Slashdot Mirror
AMD vs Intel: A Linux Bout
CrzyP writes "AnandTech puts the latest and greatest AMD and Intel CPUs, including 32-bit and 64-bit versions, to the test in their first ever "Linux Desktop CPU Roundup" to see which performs the best in various Linux applications including database, compiling, rendering, encryption, and more. They suggest the Athlon 64 3500+ over the P4 560 for "balancing price and performance". Very informative!"
Everyone KNOWS that for the best Linux performance you must use a specially optimized compile with Gentoo! It's a FACT!
So, isn't there some Bias here ?
We associated so much Intel and Windows in the past that it now seems obvious that AMD is better for Linux ?
Just a question, I have not checked the thoroughness of these tests.
Trolling using another account since 2005.
I've always been an AMD fan, this just confirms my beliefs and makes my next processor being an AMD processor decision final. Anyone know why an AMD 2400+ would be running at over 50 degrees Celsius? Check the fan and the case is at around 30 degrees Celsius.
Aside from the few times Intel released a great overclocker, I have never seen a lower-priced Intel processor beat a higher-priced AMD processor in any significant set of benchmarks..
AMD4tw!
Yet, benchmarks, until recently, always seemed to compare same clock speeds/ratings despite Intel's offerings always costing more. It's nice they're starting to be more fair to AMD.
--- We need more Ron Paul!
They suggest the Athlon 64 3500+ over the P4 560
Well yea. The AMD has more and bigger numbers. Its got to be gooder.
So this latest benchmark suggests that HyperThreading doesn't do a whole lot. Is this the case on all unixy systems (ie: is HT geared more to Windows?) or is lacklustre performance on Windows the case as well?
I'm leaning heavily to the AMD 64 stuff for my next home unixy machine, any arguments for the P4?
Trolling is a art,
Why am I surprised ?? :)
[waving hand] "You don't need to see their source code. That's not the article you're looking for..."
This one gang kept wanting me to join cause I'm pretty good with a bo staff.
I don't have an enormous pile of money to shell out all over. AMD has always done everything I need for significantly less than anything Intel has offered.
Then... how is my Athlon 64 3200 running on an NForce3 based motherboard?
Because you probably have a crap heatsink/fan combo. If you like to live richly get a Swiftech MXV462-V series heatsink for it... Works wonder but it's damn pricey (around $50.00 for the heatsink itself, big heavy copper block with a few hundred rough surface aluminum spikes driven into it in a spherical layout)
s wiftech/mcx462-v/page2.htm
It also has one of the best mounting methods I've seen, no more screwdrivers prying against super strong springs in tight corners...
http://www.cluboverclocker.com/reviews/heatsinks/
Printer friendly link
Tired of clicking next page
Im dreaming ofa big bndwdth, That can resist the
The Price/Performace of a VIA C5 (or C7)would be interesting to see here. No, I don't expect it to come even close in 'horsepower' to the players, but it would be of great interest for low-cost server appliances of sorts.
Since you didn't read the article, why are you asking the question? The testing looked pretty thorough to me, and the analysis was reasonable.
And for the record, I know a number of AMD freaks. None of them are pro-AMD because of the Windows vs Linux thing. A few of them are anti-Intel, but some of them use Windows.
So at least among those I know who voice an opinion, your thesis rings false.
Any article that bases its conclusions on price/performance ratios that leaves out AMD's socket 754 Athlon64s is overlooking a major contender. Socket 754 chips generally cost far less than the s939 ones at comparable speeds and with the current generation of chips the dual-channel memory that s939 offers doesn't provide that much of a performance boost.
Some might say that the s754 chips are an upgradability dead end but most people aren't upgrading CPUs without replacing the motherboard & RAM anyways. A s939 chip doesn't really get you much more upgrade headroom since there are no 939 boards with PCI-Express and DDR2 on them anyways...
my sig's at the bottom of the page.
Hyperthreading lets the other thread use execution units that'd otherwise be empty due to pipeline bubbles. This makes a reasonable difference on many applications on the P4, due to its absurdly long pipeline. A more sensible pipeline length (i.e. an AMD processor) means there'll be less benefit to hyperthreading. I can't think of any good reason why the effects will be different between Windows and Linux.
In soviet russia stale jokes recycle you!
Namaste
They suggest the Athlon 64 3500+ over the P4 560 for "balancing price and performance".
Naturally, I didn't RTFA, but doesn't this suggest that I, as a geek who doesn't care about the value of my money, would get better performance with the Intel? Otherwise, they would just come right out and say that the AMD is the fastest of all processors, wouldn't they? I mean, I know that I would choose a DLP HDTV for "balancing price and performance", but that LCD is just so damn cool. Hell, I don't know, maybe I'll go read the article, but this sounds like some of that marketing speak we were recently warned about.
Public use of any portable music system is a virtually guaranteed indicator of sociopathic tendencies. -- Zoso
I'd allow them to skip the Itanium processor line, but to skip the 64bit EM64T is quite odd for such a review. Handing the 64bit categories with no competition like that is almost a PR exercise. You might as well be throwing G5 performance numbers at x86 machines instead of consulting the POWER5 equivalents (closest competition).
The Swiftech MCX462-V (MVX does not exist afaik ;)) is not the best heatsink out there money can buy. There are a lot of others which beat the Swiftech in both performance, price and noise levels. I've done 2 AMD heatsink roundups, one of them was posted at /. here
:)
The latest update I made can be found here from August 2004 and includes tons of innovative Heatpipe coolers which deliver great performance at a lower price! can't beat that?
Look for a Thermalright SP-94 or Sharkoon HSP1 to get your AMD chilly
Gentoo doesn't do ANYTHING by default.
Maybe I'm blind (seems to be the case), but I stared at the OpenSSL graph results, and I see the opposite of what the written text claims about them both on the benchmark page and in the conclusion. The written statements were to the effect that the 64-bit binaries sucked and that it was probably because OpenSSL was so heavily 32-bit optimized - but when you mouse-over for the 64-bit OpenSSL graphs of AES and RSA, the 64-bit binary result numbers look like they're trouncing the 32-bit counterpart binary on the same processor, as well as everything else in the test. What gives?
11*43+456^2
I can only speak for myself and as a gentoo I dont't give a f**k about 3% performance gain by optimizing my CFLAGS. I (and probably others with me) use gentoo because we like portage and use flags. compiling software is a neccesary which I would like to take as short time as possible. So your parent question stands.
How much time is there to gain from going to 64 bit? Anybody have a clue?
According to the review, faster processors performed better than slower ones.
Gee, whodathunkit.
The debian amd64 port is still fairly experimental, but it works. It builds off the official i386 packages (almost) as often as the official packages do so there's no problem with being behind. The only thing is that a handfull of packages don't compile properly for amd64 so they're waiting for someone to patch them. As far as 32-bit applications, you have a bit of compatibility using the ia32-libs package, but it really doesn't work all that well.
So if you like debian, want the bleeding edge and don't mind a few quirks and lack of 32-bit application support (like I do) go for it. Otherwise just use normal i386 Debian or another AMD64 port.
Links: file repository, docs, wiki, and mailing list.
Recently posted: Intel's New Platform Verses AMD's 64-bit Prowess. Similar scope in benchmarks, perhaps better analyzed.
-- v --
A more sensible pipeline length (i.e. an AMD processor) means there'll be less benefit to hyperthreading.
Actually, any processor with an "overabundance" of resources (say, if the Athlon 64 had 4 FPU and had HT) can make use of HyperThreading.
Simplistic example: In the P4 case, the pipeline is long (20 stages) and there are ~4 or so execution units. That's 80 things that can be in-flight that can have stalls. HyperThreading can help keep more of the 80 "things" doing something every clock than only one thread of execution.
If the Athlon 64 had 8 different ALU/FPUs or something, the odds are that some of them would be idle every clock (that's probably more than the instruction level parallelism in most code streams). If, say, 50% of them were idle at any given clock tick, then HyperThreading would be an option to keep them busy as well.
The benchmarks would be a lot more credible and useful if scripts and data could be downloaded and run by readers.
-jim
How little in the way of data analysis skills even tech savvy people have.
Mouse over to see the 64 bit results, on a different scale? Yuck.
Do the test 3 times and take the maximum? Yuck, how about the average?
Bar charts? With non-constant widths between factors? yuck.
I think probably 3 charts would have sufficed to show the whole thing. One showing total sum of time taken to run each of 3 suites: desktop, content, and benchmark, one color per suite.
One showing the effect of 32 vs 64 bits on processors capable of doing both.
One anova of DDR1 vs DDR2 (text) and of Hyperthreading vs. Not.
One plot of performance to price ratio for the best config of each processor.
And don't even get me started on the HINT benchmark (which is hard to get anymore I guess).
((lambda (x) (x x)) (lambda (x) (x x))) http://www.endpointcomputing.com a scientific approach to custom computing.
In our first look at 64-bit Linux, we used POV-Ray and noticed that not only was the AMD64 architecture much faster at this application, but that due to 64-bit verses 32-bit precision, it produced a truer image.
What does this mean in their review? Are they claiming that 64-bit POVRay uses 64-bit integers to represent color while 32-bit uses 32-bit or something? This sounds somewhat bogus. *Maybe* POVRay uses 64-bit integers internally for calculations until it outputs the 32-bit ARGB pixels is the only thing I can think of, but I doubt that is the case.
AMD is doing something else. "Word of mouth" can be VERY powerful today. But I bet they'd be doing LOTSA publicity if the internet hadn't arrived yet.
Remember what the lack of marketing did to the Commodore Amiga (with its powerful Video Toaster), when IBM only made bleeps and creeps.
Think about it.
...a slightly larger roundup. xServe is supposedly the best bang for the buck when it comes to big cluster servers. How about: P4-, AMD-, G5-Linux roundup?
From a theoretical standpoint, considering the kinds of things a compiler does, not much. Most of the compiler's task is navigating and performing transformations on very large, branched data structures. Mostly stuff like, "Follow this pointer. Okay, does this equal that? Okay, follow this pointer. Now, does this subtree look like that one? Well, to find out, we follow this pointer..."
In other words, it's a bunch of navigation in memory with very little actual "computation." As such, it hardly benefits from doubling the width of the arithmetic units, because its task has very little to do with arithmetic.
Sure, in a very abstract sense a 64-bit CPU can do "twice as much" per clock operation, but whether that is actually useful for your intended application depends on a bunch of other factors. Compilation is not something which could benefit from having fatter integers, which is essentially what 64-bit boils down to.
Look for a Thermalright SP-94 or Sharkoon HSP1 to get your AMD chilly :)
actualy the Thermalright SP-94 is for socket 478 CPUs, he'd be after a Thermalright SP-97 for socket A CPUs. Sadly the SP-97 has been discontinued but he can probably find some around. That combined with a Panaflo or Vantec Tornado fan (depending on what performance/noise ratio he's after) will be the best air cooling solution for his socket A cpu. the Zalman CNSP7000A-ALCU is also a nice option if you're looking for a quiet, decent heatsink but sadly it cant compair with a SP-97/panaflo combo in terms of quietness/performance.
I don't know about that last part, but I would agree regarding the overlooking of the obvious. i.e. That by and large the P4 just works whereas the AMD64 is more of a headache. I went with a P4 last week. Worked great. Changed my mind because of noise and heat and returned it for an Athlon 64. The system ran slow, filesystems were corrupted when I could get a clean install, which took numerous attempts and required me to pull the ram. In the end I went back to the P4. It just works.
It means that I believe they use longs in POV-Ray and that it does produce a different image whether compiled for AMD64 or x86. Check out this article which explains what I'm talking about. This isn't made up. I've proven it by comparing the two images produced by each binary.
Reson Amd chips run better when kernel is built for amd. Intel chips are the same. Benchmark total screwed up. But no more bias than windows vs linux ones it is always the wrong kernel.
Rebuilding the kernel to match process can give upto 40% speed boast depending on the processor.
Poor windows users never see there processor chip work to its best. Hypertheading disabled ment that intel did not stand a chance since Hyperthreading is required to make up for there lack of general performace(linux kernel does support Hyperthreading if it is build right).
Ever wondered why you have to add drivers to windows to use hyperthreading yep windows does not support it at all.
I know that I shouldn't feel the trolls, but there are times when I just can't help myself.
"AMD is still having quiality control problems and there doesn't seem to be any end in sight."
Oh? I don't remember seeing anything about AMD having problems with their CPUs. I do, however, remember compiling a short list of problems Intel has had. Let's see if I can find it. Oh look, there it is! Intel churns out problem-ridden products just like anyone else. It's so funny to watch the fanboys go down in flames.
-- "Government is the great fiction through which everybody endeavors to live at the expense of everybody else."
I'd like to see something like this as well, with two incarnations:
1. a test based on what you could get for a certain amount of money
2. an all out test comparing the top of the line in each class
Depending on the application, anything from nothing to alot ;)
:)
Unlike the EMT64 Intel x86-64 processors, the AMD64's actually contain extra registers which only work on 64bit code, so be running 64bit code you get extra registers.
This really makes a difference with some CPU-heavy apps; a couple of examples that spring to mind are LAME and MySQL, which show performacne increases of ~30-50% vs. 32bit code on the same hardware. Not bad for a "free" upgrade
Moderation Total: -1 Troll, +3 Goat