The Top CPUs Under Linux

Linux Hack writes "LinuxHardware.org has published their latest review and this one covers the top processors from both the big x86 manufacturers. If you want to see who's on top under Linux, you should check out this review. There's something here for both Intel and AMD fanboys!"

Comments

Re:First post?

[QuantumRiff]

That is not a spoiler, it is a summary.. It would only be a spoiler if people actually read the articles....

Power consumption -- wow!

[dougmc]

The Intel Pentium D 840 cpu uses 130 watts. (The other cpus still use lots of power, but this one was tied for #1.)

Wow.

You know, it wasn't that long ago that the 60 mHz Pentium (1) was the chip that had massive power requirements. That behemoth used 13 whole watts of power!

At 130 watts and 1.4 volts, that's 93 amps. That's just plain crazy. All that heat in that itty bitty package ...

Re:Power consumption -- wow!

[questionlp]

The 130W number is the maximum TDP (Thermal Design Power) for that line of processors, so it may not hit 130W even in peak usage. AMD uses the same kind of logic behind their power numbers as well. Unless if you are running several CPU burn-like programs at once, the processor will average below that number, mostly when the throttler kicks in so that the processor doesn't burn up.

I believe the newer versions of the Athlon and Pentium 4/D processors will scale back clock speed and/or voltage in order to preserve power (be it through Enhanced SpeedStep or AMD's equivalent).

Re:Power consumption -- wow!

[Dark Coder]

Wait until you hear about a 1,000 watt for a typical 1Gbps Intrusion Prevention System.

Expect to see the wattage to go ten-fold increase once we go to 10Gbps Ethernet IPS/IDS.

On a side note....

All that powerful high-speed packet examination, and alas, sorry to say, that signature-delivery is a failing model for the IPS/IDS industry. Put stock in HIPS (Host-based Intrusion Prevention System) as its the right way to go.

Running Linux? SELinux or GRSecurity is one such HIPS-derivative.

Re:Power consumption -- wow!

[oncebitten]

well, i'd buy one.

winter's coming in a couple of months, and with the price of gas/oil being what it is compared to electricity, you can now heat your house and have a cutting edge machine at the same time.

Careful there, pilgrim...

[mosel-saar-ruwer]

All that heat in that itty bitty package ...

That's jailbait yer talkin' 'bout...

Why x86 only?

[TheRaven64]

Most (All?) of their benchmarks were running open source code. Surely they could have managed to test a few other CPU architectures. Of course, then the compiler optimisation would have come in, but that's understandable. If you're looking for a machine to run Linux on, then x86 isn't necessarily a requirement - and if it wins on performance or price / performance then that would be an interesting result.

Farked if you do, or not

[suitepotato]

So it is either put up with an Intel hotplate or put up with the cost of an AMD chip. Is that what I am reading there?

What I want is a chip that is fast, doesn't incinerate itself without a coolant system akin to that of a helium liquefaction plant, and doesn't cost more than two off-the-shelf boxes which I could yoke together with clustering.

I mean, isn't that massive parallelism ability of Linux clustering one of the things that makes this whole CPU arms races less relevant? I'd rather buy a bunch of 1.8Ghz Intel 1U rack units than a couple high-end multi-core machines.

Translation of Data

[jd]

Now, instead of just having n-way SMP systems, you need n-way SMP systems with different manufacturer's CPUs, if you want optimal performance.

The CPUs are specialized, which is why each does great at a few things but not so great overall. If you want a general system, then, you have to have multiple brands of CPU. How you're going to build such a monster, I don't know, but it's the only way to solve that problem if you want maximum power all-round.

specviewperf

[bloosqr]

I was curious how the opteron compares to the amd64 so I hunted around a bit. I couldn't find the other tests but I found the specviewperf tests on amd's own site.

While those may be the 8.0 tests as opposed to the 8.0.1 tests, it strikes me that the testing on linuxhardware looks a bit funny. The benchmarks on AMD's site are for the opteron 150 and the piv 3.4 ghz (w/ 1 mb of l2 cache). The ratings are about neck and neck on the amd site but about twice the speed as on linuxhardware's site.

The actual piv that linuxhardware test actually (model 670, piv) has 2 mb of l2 cache and clocks in at 3.8 ghz and for some reason is slower than what AMD got on for a slower chip?

This may be a compiler issue, which at the end of the day says benchmarks are meaningless until you use the right compilers

before anyone responds to this by saying well they used the same compiler so it is a fair benchmark, it is not. That benchmark tells you how long the compiler people spent optimizing for a particular chip in contrast to another chip.

crap comparison

[smash]

1. Did the benchmarker play with make -j arguments to enable the compile to take advantage of multiple cores?
2. Did the reviewer try splitting the povray renders into sections, to render concurrently on multiple cores (as povray has been able to do since at least 1993).

Picking, or limiting benchmarks to a single thread when testing multiple core/multiple virtual cpus per core is retarded.

smash.

Re:crap comparison

[oncebitten]

benchmarks biased? say it ain't so.