Slashdot Mirror
Latest Processors Tested Under Linux
An anonymous reader writes "I don't run into reviews featuring Linux much but here's a great review I found posted on Linux Hardware. It features the latest Intel Pentium Extreme Edition 955 and AMD Athlon 64 FX-60. Since both of these are dual-core processors, they also decided to throw in the best single-core CPUs for comparison. This is a great review for Linux folk. It gives a look at what hardware you may want to consider for your next workstation."
>It gives a look at what hardware you may want to consider for your next workstation.
Aren't we all fawning over the possibility of running Linux on a new iMac now?
It seems to dominate in the 1 test that isn't designed around games or primitive single-threaded rendering or encoding applications. You know, real life programs geared towards professionals. They say that Intel is the loser, but it seems that AMD is the one who is scrambling from the pocket like Mark Rypien (so to speak).
Intel's dual core is on 65nm so not only is AMD generally faster but will reap MHz benefits when moving to 65nm later this year.
Go AMD !!
There are places where the networks are not touching,and there are places where they are-Boeing's Lori Gunter
The first picture in the article - showing the substrate against the gate. Are the very visible lumps in that ... ummmm .... atoms?
I write a blog now, you should be afraid.
The new iMacs do not use an antiquated PC bios, they use EFI. New game.
But AMD and Intel aren't the only manufacturers with interesting new chips which Linux runs on.
Freescale has the MPC8641D a dual core e600
IBM has their PowerPC 970 dual core
What about Sun's UltraSparc T1?
I'm just really tired of hearing the same old Intel vs. AMD fan boy flame wars.
And what I'd really like to here about is how well the compilers and or OS is actually using these new processor features or are they just being treated like faster 486s.
Nothing in the world is more dangerous than sincere ignorance and conscientious stupidity.
Okay, then :) Ignoring for the moment that nobody held a gun to your head and made 'ya read the "fanboy flame wars," let's talk about those other CPUs. Feel free to link to reviews or benchmarks of these CPUs, doing useful stuff like hosting lots of dynamic web sites, acting as file servers for big-ass files, running huge databases, or other things such machines are used for. Be sure to include price tags, please. I am actually pretty curious how these other chips and architectures stand up against Intel and AMD. Obviously the UltraSparcs aren't going to be running many games, but they can encode video like the others, so that'd be an interesting benchmark.
It should already be clear that the 64-bit CPUs (the AMD ones at least) aren't glorified 486 upgrades. Last I checked amd64 doesn't even have that obnoxious x86 legacy memory model crap (640k, then 384k, then 1MB-900(ish)MB, then high-memory to 2GB, then spiffy-uber-high-memory to 4GB or higher), and just has flat memory access. Other improvements supposedly include a wider, faster bus between CPU, memory, and peripherals, and better/faster memory access.
Obviously with dual-core any SMP-capable OS (even good ol' Windows) picks up major performance gains since, well, there's two CPUs. Even with non-multithreaded apps, you get performance gains since the OS starts new processes (and migrates them as needed) to the most-idle CPU.
What do these other non-Intel/non-AMD CPUs bring to the table for spiffy performance?
For what it's worth, I'm not flaming; I'm actually curious :)
Read my stuff.
I originally thought the same thing, but as is often the case with electronics and automobiles, these look much better in real life than in the pictures you see on the web. I am continually amazed by this. If you get a chance, check one out in person.
Nothing is inexplicable; only unexplained -Tom Baker, Doctor Who
How can they compare CPU power comsumption by TDP?! This is nonsense. An athlon FX57 merely goes over 60-70W, while a dual core P4 can reach a maximum power consumption over 150W.
Statistic says: if you've got your head into a freezer and your butt into a oven you are at optimal temperature.
Well, it would be better if they also did the following tests:
/dev/null. /dev/null on another machine over ssh (using default cipher) and gigabit LAN. e.g. time cat datafile | ssh user@hostb cat > /dev/null; (where hostb has appropriate key installed so no password prompt, and datafile is fully cached in RAM e.g. 256MB file file ).
;)
openssl speed (see which crypto tasks/ciphers the CPUs are better at)
Building the linux kernel and modules
apache bench (ab, ab2) of apache serving static file with 1, 5, 10, 50, 100 concurrency
apache bench of apache running a simple perl cgi with 1, 5, 10, 50, 100 concurrency
(may wish to turn off logging for apache - to avoid disk write bottlenecks)
data compression and decompression:
gzip of a text file (cached in memory)
gzip of a binary file (cached in memory)
and corresponding gunzips to
(may also wish to test bzip and lzop)
transferring cached data from one machine to
The MySQL and Postgresql database benchmarks/regression tests.
The maximum rate of tiny udp packets per second (over gigabit LAN) the machine can take via INPUT with zero firewall rules, and with 5, 10, 50, 100 firewall rules the packet passes through, and the similar maximum rate for packet _forwarding_ (to another interface). Anyone remember Slammer?
Simple java benchmarks, compile, run, web? running apache bench tests in a vmware virtual machine?
By the way, I notice an annoying thing when people benchmark kernel builds.
Most sites use a smaller value like make -j 3 for single core CPUs and skip the higher values (like -j 4 , -j 8 etc) which they do use for the multicores/chips. If they don't want to do all values of X, they should stop increasing X when performance stops improving - and _show_ the point where things stop improving.