Slashdot Mirror
C`t Throws Athlons And P4s In The Gladiator Pit
An unnamed correspondent writes: "In the most recent C`T "Computer technik" there is a great benchmark with a pentium 4 (1,5 and 1,4 ghz)vs a athlon thunderbird (1,2 ghz and 1,2 ghz ddr memory with the 760 chipset).
If you think that that isn`t a fair race ... then read it now here and here in English.
You should get a copy of the German paper version anyway -- great magazine, even beter benchmark.
Now does anyone know where to get a 760 mainboard ;-)" Unnamed's cousin Noname also contributes a link to GamePC, which
reviews in grand 13-page SE-style the 1.4 and 1.5 GHz P4 chips.
I'm not a very demanding fellow. I just want 64-bit systems to be everywhere before 32-bit time_t overflows in 2039.
Personally, I think it's gonna be tight. We've been hearing about 64-bit for a long time now and yet most of us are still stuck with 32-bit.
And I really hope MS moves their OS from 32->64 in less time than it took them to go 16->32. Wasn't the 80386 released some time around 1987? Past experience suggests that the "fully 64-bit" Windows 2015 will still run some 32-bit code under the hood.
There are a small handful of folks there who do benchmarks I trust
(eg. C't and Ars Technica). It isn't always so easy to do your own
benchmarks since one needs access to all the sets of hardware you want
to compare.
I don't know if I agree with this.. These benchmarks show that Intel's 1.5 is roughly equivalent to AMD's 1.2, and we were told to expect this by the previewers.
.13u before Intel), then they can start pulling away.
.13u is going to be hard to beat. There are no benchmarks for AMD's sledgehammer, so there's no point in speculating about it.
I believe AMD uses a fully pipelined FPU (multiple ones at that). I'm sure that the P4 also fully pipelined their FPU, BUT, they added several additional stages to the basic instruction as well.
The addition of stages does two things: first it increases your max clock rate all-else-held-equal. And second it increases the latency for missed branch predictions. Again, all-things-being-equal, the missed branchs will tend to hurt more than the higher clock helps, except in a few special cases.
Intel, therefore included with those extra stages a highly advanced branch predictor that lives in the MIDDLE of the pipe. More-so, successfully predicted branches can skip the first several stages thanks to the branch cache. Thus, well-behaved code will get the benifit of heavier pipelining with fewer of the pit-falls. To make things even more tantelizing, they're using a 2x clocked Integer Unit. Thus they have a 3GHZ integer unit on these benchmarks. That means that they can further extend their pipelines with almost no visible penalty (even in branch-misses).
Unfortunately, they still seem to be plagued with branch misses (the only logical explanation for why AMD can still keep up or even surpase them). Obviously the memory played an important role in these benchmarks.. The KT133 v.s. AMD560 really only differ in memory speed, and that was enough to sway several percentage points. A more fair comparison would be between VIA's up and comming DDR-SDRAM P4 chipset.
But, as was pointed out; if Intel can get the P4 up to 2GHZ before AMD can (last rumor I heard was that AMD was going to hit
Unfortunately, as several sites pointed out, buyers don't look at benchmarks, they look at CPU speed, so Intel should be able to wrongfully win people over on this synthetic basis. Thankfully, the only people that are going to be willing to buy P4's are people needing servers (or maybe even Q3). We'd have to see NT ASP/Sql Server and or Linux+Apache+PHP+Oracle, etc to determine who's king (including memory types). Unfortunately I rarely see benchmarks on these grounds.
Sooner or later AMD is going to come out with their 64bit proc. With Mustang gone, this is their only next-great-hope. An all new design - hopefully without a tremendous cost - that has started from scratch (as the P4 did). I'm sure it too will have a heavy pipeline, but several of it's new features (such as the flat-memory archetecture) should enhance the playing field.
By then, however, the P4 will have found a new chipset that handles DDR-SDRAM and will have enough volume MB's and cases that it'll be cheap enough for the hard core gamer and possibly even casual gamer to purchase. A 2 - 3GHZ processor running at
I totally agree with another poster that said we should be rooting for both Intel AND AMD since competition is good.
-Michael
-Michael
Intel has a good public perception, but Rambus has a hideous one. I would think that until the SDRAM boards come out for the P4, the fact that it requires dual channel Rambus is going to hurt it somewhat.
(thats another thing that I find interesting, that they need to use a *dual* channel 800mhz rambus setup to be able to compete witha single channel sdram setup. Thats pathetic.)
-- "So they told me that using the download page to download something was not something they anticipated." - Bill Gates
The more benchmarks I read, the more I steadfastly believe that you must test yourself. Benchmarks don't mean squat if the machine sucks running YOUR app. Get your vendor to supply 2 test machines, and then pick the better one for you. Oh, and this also solves that "other" problem. The vendors can only get you machines THAT THEY HAVE. Machines that aren't "really" avaialble to the rest of the public for another 6 months don't mean CRAP to the guy that needs his system now.
"Science is about ego as much as it is about discovery and truth " - I said it, so sue me.
Here's a question I've got.
I have AMD machines now and when the dual-Athlon DDR SMP motherboards come out (purr purr) I will be getting another one.
A lot of the benchmarks, etc. claim that certain things are bettered by optimisations, saying that recompiling or rebuilding with P4 or Athlon optimising in place will radically change the numbers.
So for the Linux/FreeBSD crowd in the know: given that we rebuild kernels, what are going to be the chances that gcc and/or buildscripts are going to support/offer optimisations for either the P4 or the Athlon? I think the PentiumGCC people are working on K6/Pentium optimisation, any chance of it going further?
I'd hate to think that but for the want of code optimisation options for my silicon, I'd be unable to take full advantage...
--- Jump!! Fire!! Bullet time!! - Lego version of the Matrix
So what you're saying is that if I could buy the P4, It would cost twice as much as an Athlon that can beat the hell out of it in most performance benchmarks.
Thanks for clearing that up for me.
-atrowe: Card-carrying Mensa member. I have no toleranse for stupidity.
(And I was getting worried...)
For example, if you compared a Z80 running at 1MHz with a P3 running at 1MHz, you would find that the Z*) does much more work.
KTB:Lover, Poet, Artiste, Aesthete, Programmer.
KTB:Lover, Poet, Artiste, Aesthete, Programmer.
There is no
Thanks for any information!
Ah well. Right now I'm really waiting for Itanium anyway. Once that comes out, I'm hoping someone'll do a price/performance comparason of the assorted 64 bit chips on the market that will run Linux. I'm also hoping it'll push 64 bit processor prices down a bit. I'll happily go for whoever offers me the most bang for my buck.
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
As the GamePC article points out, a comparison between the Athon's and the P4's cannot be "apples to apples" at present. The P4's have significant hardware changes (such as a larger pipeline) that present software doesn't take advantage of. Since the P4 is "serialized" it really shouldn't come as a shock that the Athlon's performed better in this test. Let's not jump the gun here. Do another test in a few months and see what happens. Besides that, I don't really think game playing is the best benchmark that could be done. Most of the benchmarks involve 3D graphics. Quite frankly, with a processor like these, there are much better ways to use those cycles that would be more indicative of their power.
"I believe that a scientist looking at nonscientific problems is just as dumb as the next guy." -Richard Feynman
And people seem really happy with AMD, sure there was a minor flap when they were accused of covering up a bug with their chipset not doing full AGP speeds with Nvidia boards, but over all I see people who know what they are doing with hardware drooling over AMD and raising thier eyebrows at Intel.
Until now, it's been kind of hard to tell someone who doesn't understand technology why they should like AMD (except for the price). Not every one is willing to listen to a lecture about the evils of unfair patent law and the whole Rambus affair. Not everybody can even understand or care, how schizophrenic Intel as a company has become, shipping chips with no decent chipset support (i820 anyone?), announcing releases of high-speed chips that they can't supply in any reasonable quantity, and ignoring the needs of not only large accounts (I worked at a school board last year and had to fight tooth an nail to gaurantee a supply of celerons for student workstations) but also niche accounts that hold the strength of their image in their hands.
I'm a big believer that, public perception be damned, if your deailing in tech and you lose the respect of the technical community then, over time, you will lose the respect of the rest of the market.
I don't think the public has had much bad intel publicity that they can fully understand. However, I think that Intel's move to increase clock speed at the expense of performance will ultimately have a negative effect across all segments of the market.
What is the message going to be from you people when you are asked about which computer people should buy? It will be, "yeah you could get the intel system, but it's actually way slower than the AMD that costs a lot less too." And the psuedo-experts who read the free computer monthly will pick the argument up and spread it even further.
And then Joe Lunchpail goes to work and tells his buddies, "yeah, I never heard of this AMD, but I guess they are making faster computers than intel, even though intel says they're faster, so that's what I bought."
And this kernel of information, meme if you must, will start to weaken the Intel brand and the public's perception of Mhz. It isn't hard to understand. Faster clock speeds are just for marketing. Even John Dvorak could bold that entire line in his zdnet column.
And for all the PowerPC zeolots, give it up, you can't actually buy those chips yet either.
I'm not saying that this issue will kill Intel, but it will damage them. It is a short-sighted and ignorant move by their marketing department, how, like most marketing departments, overestimates word-of-mouth when it's in their favour, and underestimates it when it's potentially negative.
Metamuscle.com - News in the Iro
What I want to know is this: how will these two processors perform against each other in dual-processor configurations?
The answer is that the Pentium 4s were designed to not be SMP capable, while the Athlons will be using the same SMP architecture that is used currently on DEC Alpha systems, which means that each processor has two dedicated connections to the North Bridge of the motherboard, as opposed to Intel's Xeon SMP configurations, which require all the processors to share bandwidth to the North Bridge.
Friends don't let friends use multiple inheritance.
It seems as though the athlon processor kicked Intel's butt according to the numbers, but there are some other things that put athlon way ahead of the p4....for instance... PRICE! athlon chips are a lot cheaper than these P4s.....also...the p4s require you to buy an entire new system...new mobo, new powersupply, new case, and a new 454 gram heatsink (454grams is about a 1 pound). I think that if these things were added in, there is no way that anyone in their right mind would take a P4 over the AMD chips.
The anti-salmon
I hope you realize that this is a fundimental CPU design. If AMD wants the Athlon to go beyond 2GHz they will have to make a deeoper pipeline. The Athlon has already incresed the pipe from ten stages to twelve. If any CPU maker decides to make a fast CPU it must have something to feed it data. Pipelining is the way bot AMD and Intel solve this problem.
A lot of the discrepancy between GHz and performance seen in P4 chips is explainable by Intel's choice of pipeline design. Intel chose to make extraordinarily deep pipelines on the P4 chips, which allows them to crank the clock speed up up up. Sounds great, huh?
The problem is that it's a bit of a false gain. Most of the performance gained in clock speed is lost again to the serious hit you take at each branch misprediction. If you could keep your ultra-long pipe full, you'd be cruising, but you can't. Occasionally you will mispredict, and have to flush that pipe. One your pipe becomes as deep at the P4, that performance hit starts eating your lunch. Suddenly, most of your processor is sitting empty most of the time.
So, clock-for-clock P4's get slaughtered by Athlons or PIII's. But Intel doesn't care. They know that the majority of consumers buy based solely on that magical MHz/GHz number. Most consumers are not sophisticated enough to realize that there is more to performance to clock rate.
This move on Intel's part was motivated by marketing rather than management. They are playing on the uneducated masses. It is all but directly deceptive, and I hope they get their clock cleaned by the press for it.
Buyer beware.
--Lenny
I hope you realize that this is a fundimental CPU design. If AMD wants the Athlon to go beyond 2GHz they will have to make a deeoper pipeline.
:).
...Or move to a finer linewidth.
Pipelining lets you increase the clock rate at a given linewidth. It isn't a requirement for faster clock rates in general.
Sometimes it's a good idea to use a larger pipeline, and sometimes not. For a given linewidth, increasing the pipeline depth will increase the clock speed at the expense of mispredict penalty and hardware complexity and timing sensitivity. Sometimes the increase in clock speed is enough to offset the disadvantages, but beyond a certain point, a deeper pipeline makes performance _worse_. What pipeline depth makes sense depends on your branch predictor, your cache, and a few other things.
However, shrinking linewidth will always let you increase clock speed, regardless of pipeline depth. It gives a straight factor-of-x speedup of all logic, no matter how the pipeline of the chip is set up.
I've been studying this for five years, so I have a good idea of what the tradeoffs are
Its all great to look at benchmarks but a chip that is unavailable scores a 0 each time you test it. There's two reasons a chip is usually unavailable. It is priced well beyond reach and reason or it is being produced in such low quantities that they might as well not make it
Since the release of the Athlon, AMD's chips are more readily available at higher clock speeds. Right now, there are 4 full pages of vendors selling the AMD 1.1 Ghz Thunderbird. The chip sells for as low as $341. That is an available chip.
Intel's 1.4 and 1.5 Ghz chips are available from 8 vendors and will cost you between $950 and $1100. In my book that chip is not available.
Pentium 4 1.5 GHz: $1099 (Pricewatch)
AMD T-Bird 1.2 GHz: $488 (Pricewatch)
Marketing to convince consumers that Pentium 4 is faster: $4 million
Look on Intel managers' face after seeing sales statistics: Priceless
------------------
A picture is worth 500 DWORDS.