal management in my buying decisions. That's why I have a G3 upgrade card in my PowerMac :) cute little (no fan) heatsink on it, much smaller and tidier than the huge (no fan) heatsink on the old 604e :)
I realize this is dramatically off topic, but in the early days of portable transistor radios (early '60s - ancient history), one measure of the quality of the radio was the number of transistors used. This lasted exactly as long as it took for some sly company to start mounting dead transistors in the radio (not even bothering to connect them to the rest of the circuits), just so they could up the transistor count and therefore be counted as a "higher quality" radio.
...phil
...phil
"For a list of the ways which technology has failed to improve our quality of life, press 3."
You guys seem to like throwing out numbers. Could you do us all a favor and give us some links or at least explain where you managed to come by these numbers?
It's not enough to bash in heads, you've got to bash in minds. - Captain Hammer
Not nearly as much as far as I can tell. Then again it doesn't need as much.
To run IA64 code you need a IA64 OS. To run 64 bit AMD code you need a minor change to the context switching code (you need to save 64 bits of data per GPR, and twice as many GPRs). Of corse you need more OS support for a 64 bit address space.
To produce good code for the IA64 you need some extremely advanced compiler technology. Unrolling loops isn't enough, you have to modulo schedule them. Good AMD64 code is hardly different from normal x86 code.
Now there are IA64 OSes, and at least a compiler or two. Intel did get something for their hype machine efforts, and for having some machines around for people to test on. I don't think the AMD 64 bit stuff is as far along, but it won't take long to catch up.
Besides a lot of the IA64 mind share is folks calling it the "iTanic", AMD may not want that kind of mind share :-)
I don't think anything in the IA64 is as useful as SMT (as seen in the next gen Alpha, and rumored to be in the next IBM Power CPU). Pretty much everything in the IA64 has been seen before BTW. I do admit it is more innovative then the 64 bit AMD CPU though, but not all innovations are worth the price of buy in...
Of corse two years ago I felt differently about the IA64 (and I have a stack of IA64 docs to show for it), and in two years I may change my mind again...
If you read Tom's Hardware fairly regularly you should know that the Pentium IV is not up to snuff yet. I think that, just like the Pentium Pro, the Pentium IV is just the first iteration of what will eventually be a great CPU.
Of course, I'm really rooting for AMD here, they're pushing Intel to its limits and the consumer is benefitting.
pw:secret
Thing is though, the increase you are seeing is due to Rambus memory, not the CPU. And there are *very* few applications where higher bandwidth of Rambus matters. In 99% of applications the, the higher latency of Rambus totally negates any performance gain you might get due to higher bandwidth.
Oh, btw, redo your benchmark taking price/performance into account...
___
___
If you think big enough, you'll never have to do it.
Uhhm, please pull your head out of your ass and re-read my comment. Specifically I said that the performance increase has absolutely nothing to do with the CPU. It is due solely to the Rambus memory. For that matter, if you could put Rambus memory into an Athlon board, you'd see the same increase. In fact, the Athlon system would be even faster than the P4. And speaking of FPU -- it has been shown to be abyssmal in P4. Even P3 outperforms P4 on FPU!
___
___
If you think big enough, you'll never have to do it.
It's too bad these benchmarks are useless, as 99% of applications don't require anywhere near as much bandwidth, and are in fact hurt by Rambus's high latency. The one exception, as you pointed out in your other comment, is scientific computing. Other, more realistic benchmarks showed a different picture. I especially liked the ones where 1GHz P3 beat 1.5GHz P4 :-)
BTW, do you work for Intel?
___
___
If you think big enough, you'll never have to do it.
You know it's bad enough that you can't use AMD or Intel in the same motherboard anymore, and I wonder if the hardware companies are happy about that or not, but now it's getting to where upgrading your motherboard with a better CPU is going to be an exercise in futility. I wonder how soon we're going to see motherboard, power supply and case form factors only compatible with certain CPUs from certain companies, not to mention all the overtime work they're going to do to make your current RAM useless.
Anybody want to buy some 5volt DIMMs?
I see even classic Slashdot is now pretty much unusable on dial up anymore.
Intel to cut up to 60 per cent off P4 prices
I see even classic Slashdot is now pretty much unusable on dial up anymore.
So, we need to factor in the price of additional heatsinks, fans, and peltier coolers when we do price comparisons.
Yes, and you also need to factor in the 50% drop in clock speed the P4 makes to avoid needing that fan when you do price/performance comparisons.
Though what I really wanted to see was benchmarks demonstrating this effect. As in -- run UT for a while, watch temperature rise, watch framerate drop when temp hits 55 degrees.
The enemies of Democracy are
But eventually the x86 architecture will run out of steam
Okay, I've said this before, I'll say it again, but no more.
x86 was supposed to 'run out of steam' 10 years ago when the RISC revolution started then died. It didn't. RISC lost, CISC won.
But then again, RISC won, because CISC only won by keeping its external form but becoming RISC on the inside. Every x86 chip since the PPro has had a RISC core with an x86 decoder front-end. Thus the limitations of x86 as an ISA have been decoupled from the performance of the chip.
There is no reason for x86 to 'run out of steam'. There is no inherent limitation in the ISA that prevents faster processing. Yes, the decoders are large and complicated, but Intel and AMD have a handle on that. The P4's trace cache is a beautiful example of getting x86 entirely out of the critical path.
Given that, why are both AMD and Intel making 64-bit ISA's instead of sticking with ol' IA-32? Because servers need bigger address spaces for things like terrabyte databases. That's all. The only motivating reason.
AMD won't ever be making IA-64 machines. Why? They don't have a license, and intel sure as hell isn't going to give them one. I suspect this is a major reason why Intel isn't making their own 64-bit extension to x86.
Lastly, while IA-64 is both VLIW and RISC, it does involve new technology not encompased by either term, so the new phrase 'EPIC' is suitable. Not that I think it is a good idea in the first place.
The enemies of Democracy are
The number of registers is being increased to a healthy 16 by AMD. I never said that the ISA shouldn't advance, I said it needn't be scrapped due to inherent limitations. Though the lack of registers can be largely eliminated by having a 1-cycle access l1 d-cache.
The floating point architecture is the same as the rest of the ISA - an illusion provided to the external world, re-arranged into something sane by the decoders. This is why the Athlon's FP performance rocks so much.
Sure, having this baggage isn't _good_, but it isn't bad either, outside of the extra die area it takes up. But what having that baggage does let you do is take advantage of the dearth of x86 software out there. That's the whole reason it's still around.
The enemies of Democracy are
Yabbut, VA Linux is working on new machines using two AMD CPUs and the AMD 760MP chipset. The news has been on several hardware sites and VA employees have been posting to linux-kernel their patches to make the AMD 760 MP work well in Linux.
For those who didn't click the link above, let me iterate it here: [why the P4 is a dog]. It's a well-written, accessible technical article that documents the development of x86 generations, and describes why the design choices in the P4 are sub-optimal (and why design choices in the K7 family have been optimal).
It's a pretty factual examination of things. Worth the read -- it might save you from pissing your money away on something that's not any good.
--
--
Don't like it? Respond with words, not karma.
Intel has a market cap of 24x AMD, actually. INTC has 6.7B outstanding shares; AMD has only 314M.
:-)
But because Intel has more shares outstanding than there are greenbacks in North America, it has essentially devalued its share value:
AMD: basic income of $3.25 per share.
INTC: basic income of $1.57 per share.
Put quite simply, AMD's share is providing a better bang for your buck. AMD's share should, if one were to use Intel as the benchmark, be priced at $150...
Guess I should buy some.
--
--
Don't like it? Respond with words, not karma.
Why would you say Intel is likely to dominate the platform?
Fact: AMD now holds twenty- to thirty-percent marketshare.
Fact: AMD Duron and Athlon processors are now spec'd by all but one mainstream, brand name supplier.
Fact: The latest AMD processors outperform -- and are widely reported as such -- the latest Intel processors.
Fact: AMD K6-2 mobile processors have a 20% marketshare, and AMD K7-family processors are gaining marketshare.
Fact: Intel keeps fucking things up. And I mean *seriously* fucking up. From backing Rambus to failing chipsets to bolloxing up its relationship with their customers... oh, gahd, the list is endless.
So, how do you foresee Intel dominating the platform war? What I foresee is a an ungainly beast of a has-been chip design and manufacturing company, shooting itself in the foot and chopping its own legs off, while an upstart executes cleanly and quickly, soon to overtake it...
--
--
Don't like it? Respond with words, not karma.
Why would you put a fuse or circuit breaker in a house? Why not just draw less current?
By an amazing coincidence, I was just reading this article when this story came up on /.
If tits were wings it'd be flying around.
Actually, one big problem about Pentium 4 systems is the fact you need a completely NEW system case design and a new power supply with a totally new type of motherbord power connector--cheap it isn't!
Pentium III and Athlon systems can still get by with standard ATX cases, provided you have a 300 watt power supply and decent system case cooling.
It'll be a while before Pentium 4 systems become common.
Raymond in Mountain View, CA
I built my P4 system into a two-year old generic ATX case, which previously had a 100 MHz Pentium in it. The only thing I needed to change was the power supply (which you likely would need to do if you got an Athlon also, since it has such high power requirements).
:(
If you're using an Intel-brand motheboard for the Pentium 4, sticking it in a regular ATX case ain't going to work.
The problem is that the Intel P4 boards require a special motherboard mount, and the Intel boards also use a new-style power supply connector, too. That's why they're not cheap.
Raymond in Mountain View, CA
The P4 is slower running current apps, but if you re-compile and optimize for the P4 it runs much better. Too bad it'll be years before there is enough software optimized for the P4 to make it worth buying, and by that time there will be other CPUs.
I do agree that for current operating systems and applications, they are not written to take advantage of the architecture of the Pentium 4 CPU.
Yet, remember what I said about the Pentium Pro from six years ago: refinements to the CPU core pioneered by the Pentium Pro led to much better improvements in 16-bit app speeds, starting with the Pentium II, then Celeron, then Pentium III. Remember when the Pentium II first came out operating systems such as Windows 95 OEM Service Release 2 recognized it as a Pentium Pro CPU.
I expect the second-generation Pentium 4's due this fall to be a bit faster as they will "tweak" the Pentium 4 CPU core for faster performance.
Raymond in Mountain View, CA
SSE2 will become popular, because even AMD has licensed it. If you buy an Athlon now, you may accomplish more FPS in 3D games (amazingly enough, the P4 still wins clock-for-clock in Quake III framerates compared w/ the Athlon). If you get a P4 (I'd wait for the new socket version ones, though), you'll have SSE2 support and will most likely get a performance boost in FUTURE apps, when it'll actually press your hardware.
However, unless it's a high-end game, I don't see any real apps coming soon that will truly take advantage of SSE2 (well, maybe Adobe Illustrator and PhotoShop, but who knows when will fully SSE2-supported versions will be available).
Besides, remember that AMD plans to have the third-generation Athlon based on the Palomino core coming this fall. I'm sure AMD will make many improvements in order to keep up with the Pentium 4 in regards to integer and FPU speed.
Raymond in Mountain View, CA
I still think it is still too early to pass any judgements on the Pentium 4 CPU.
The reason is simple: we are repeating history in terms of Intel CPU architecture development.
I think many of you remember the Pentium Pro CPU, which came out in 1995. While it was very fast for its day in terms of 32-bit applications, it was a bit poor for 16-bit applications. Yet, the P6 CPU core that the Pentium Pro pioneered became the basis after numerous refinements for the Pentium II/III and Celeron CPU's, which run very quickly with both 16-bit and 32-bit applications and was not matched until the AMD Athlon came out in 1999.
As it stands, the Pentium 4 CPU core design--which is brand new in many aspects--has only begun its development curve. I expect dramatic improvements in performance as this new core design is improved over the next few years.
Raymond in Mountain View, CA
Correct. And the real money corporate customers buy Intel primarily because
1) Intel has better OEM vendors lined up for their market (Dell/Compaq/IBM). This channel is the key. AMD primarily gets what's left (home/soho/BYOB).
2) Intel has a much better reputation. Little things like that "What chipset bugs?" scene when the K7 was launched don't help. The key here is that if Intel tries to force RAMBUST or buggy shit like the i820 down Mr. MIS's throat they are dead in the water, so they are dancing on pinheads.
3) The corporate market could give a shit which processor can do 10fps better in Quake or if some CPU is slightly suboptimal at running legacy code (which after all was designed to run much slower computers to begin with). They are simply looking for a price/performance/supportability sweet spot that they can standardize on.
--
Business. Numbers. Money. People. Computer World.
- they pointed out the fact that P4 performance is no greater than lower-clocked Athlon. This is true. However, with Northwood P4 should be scaling to 2G and above where it will eventually beat Athlon. (Of course, AMD is not sitting still, but P4 is supposed to have more headroom than Athlon due to its superlong pipes.)
- They criticize the power throttling; this would be a moot issue after the processor is shrunk, at least for the lower end P4s.
- They point out that nobody wants to buy the current P4 due to its dead-end packaging. This is also true, but presumably the new socket for Northwood will be around for a while. (I have no idea why Intel thought to introduce the 850 while admitting that it would be dropped in less than a year, but that's what they did.)
They even acknowledge the price drop. AMD competed for years with inferior processors priced low. Intel can certainly compete with Northwood.Anyone care to comment on this seeming discrepancy?
Assuming that it really is thermal throttling, I would love to see what a good tech site like Tom's might be able to determine about the throttled down CPU when using various heatsinks. If that feature is really there then you should expect more powerful heatsinks give the same temperature as lesser heatsinks, but higher performance.
In other words, it is possible to see this as a feature, not a bug. You get 1.5G when the processor is capable of it. You get half that when you are running hot; but with good enough cooling you should always get the highest performance possible.
"Overclocking" may go away, replaced by "overcooling".
Otherwise, why would the sub $400 USD computers be so popular? Show a consumer a 1.2 GHz AMD based computer and a 1.5 GHz computer, at a even $200 price difference and expect to see the AMD win.
Nobody cares. Consumers just see "1.5ghz!!!!" and say "thats fast!!". These studies and all are preaching to the choir. We already know that the p4 isn't clock-for-clock as fast as the thunderbird. Sigh..
This has been a problem from the get-go...there was a recent slashdot article awhile ago (Like 4 months back) where the guy details all the problems that the Pentium IV is going to have. Basically, it sounds like they sacrificed all the gains of the last 10+ years for raw clock speed. Now, it seems, it is coming back to bite them in the ass. Too bad. Unlike MickeySoft, Intel actually had some competition, and managed to stay afloat. Looks like the tides gonna be 'aturnin'.
JoeLinux
The world is coming to an end. Please log off.
The K6-2 was very crippled by the chipsets out for it at the time. It wasn't a stellar performer on its own, but the chipsets would have held anything back.
I played with an ASUS P5A (Super 7, Ali chipset) at work recently when I put a 233MMX in. What a dog. Slower than the 166MMX next to it on a XP55-T2P4. (A bunch of data entry terminals, speed not terribly important.)
You need special IDE drivers, AGP Drivers, etc. It's like a Via chipset, except that Via actually makes decent drivers. Maybe Ali and ATI are the same company? They both make hardware that should be good, yet prevent anyone from ever using it properly by withholding stable drivers.
But I've also seen a K6-2 do very well on some real-world things... I think it's largely the mobo.
Really? You make over $100 an hour? Wow!
The OEMs always charge $100 - $250 more for a system than it could be built for, with quality parts, if you did it yourself.
And maybe I'm just really fast, but I've rarely had a new PC take more than an hour to assemble from parts.
You could be talking about how they come with the OS pre-installed or something, but I've rarely seen a business that's used the pre-installed OS. Everyone I've worked for has wiped it and reinstalled something else. They've also usually had something like Norton's Ghost to automate this procedure. (Not like installing takes very long, base Win2k is a 30 minute install, Win98 takes 15, and *nobody* uses WinME...)
But, you go ahead and pay those companies an extra couple hundred dollars, and accept the low quality parts you're likely to get. (Dell and Compaq both have custom mobos that are trash, and tend to ship with the slowest HD and Video you can buy, just to save a buck.)
I'd rather spend a few hours on research (not too much, because I keep caught up for my own purchases) and then specify, to the specific parts, the exact computer I want. And then get a local store to send me exactly that. Hell, assembly is free on a full system, so there's that much less to do.
It's nice, knowing you've got quality parts that you'll be able to find a BIOS upgrade for, or replace with off-the-shelf components in the future. Nicer yet when you realize it was much cheaper.
We got in a single p4 1.5 Ghz machine with 512 megs of RDRAM along with a dual 1 ghz machine with a like amount and kind of ram. We ran some benchmarks. Actually, we ran real applications that the scientists at our lab run (chemistry codes).
The short of the story is this...the p4 plastered the p3, the athlon, and everything else, save for an alpha, which it gave a run for the money.
The first test we did was a large matrix inversion in octave. Same version, same kernels, etc. Same disks (scsi/160). The p4 ran it in 1/3 the time of the p3, with only 1/2 again as many ghz. Same speed ram. That memory bus mooooves things. Boy does it move.
We ran Gaussian on it. We saw a linear increase in relation to processor speed. This may seem "normal" but it is not. You almost never see that in this particular application.
We ran lmbench on it vs a ultrasparcIII. It embarassed the US3 so badly that we felt that it just wasn't a fair benchmark (gotta find something better for the suns).
I like the p4. It'll be a win for scientific ocmputing, regardless of what the consumer market thinks.
-- Who is the bigger fool? The fool or the fool who follows him? --
So Intel released a beta processor, a prerelease effort. The P4's architecture has many good points, but only above 2GHz. I am ignoring the CPU-protect feature (halving the speed when the CPU gets too hot - this is a cooler issue, not a CPU issue, I would like my CPU to protect itself! However Intel should have written that their CPU dumps 73W max at 1.5GHz).
What I would hope for is an optimised P4 later this year, i.e., the real release version. At 0.13u, with more L2 cache, etc, the P4 will actually start to be a better platform. Shame that they are going to couple it will PC100 SDRAM with the first SDRAM chipset, but then again, maybe they have fixed that bandwidth hogging problem...
Still, the extra problems are not doing anything for Intels reputation. IT managers are starting to notice that there is more beyond Intel, Athlons are starting to ship to corporates instead of PIIIs. Integrated motherboards for AMD processors are starting to appear by the bucket load (KM133, SiS730, etc).
So, AMD for me this year. Can't wait for summer and beyond - Dual AMD processor action! Now, what do I need one of those for?
. Furthermore, the 1.5GHz P4's thermal diode throttles the part to effectively 750Mhz as soon as power consumption exceeds 54.7W. Without this limitation, a P4's maximum consumption would be 72.9 watts, similar to a 1.33 GHz Athlon's 73 watts.
Yep, but I'd rather have a chip slow down a bit when overheating than melt like Athlons do. AMD managed to make use power hungry/heat generator chips, yet didn't even think about emebedding a thermal diode and protection like all P3 and P4 have. Results ? Lots of burned Athlon at 1 Ghz and more.
Since it came out I've been advising friends and coworkers to stick with the III's or Athlons until the IV gets its act together. Nice to see I wasn't blowing smoke.
But the sheer percentage of these people who were going to buy a P IV if I hadn't said something is a testament to Intel's strategy. People are going to buy it just because of the name.
InQuest is a paid "Marketing Consulting" company. The article is a crock, nicely documented here:
p hp ?message_id=30014614
http://www.aceshardware.com/board/general/read.
---
It's all well and good that the Pentium IV has some design "issues", and I think many of us here at /. expected this. Unfortunately, much of the purchasing decisions in the real world are based on sheer numbers: Dollars, Gigahertz, or both.
Look at the success of Microsoft Dos and Windows. There were certainly better alternatives out at the time for everything DOS and Windows did (PC-DOS, DR-DOS, even Macs). Microsoft primarily won the OS and Apps market because of its hugely successful marketting push. The only thing the average person heard in conjunction with "Computers" became "Microsoft".
Or consider America Online. There were many ISPs before them (remember netcom?), but the veritable hailstorm of "Free" floppies and CDs bought AOL the lion's share of Internet Service Provision.
The fact is that unless the Pentium has a serious flaw in it (fdiv, F00F), it will do reasonably well just because it has "1.5ghz". And as we all know, that more gigahertz must mean better technology!
In future news... Watch as Intel attaches a Ring Oscillator to their P5 chip running at 10ghz, unused by the rest of the chip.
-Ted
I don't have the direct link for this, but in order for Intel to reach 2 Ghz they are going to have to change the Socket design again (something like Socket 473). There are not enough pins to reach 2 Ghz. I am waiting for the intel processor that needs a separate case and a heatsink the size of my house.
"Imagination is the only weapon in the war against reality." -Jules de Gautier
I think we are seeing the end of 'upgrading' your computer.
Things are moving so fast, becoming so cheap...that by the time you want a new computer,, you might aswell just replace the whole damned thing!
Its only the true hardware junkies who spend the money to get the latest and greatest hardware every week.
"Imagination is the only weapon in the war against reality." -Jules de Gautier
For the last few years the Mac hardware has been pretty cool...and ever since OSX, its started to look really promising for Macintosh...I hjope they don't screw it up.
"Imagination is the only weapon in the war against reality." -Jules de Gautier
Intel... lure the mainstream... just plain bad... lost market share and customer respect... unwise... poor... many new alternatives... not...thrilled.
Intel...predicted...fastest... shipments...below expectations... bad omen.
The market...will not pay a premium... Intel is hoping... frustration.
Intel...criticizing AMD...Athlon. P3...well supported.
But now...instability of the P4 platform...P4 lacks...market...confidence.
P4...will undergo...five major changes...
Users...aware...weaknesses... Intel's...problem is the platform... band-aid.
Brookdale...difficult...DDR...competition.
Intel...frustrated...cripple... wide open for all...competitors.
P4 needs...strategy... DDR... escapes...Intel's marketing folks... familiar territory.
Intel...deny...problem... vigorous defense of Rambus.
elaborate thermal and power regulation requirements... toaster oven.
Intel... publicizing... amazement... only 54.7 Watts... below the fastest Athlon.
Pentium 4... unbalanced... weakest.
Intel's... elevated costs... horrific... price slashing.
--
314-15-9265
Well, AMD has already thought of this. In case you haven't noticed, AMD is no longer in the business of catching up with Intel. They are poised to take a bite out of Intel's ass.
Need evidence? Okay. The reasons Intel became so wildly successful in the processor market is due to a number of factors. But I think the most obvious one should be that they did what no one else had done at the time; they made their next-gen processors backward-compatible. This, of course, elated consumers because it meant that they didn't have to throw away all of their familiar software (and sometimes hardware) to get the performance boost that a new CPU would offer.
Now, fast forward to the late 90's. Intel dominates the market, but sees a threat, so it does what nearly every company does when in fear of seeing their previously solid sales figures start to dwindle due to competition. They change their product to make it incompatible with their old product line, but most importantly their competitor's product line. Then then sit and pray that brand recognition will propel them through the transition.
In the current case of AMD and Intel, this may or may not work. But AMD has a distinct advantage. They didn't wait for Intel to make their move first, but set to work on their next major cashcow product. We know it as Sledgehammer: AMD's 64-bit chip to compete directly with Intel's. The advantage lies in that AMD's chip will be backwards compatible with the 32-bit x86 architecture, while Intel's will not. To see why this is an advantage, I direct you towards the first paragraph of this post.
(In Intel's defense, they claim that they will have an x86 emulator ready at the time their IA64 chip ships. How well this will work in comparison to backwards-compatible 32-bit hardware instructions remains to be seen...)
On comp.arch we were kicking this around a few weeks back, trying to access the average roundtrip latency of memory. Some of the same ideas for defeating the cache were employed. The truth is that these benchmarks are particularly stupid in that they test a random sequence of accesses, which, hello, defeats the purpose of a cache anyway! What exactly were they trying to prove? That memory is slow? Of course. We already knew that.
Why don't they mention something like bandwidth scores or something similar? Take a look at the PIV's scores in Sisoft Sandra (a widely used metric in the overclocking community). It absolutely crushes both the Pentium III and the Athlon in memory bandwidth.
Yes the PIV is slow. Yes it has an overly long pipeline. Yes the cachelines are big. And yes, random accesses are bad on caches. ALL CACHES.
120 characters isn't enough to explain it.
The P IV requires RAMBUS, much more cost with little or no performance benefit AND a manufacturer that is sersiously ethically challenged. (Witness the suits stating they patented ideas that came out of JEDEC meetings, which they are now using to extort..uh, extract license fees form other manufacturers.) Since Intel is limiting its cusomers's choices with P IV, I will limit my choice to non P IV processors ( and maybe to non-Intel processors).
--
"that's not encryption - it's a new perl script that I'm working on..." - from some Matrix parody
Hardly anybody cares about wasted memory bandwidth on a single processor machine, as long as it doesn't affect performance (and the data doesn't suggest this).
However, suppose you've got a SMP machine with four CPUs, I guess you'll get into big trouble. The Intel architectures are not really known for impressive I/O bandwidth, and if the memory subsystem has to deal with 1.6 GB/sec instead of 400 MB/sec, performance is probably affected.
So Intel should fix this rather quickly. May be this is thust an optimization issue, the 32 byte cache line is rather old and many programmers have adjusted their data structures accordingly because you can gain a few percents just by looking closely at the data access patterns. More recently, Intel has added explicit cache line prefetch instructions, and if these are used improperly (assuming 32 byte cache lines), the effect can be pretty devastating on the P4, I guess.
This is incorrect. Intel has 84% of the market, and AMD has 14%. Contrary to popular belief, Intel actually gained market share in 2000. The real market change is that Intel has almost completely captured the sub-$1000 market (it now has approximately 95% of it now, and had less than 50% in the 1998 timeframe). AMD has shifted away from the low-end into the mid-tier (where it has about 25% share - by far its largest share of any segment). This has been good for AMD because it has given it a higher ASP. Intel still has a virtual monopoly on high-end, X86 workstation, servers, and mobile (I don't know where you get 20% share for AMD in mobile; AFAIK it's less than 5%).
Intel's main strength is its manufacturing capacity. Intel has 7 or 8 fabs, and AMD has 2 (which is not enough to feed more than 20% of the market at current demand, which puts an upper limit on their market share). It would be in AMD's interest to continue to trend toward higher-end and leave the low end (but high volume) to Intel.
Interestingly, despite the fact that the Celeron is more than double the price of the equivalent Duron, Intel has a virtual monopoly on the sub-$1000 market, which makes me very seriously question AMD's marketing abilities. The Celeron is smaller (i.e. cheaper to manufacture), while the Duron has a relatively high opportunity cost (manufacturing Athlon's). Intel has been quietly cleaning up on this market (low margin, but very high volume), while AMD has almost complete lost it.
Intel's biggest weapon is its 0.13um process on 300mm wafers, which is coming early next year. This technology literally quadruples the capacity of a fab (AMD won't have 300mm until FAB35 is completed, which is projected to be in 2005). This means that one single Intel fab has the double the capacity of all of AMD, _and_ it lowers prices. Basically, Intel will be manufacturing tiny 2.5GHz Northwood's in tremendous quantity at a lower price than AMD's 1.5 GHz Athlon's (or whatever they will have in early 2002). I'd be pretty surprised if AMD didn't go back into negative earnings by Q3'01, and shocked if not by the Q1'02.
In other words, they're having problems moving them out, so they are going to slash prices.
So that's why AMD almost always cuts prices first?
If the P4 even has any SMP circuitry, what makes you think it was ever debugged by Intel? Not to mention the lack of SMP chipsets.
Well, for starters, according to almost all industry rumor sites (the register, amdzone, etc.), dual P4 is going to be launched within a matter of weeks.
I just hope there's some way of alerting the OS (or BIOS, or something) that you've had a heat related speed change. Otherwise, you might not realize that things are screwed.
Apparently, of the rich, by the rich, for the rich.
4 in roman numerals is IV. When the average consumer realizes he spent triple the money on something worth half as much, the paramedics may need to give him one of those. Cooincidence?
I am !amused.
You have to take anything that Bert McComas says regarding Intel with a grain of salt. He is probably one of their biggest critics and I've never heard him say anything positive about their technology.
I'll give you an example: He says that the chip will slow down when you try to extract useful performance from it. He bases this on his misunderstanding of the THERMTRIP circuitry. Yes, it is true that the P4 will dissapate more energy when you make it run a heavier load. Yes, it is true that if the die on the chip gets too hot, the chip will slow down to protect itself. But that completely ignores the fact that it is up to the system vendor to insure that the chip never gets too hot. Right now, I'm looking at a P4 system that I was a designer on. Not only do we have a fansink on the P4 itself, the chassis has four separate fans to dissapate heat, and one of them is a monster 5-incher. There is no way that one of our systems is going to overheat the P4 and cause it to slow down.
Unfortunately, customers don't seem to appreciate thermal management in their buying decisions. But keep in mind that all 1.5GHz P4 systems are not alike.
despite the fact that the Celeron is more than double the price of the equivalent Duron, Intel has a virtual monopoly on the sub-$1000 market, which makes me very seriously question AMD's marketing abilities.
Consumers probably think that "Duron" is a paint not a processor. Plus, AMD doesn't have the Blue Man Group doing cheesy commercials (QuickTime).
Will I retire or break 10K?
Try buying a MP AMD system from your local vendor. If a corporation needs SMP, they _must_ buy Intel (for x86 servers).
Recent Athlon processors use a bus protocol similar to that of Compaq Alpha processors. Somebody else wrote that VA Linux Systems (Slashdot's parent company) is building an Athlon-based SMP server and patching Linux to improve its SMP performance. But still, can you imagine...
Will I retire or break 10K?
OK, you have that "effective SMP" web cluster of maybe 100s of machines. Fine. Now what's your database server?
This is also an Athlon cluster. Specifically, an Oracle8i Enterprise Edition Parallel Server cluster. (MySQL doesn't seem to support clustering.)
Will I retire or break 10K?
As someone else mentioned (I don't know if is true) some cips will melt in 8 seconds without cooling. That gives VERY little time for a fan monitor (like the lmsensors for Linux or LanDesk for Windows) to detect the condition and initiate a shutdown. That is bad. If you poll every 30 seconds (common) you fry. If you poll every 1.5 seconds (fastest the system monitoring chips can be polled) you have 6.5 seconds left to initiate a power off. Forget about controlled shutdown. And if the machine is running a lot of processes or thrashing, the poweroff signal might not get sent it time.
A melted blob of silicon would be all the remained of your chip.
Ideally you'd have enough time to have the OS detect a failure, send a page or an email and do a controlled shutdown before burn up. Or allow it to survive at full speed with no fan long enough for the OS to detect it and cut the speed - at which point all the above (paging/email/shutdown) could be done at reduced speed.
Cooling failures happen. Fans stop. Cars have overheat protection modes on the engine controller, so should CPUs.
That said, a CPU should NOT go into its "overheat protection mode" or thermal throlling as part of normal operation. I.e. room under 100 F and even at 100% CPU load, if the fan is turning at least minimum spec'd speed, it should NOT overheat.
Simple as that.
Just because it CAN be done, doesn't mean it should!
Games are a prime example of intense periods of computation. So what I'd like to know is if anyone has noticed a sudden drop off in frame rates a few minutes into a game, possibly following by an alternating cycle of high and low performance priods?
Or is the cycling between full performance and throttled performance so small that it is perceived to average out to a constant FPS in a game?
I could imagine it being difficult to certainly attribute FPS changes to this, but I'm very curios to know if anyone has firsthand knowledge.
Thankfully when I bought my Athlon, the P4 cost about 5 or 6 times as much (when purchased w/lots of memory)
I can say one thing against the P4. The two people at work that got brand spanking new Dell w2k boxes with it weren't much impressed with it over their old 500MHz machines.
---
Overheating? It is called a heatsink and fan.
Thermal grease also helps, you know, the white/silver/copper shit that costs $2 an oz?
Mounting the friggin heatsinks right would also be a plus here. Using one meant for the processor would help too, i.e. not a heatsink for a pentium pro.
I suppose geniuses like you who burned up their athlons/tbirds are bitter.
AMD makes good products, you just have to know a few basic things about them, like how to properly install a processor.
Oh, the P4 throttles when it reaches a specific wattage, that means, even though it is running cool, safe temperature, it still throttles down.
oh.. and in case you didn't pass grade 3 math, "a bit" is not 1.5ghz to 750 mhz, thats "half", which isn't "a bit".
What the hell is with this intel fetish with some people? Intel, at the current time, makes sub standard stuff. AMD makes really great, and cheap products. Your choice should be clear.
And FYI, a p4 will cook just as good as an athlon if you do something stupid.
Fuckhead.
I have a shotgun, a shovel and 30 acres behind the barn.
1q2w3e4r5t6y7u8i9o0pqawsedrftgthyjukilo;p'azsxdcf
In other words, they're having problems moving them out, so they are going to slash prices.
Check out the Vinny the Vampire comic strip
"It is a greater offense to steal men's labor, than their clothes"
Why would you say Intel is likely to dominate the platform?
Fact: AMD now holds twenty- to thirty-percent marketshare.
Fact: AMD Duron and Athlon processors are now spec'd by all but one mainstream, brand name supplier.
Fact: The latest AMD processors outperform -- and are widely reported as such -- the latest Intel processors.
Fact: AMD K6-2 mobile processors have a 20% marketshare, and AMD K7-family processors are gaining marketshare.
Fact: Intel keeps fucking things up. And I mean *seriously* fucking up. From backing Rambus to failing chipsets to bolloxing up its relationship with their customers... oh, gahd, the list is endless.
Just because you, and every other computer guy/gal knows this, does not mean your average consumer (assumed to be reasonably informed -- we'll ignore the idiot consumer for now) also knows.
According to the average consumer, Intel is the only game in town. And don't tell me that they're stupid for thinking that, or anything equally naive. They are simply uninformed, and it is not economical to them to become properly informed. For them, anything being sold now will be grossly overpowered for their web surfing and email writing.
What I don't understand, is why companies don't sell decent computers that are low powered and quiet for web surfing. I mean a "real" computer, not a stupid internet terminal (that are only useful for haxoring and installing free OSes). Something like a Celeron 400, 128 MB RAM, and 20 GB drive.
That the Pentium IV is a marketing ploy by Intel to keep AMD from grabbing all the market share with their superior products.
Didn't we already know that?
The only plus side from my viewpoint is I was going to buy some 500 MHz Pentium II chips, which should drop in price.
What about the power consumption claims from Intel on the Pentium IV - anyone got any specs on that?
--- Will in Seattle - What are you doing to fight the War?
Note they say: "Perhaps Intel was a little frustrated that its own 815+PC133 platform beat RDRAM last year. This time Intel will not screw up - they will doubly cripple SDRAM to make certain that it cannot match RDRAM. In other words, Intel seems willing to publicly impale Northwood on an entirely inadequate Brookdale platform in order to make its point. It is this type of agenda that causes further doubt about Intel's commitment to the P4 platform and leaves the door wide open for all of Intel's competitors."
..."
...
And note the CPU rampup - yeah, it will run faster than a P3, so long as you don't ask it to run fast for very long. Kind of like a car that goes 0-60 mph in 0.1 seconds, but if you actually try to get to top speed of 125 mph, it will suddenly kick in the limiter and drop you back down to 115 mph. Sure, the car's rated to do 250 mph, but it can't actually run at 250 mph.
Translation - very sucky. Trying to fight AMD on "bandwidth" by faking it and also fight Transmeta on low-power by using a heat limiter (faking it). It saves power when demand gets high by cutting throughput, instead of Transmeta which saves power when demand is low, but lets you use full power when you need it.
"Scotty, I need more power!" "Sorry, Cap'n, but the speed limiter just kicked in - I can give you warp 1.3, or if you turn off the shields and the phasers, I can give you our full rating of warp 9. Of course, then we'll die anyway
Yup, sounds like Intel
--- Will in Seattle - What are you doing to fight the War?
Maybe you weren't around back in 1981 when IBM was advertising its 8088-based PC with the 8-bit data bus as a "16-bit" machine. Of course, this was in the days when CPU clock x RAM width = speed. Sure, it had a "16-bit core" which they advertised as making it "faster," except, waitaminute, in benchmarks with real software the PC was slower than, say, a Z-80. Oh, the distance that slippery slope can lead...
Brackets contain world's first nanosig, highly magnified:[.]
Once I finish this reply, I will return to the air-cooled T-bird on my dining room table that is being happy as happy can be running at 1.6 GHz. This is with a $140 mobo, a $220 chip, and a $10 fan in a good case. After I get an OS on it, I will run Tim's benchmark. Dare say if past is prologue, it will surpass the 2.2 GHz P4 when that turkey is released.
Sorry to say, but Intel has let the marketing types run their company a bit too long. The blue man group is probably going to be the folks that are blue because their investment in Intel stinks so maybe their use in their advertising is more than appropriate.
Sex is heriditary, if your parents didn't have it chances are good you won't either.
If I had to pick one reason why Intel is likely to sell lots of inferior P4s at higher prices than AMD's superior chips, I would have to say: GHz. P4s simply run at higher clock speeds than AMD's and Joe Consumer doesn't know that AMD chips are faster per Hz than Intel's. He'll simply buy the one with the bigger number. And the clueless sales people at consumer electronics stores don't help much either. I'm sure the following conversation is happening in stores across the country right now:
_________________________________
Clerk: Can I help you?
Joe: Yeah, I'm looking for a computer.
Clerk: Oh, what you want is the new spiffy Intel Pentium IV that has 1.5 GHz!!!
Joe: What's a GHz?
Clerk: It's, uhhh, a thing that tells you how good your computer is. Having 1.5 is really, really good!!!
Joe: Oh, OK. What about that one over there? [points to a less expensive AMD-powered computer]
Clerk: Oh, well, that one only has 1.2 GHz. It's not a bad computer, but I'll tell you, personally, I'd never buy it. 1.2 GHz aren't nearly enough for today's demanding applications such as web browsing and e-mail!!!
_________________________________
As much as I hate to admit it, Intel made a good decision when they increased the depth of that pipeline: They decreased the speed, but they increased the only thing that matters to computer buyers, the GHz number. What AMD really needs to do is start an advertising blitz showing how much faster their chips are at a lower clock speed (and a lower price). Otherwise, Joe Consumer (and those clueless clerks) will never hear about it.
main(c,r){for(r=32;r;) printf(++c>31?c=!r--,"\n":c<r?" ":~c&r?" `":" #");}
I don't like the P4 design. It's complex, has a messy instruction set, and consumes too much power for the performance they deliver. But the same is true of the whole Pentium series, and we have learned to live with it.
Overall, I think the Athlon may be a somewhat better deal than the P4, but the P4 isn't a slouch either. Now, I am looking forward to Sledgehammer: the 64bit AMD chip might end up being a much better compromise than Intel's 64bit offerings.
What I was alluding to is that in the future, AMD will have to support a non-x86 platform. After all these years, the x86 platform is truly a well-documented standard. Even its undocumented features are very well-known.
However, when Intel moves to their own VLIW RISC solutions, (which they call EPIC instead) AMD will likely be forced to follow them, sooner or later. And cloning that chip (or even creating a compatible interface) should prove to be a far larger challenge than merely making yet another x86-compatible chipset.
AMD hopes to avoid all of this with Sledgehammer, and keep people on at least a quasi-x86 platform. And that might work for a couple of years. But eventually the x86 architecture will run out of steam, and when that happens, Intel will be there to take over again, for at least a few more years, until everyone else catches up again.
The Pentium IV is essentially still in development. And considering that it debuted at 1.5Ghz, I don't think this should be any surprise.
We already knew that the P4 offered less performance, clock-for-clock, than other modern chipsets, but it is nice that someone has identified why. Also, it's about time that more people realize just how important cache performance can be.
However, the real question here is, what can we do about it? I mean, switching to AMD is fine in the short term, but Intel is likely to dominate the platform war in the long-term. And the P4 will likely be the first step in migrating people to their new architectures. Unless AMD's Sledgehammer is wildly successful, they will have an enormously more complicated cloning job on their hands.
As usual, tom's hardware has a really good story on the P4.