With all of the attention on dual core processors lately, it has been real easy to overlook the one application that might benefit more from multiple cores than any other; Linux. OK, so technically Linux isn’t an application, but the kernel has supported SMP for nine years almost to the date. But does any of this translate to great desktop performance for dual core processors?
I thought they said they were going to compare the performance of linux in all those tasks to the performance of windows.. I don’t see a single windows result. Maybe I read what they said wrong..
especially if your system can use smp job delaying…
I’ve never seen a bench so useless
Bad review.A better approach would have been testing first windows installed with the described apps and then Linux installed on both machines with again described apps.This way you possibly eliminate the deviances in performanced only app related.
(Some benchmarks,windows installed on both Intel and AMD):
The benchmarks speak for themselves,sry the text is in dutch.
http://www.hardware.info/reviews.php?id=565&page=6
http://www.hardware.info/reviews.php?id=565&page=7
http://www.hardware.info/reviews.php?id=565&page=8
http://www.hardware.info/reviews.php?id=565&page=9
conclusion:
Not hard to see the AMD’s rule in almost all diciplines.
As a desert you have the fastest gaming CPU known too.A X2 4200+ or X2 4400 .. are superb CPU’s.
The methods of utilizing dual cores and smp systems are mostly the same – multithreading, as it already has been pointed out.
You would probably optimize a little bit different at best.
So now there are doing the same close to useless benches for daily usage that with windows. sad
What I really liked to see is the growth when compiling using make -jX with X range from 1 to 4.
strange no povray test, it has got a bench path and support multi process rendering
With dual cores, if one core is stalled, its hardware resources are wasted until whatever it’s waiting for becomes available. AMD’s solution suffers less from that, because the two cores share more resources, whereas Intel’s current chips basically just stick two P4s onto a single die (but the upcoming Pentium-M based chips should fix that.)
But why didn’t either of them pursue the “hyperthreading” idea further, i.e. instead of having more or less separate cores, spend the additional transistors on more execution units for a single core that is able to process multiple concurrent threads? Wouldn’t that yield better hardware utilisation?
Since 939 based motherboards support both amd64 and X2, wouldn’t a better test be to perform various test with the single amd64 chip and then swap in the X2 and run the same tests. (Of course, reinstalling Linux to gain smp advantages)
Hyperthreading didn’t cut it at all like it should have.
That is why the invention of dual cores become necessary after hyperthreading at all.
Probably it might have been an idea to give each indivdual core hyper threading, but that might have become complex.
Since 939 based motherboards support both amd64 and X2, wouldn’t a better test be to perform various test with the single amd64 chip and then swap in the X2 and run the same tests. (Of course, reinstalling Linux to gain smp advantages)
Or just recompile the kernel/install an smp kernel? I can switch from a non-smp kernel to an smp kernel in RHEL, for instance, with one command and a reboot. You don’t need to reinstall the whole system.
I’m using a single core hyperthreaded processor (P4 3ghz), and enabling hyperthreading has definately helped the system be smooth under heavy load vs. hyperthreading disabled. Using two seperate cores should be hands down superior to my configuration.
The only thing that I’d upgrade is the slowest part of the system; hard drive configuration. Processor speed is plenty fine for normal and even heavy loads.
Hyperthreading didn’t cut it at all like it should have.
A few applications and general responsiveness did benefit. And those improvements came almost free because Intel only had to extend the processor’s frontend a bit in order to accommodate two threads.
I just wonder what performance would be like if they spent anything like the transistors required for dual-core on additional execution units and support for more threads in a single core.
It’s not quite as simple to make a single core 2x as wide as to make two single cores. Having a really wide execution core (it would be 18 units wide in the case of an Athlon 64) makes it very complicated to do routing of buses between various components (especially the register file). It also requires a bigger reorder buffer and scheduler, which again makes the chip more complex. These complicated components also make it harder to clock the chip as fast, because its more likely that there will be a component that can’t keep up with the rest of the chip.
With more and more “virtual” cpus, it will get harder to keep every cpu busy. It is easier to utilize 2 cpus in a dual core, then like 8 virtual cpus in hyperthreading. (although I guess I’m far off the real value at any possible ratio)
this was a review not a benchmark as they claimed and a crappy one at that they should have benched windows against linux on each cpu using the same identical programs …. gimp, open office, maya.. etc, each of these apps can be used for benching and is available for windows and linux.
there was no control group here… how can they choose any difinitive victor?
i would love to see a pure compiling bench and maybe throw in some games, i do tons of compiling, to substantially speed that up would be huge for me
I did it once going from my p4 2.6 to a athalon 64 2800 which essentially cut my compile times in half…. to cut them in half again would be saweeeeet.
Reasons why I think it was a bad benchmark method to evaluate Linux with dual core processors are as follows:
1. Even though the author comments on benchmarks for Windows with dual core processors he offers no comparision with other OS (Windows XP 64 or OSX Tiger) when benchmarking Novell SuSE Linux.
2. The applications tested were a bad choice by the author. Most consumers would like to see benchmarks other than games. How about showing Mental Ray render benchmarks which runs on Linux, Windows and OSX? Screamer.net renderer from Newtek has a very small market share in film and game design. If comparing Linux against OSX then Shake benchmarks could of been used. Actually the best software to do this type of evaluation would be SPECviewperf 8.1 which runs on Linux, OSX and Windows.
i totally agree dark, if he was benching linux why din’t he say he was benching linux instead of linux vs microsoft
this bench sucked, no real control group and he really didn’t even comment on any windows performance only linux performance, I am a Linux Man myself, but I would at least like an honest comparison,….. to me this looks like windows was faster so they didn’t post the bench for it… i may be wrong, .. either way, even if linux is faster or the same, to the person who is a windows user it looks as if they are hiding something
honestly.. i believe when it comes to 64 bit linux is much faster and more efficient, mostly because windows only offers a 64 bit kernel and very few 64 bit apps
this may and probably will change in the future, but linux has a head start…. I like linux but… competition is good…. if linux getts better, MS will get better and vice-versa. Linux Keeps MS on it’s feet (but i still like linux) lol
I have a few other things to say, I am no Linux zealot, i like linux, i use linux, not because i think it is better in every respect, but because i like to learn and know how my computer works (it is better in some respects).
My main thing is, I want a real comparison, if linux comes out on top, cool, if it dosen’t, i contribute to the development community and i will do what i can to change it so linux does come out on top
either way, this bench sucked and i ask you all, please don’t judge linux based on this bench… this guy who performed this test was an idiot, i am a linux guy and while i want to hear “linux is the best, it smokes windows”, it dosen’t always,…….. sometimes it does, sometimes it dosen’t…. depends on the application, but in the end…. it will be faster, more stable, and more secure… that is a promise
I still use windows once in a while, … I use OSX a lot more (i run the media dept at a chruch), and windows has it’s strengths, OSX has it’s strnengths and linux has it’s strengths also (along with the bsd’s and others)
gotta love choice
You mean to tell me that a site catering mainly to clueless gamers posts a crappy set of benchmarks of a workstation setup? I’m shocked I tell you, shocked!
Actually the site has in the past posted benchmarks comparing SuSE Linux Professional (64-bit) vs Windows XP (64-bit). Even though Windows XP 64 at the time was in Beta release it still gave the reader an indication how the OS would perform with equal hardware against other OS such as a Linux distribution like SuSE Linux.
Thanks for your answer.
How much do you think the extra logic would cost? I.e. how much wider a single core (1.5x? 1.9x?) could you fit into the space taken by two cores and how much slower would you have to clock it?
I’d suspect it’s quite a finely balanced tradeoff in the end.
The apps chosen are not directly comparable:
Outlook is compared to Thunderbird, when Evolution would have been a better comparison. Further, Thunderbird can run on both OSs so why not use it for both?
Also for The Gimp, use the same app for both OSs. Throw in Photoshop on Windows as an extra comparison.
And iTunes is compared to XMMS? Winamp would have better!
It seems this guy is out to get Windows: “Ultimately, we would love to see a Linux configuration perform the same task as a Windows machine but faster.” (First page) So much for impartiality!
I think that a interesting benchmark is one that compares the execution of “heavy” processes in a uni core processor vs. a dual core one using optimizations such as “Auto-Parallelization” form the ICC.
No doubt that when executing more than one “heavy” process with a dual core processor thing works faster; but most people do one thing at a time, or at least one “heavy” task at a time. How often do you compile the Linux kernel and encode a movie at the same time?.
How much do you think the extra logic would cost? I.e. how much wider a single core (1.5x? 1.9x?) could you fit into the space taken by two cores and how much slower would you have to clock it?
You’d have to ask an EE. I’d imagine it depends very heavily on the core, but you can get an idea of the complexity from existing chip designs. The G5, for example, is a 5-issue 8-way core. In other words, it has 8 execution units of different sorts and can send them 5 instructions per clock cycle. Scheduling for that is complicated enough that the G5 uses tricks like instruction grouping to lower the complexity of the scheduler.
I’d suspect it’s quite a finely balanced tradeoff in the end.
Undoubtedly.
Actually, Newtek’s Screamernet renderer is an excellent choice for benchmarking 3D:
1. It has one of the largest market shares. Most animation houses rely on it on one way or another (if not completely) for previs, modeling, animating, and rendering. Despite the hype, Maya actually has lower market penetration, albiet a high popularity profile, and is actually not as well suited as an out-of-the-box solution. For example, even devout Maya artists will tell you that its modeling workflow and render engine are in many ways inferior to Lightwave’s. (By the way, that’s the reason why they finally gave up and licensed mental ray as their bundled renderer.)
2. The render engine runs natively on all three OSes, and the development team works very hard to fine-tune it for each OS and its architecture.
3. At $1,500 for the entire LightWave package, it’s actually feasible that many people would buy it if they wanted to, versus just the standalone Renderman render engine at $2,000 per license (plus $400 annual “maintenance” fee.)
I’ve relied professionally on Newtek’s animation package for eight years with no complaints, having dumped Softimage in favor of LightWave along the way.
Whoops. Got so carried away with my LightWave pimping, I completely forgot to mention Shake. Since Apple currently only sells Shake for OSX, how would somebody go about benchmarking the Windows and Linux versions?
This review would had been a lot better if they compared it against some other OS on performance. This review does’nt give much at all.
—
http://bitsofnews.com Giving you the latest bits.
http://.com Giving you the latest bits.
Weren’t you warned about that repeatedly?