Slashdot Mirror
AMD Duron vs. Intel Celeron
DeadBugs writes: "With all the hype surrounding the new Athlon XP and P4 2.2 GHz, the more affordable processors have been ignored. Tech-Report has a great article comparing the new AMD Duron and Intel Celeron. Both are now running at 1.2 GHz and have upgraded cache. The new Duron contains XP technology, while the Celeron is a PIII Tulatin with a 100MHz bus and built on the .13 micron process."
I wish Intel wouldn't have cut off the Tualatin P3 so quickly, as it would make a decent dual processor system... but now I'll be getting a dual Athlon instead :)
A decent review, I suppose.
I think it was a tad unfair to compare a Duron using DDR to a Celeron on PC100/133 (depends on the motherboard and how they set it up). They did acknowledge it directly when discussing the memory bandwidth (which showed the expected numbers, Duron was around 2x Celeron), but I think it shows only part of the picture (especially with DDR prices back up in the stratosphere compared to say, a month or two ago). This is one reason I take benchmarks with a grain of salt...it's very difficult to objectively compare AMD and Intel CPU's now due to the drastically different architectures.
The article also mentioned the Intel headspreaders...these should be reflexive on all processors. I can't count how many "Cracked core" thread's I've read on the [H]ardOCP forums...and a reasonable number of these guys are shall we say slightly above your average user.
My $0.02...
-Colin
Tom's Hardware already did a review [tomshardware.com] of the Celeron 1300 vs the Duron 1200 (benchmarks included the Celeron 1200) where the Duron simply spanks the Celeron.
The test shows why the Celeron is inferior to the Duron: the Duron's vastly superior FPU unit allows it to substantially outrun the Celeron on FPU-intensive tasks. That is the reason why the Duron has become the choice for many do-it-yourself computer builders.
faster bus speed, faster instruction execution, faster floating point, there is now way it can't.
Pardon me if I got any of the neuroscience wrong on that description, but I think the general idea is correct.
Cheers
Athlons may be only $5 more for the same speed, but Athlon XPs are not. The older T-Bird Athlons are slower than the new Durons. And the new Athlons run MUCH cooler. So you can pay $5 more for a 1.2ghz T-bird and get a slower CPU that runs a lot hotter.
it's poor software authors and poor compiler writers that are to blame for the lackluster performance of code on Intel's CPUs.
I think this one can be blamed on Intel too. If Intel really wanted to sell its processors, they'd invest a little money in helping push compiler improvements to take advantage of their processors--such as contributing to GCC. Instead, they do invest money in compiler technology, but only in their own proprietary compiler, and then try to sell that as competition for the other two mainsteam, more popular compilers that everyone uses (GCC and MS VC++). Then they wonder why software isn't optimized for their processors.
The compiler authors don't have time to make processor-specific optimizations for every single flavor of x86 architecture out there; they already have to deal with P5, MMX, P6, K6, 3D-now, etc. Why is Intel's newest fad so special that they should get extra attention? It's not. Compiler authors are going to write their compilers to perform the best on the majority of processors out there, instead of concentrating too much on one specific technology.
How is this news? The Duron has be kicking the Celeron's tail end since the Duron's release. The Duron can usually keep up with a P3 of the same speed, or at least trail very closely behind. The fact that they are both over 1ghz is not new either, not to mention that the so-called advances for the Celeron trully haven't helped it much, as it's an aging architecture. Please people, save money, get a better product from the AMD processors. (I'd still prefer a XP over a P4 personally, I'm yet to see a P4 that didn't give me this feeling that everything for some reason loads slower, though benchmarks seem to say it can be faster in some cases...yet it's only slightly faster than a much lower clocked Athlon XP, see a problem here?)
Derek Greene
The average consumer would assume that the Pentium chips are much faster as Intel has branded a 'fancy' new 2.2 GHz chip. Even that AMD chips *model names* only reach to +1900. (about 1.63 GHz).
Almost hilariously, AMD doesn't have to get their chips running at a 2.2 GHz frequency to get nearly the same performance.
The same speed differences per frequency show up in the lower bus-speed chips (Duron / Celeron).
The average consumer is completely unaware of the closeness between speed of the chips of each company.
AMD chips are much better priced, and carry more value for their money. Stability is excellent, speed is unmatchable in identical frequency ranges. It has been this way for a couple years now.
Aside, AMD has likely changed their naming system to make their chips 'sound' competitive compared to Intel chips? (i.e. Athlon XP 1600+ sure sounds like 1600 MHz doesn't it!).
orgnine
Comment removed based on user account deletion
It's saddening to me to see the optimization skills software engineers *used* to have back in the day diminishing year by year as the ability to right crappy code is justified by ever-faster CPU's.
Well, its just a matter of economics. In the beginning of computing, you could get maybe 50 programmers for the price of one computer. So the time of the computer was valued more than the time of the programmers. The programmers had to spend a lot of time optimizing.
Now, the price of computers has fallen, until you get a lot of computing power for the price of one programmer. The time of programmers is valued a *lot* more than the time of computers. So the rational economic choice is to buy more (or more powerful computers) to make life easier for programmers.
When you look at how some of the old time programmers react to this change, I think it is insteresting to look at the medieval guilds. The programmers are angry at the newcomers and try to put them down, and make it as hard as possible for them to advance ("RTFM!"). The guilds on the other hand made it forbidden by law to practice the craft in question outside the guild. Both guilds and programmers occasionally justify their behaviour by the need to preserve the fine traditions of the art, and distrust new techniques and technologies that make things "too easy".
But what it really is, is a fear of competition. Instead of trying to improve themselves and keep up with the times they try to stomp out the competition.
This may cost me some karma...
Being bitter is drinking poison and hoping someone else will die
Oh, but I think it is. Any new technology that can give you a 10:1 performance improvement is worth ANY persons time to investigate. That can make the difference between an MPEG2 video stream compiling in 10 hours or 4 hours.
You seem to forget the whole point of optimizing the compiler is to make the compiled code perform faster and better (and possibly with fewer instructions, thus decreasing bloat). Now sure, SSE and SSE2 (and really, even MMX) require more thought from the actual developer as well, but implementing these enhancements in standard libraries and in commonly used functions would surely help people realize some of these benefits up-front with a simple re-compile of their code.
There's also your issue of considering it a "fad". This technology is here to stay, and Intel has pretty much said that using the FPU is the wrong way to go about things in their newer CPU's. By this very virtue, one could pretty much surmise out of the gate that if a test or benchmark uses *only* the FPU, Intel will likely fail, and fail miserably. Intel has really only introduced three new technologies since the release of the Pentium--
MMX (MultiMedia eXtensions) - SIMD (single instruction multiple data; basically performing the same operation repeatedly on a set of data, by only executing a single instruction) integer routines, useful in certain video functions, and in some string functions. Introduced in some later Pentium processors (not in the Pentium Pro, I believe, not sure though).
SSE (Streaming SIMD Extensions) - SIMD again, but applied to floating point work. Introduced in the Pentium III.
SSE2 (Streaming SIMD Extensions 2) - More SIMD, again applied to floating point (I believe, I've only begun to read the info on this from Intel's developer website (developer.intel.com has tons of manuals in PDF format you can read on the subject though). Introduced in the Pentium 4.
Only three "fads", as you call them. Other than this, very few instructions have been added that a compiler team need worry about; the only ones I can even think of worth dealing with in a compiler are the CMOVcc instructions (and with these, it depends on the amount of time you save using these instructions; I assume they're an improvement though since it saves you the trouble of branching in your code).
BTW: The compiler issue is really more of a Wintel based problem (afterall, the benchmarks being run are usually on Windows based systems).. take a look at Borland's compilers for example; Borland Delphi offers little optimization configurations. Borland CBuilder offers more options, but as far as I recall, no ability to compile with MMX/SSE enhancements. As someone else pointed out, you can in fact purchase Intel's compiler technology as a plugin for Visual Studio, but Microsoft should *really* be adding these features directly to their compiler.
And the latest optimization issue (which I don't believe can be addressed through compiler changes, but who knows) will likely be when Intel releases SMT-capable processors.. suddenly everyone running Windows XP Professional (or if MS decides to allow dual processors on WXP Home Edition) will be able to run multiple threads at the same time-- the problem is, not a lot of software utilizes multiple threads, even though the technology has been around since Windows '95 (obviously in a single processor environment, you don't get a speed increase, but it does make your app more responsive (the UI doesn't freeze during long operations, for example)). Benchmarking an AMD processor without any form of SMT against an SMT capable Intel processor would be unfair at that point.
All I know about Bush is I had a good job when Clinton was president.