Opteron Consumes less power than Xeon

[xlink]

http://www.worlds-fastest.com/wfz994.html
Chicago, IL, May 16, 2007 -- An AMD Opteron processor-based server consumed from 10 to 63 percent less power when compared to an Intel Xeon (Woodcrest) processor-based server during some recent tests at the Neal Nelson & Associates computer testing laboratory.

"These tests of AMD's Opteron show a measurable reduction in power consumed in both the idle state and under heavy processing load," observed Neal Nelson, president of the independent consulting group.

The measured power reduction, when applied to an estimate of the Xeon/Opteron market size, represent potential lifetime energy savings of 175,970,880,000 kilowatt hours of electricity with an estimated reduction of 199,199,036 tons of greenhouse (CO2) gas emissions.

The tests were run with the Neal Nelson SQL Transaction Benchmark which generates traffic from up to 500 simultaneous World Wide Web clients and submits the transactions to a server system. It stresses all major sub-systems in a computer's architecture including memory access, inter-process communication, context switching, disk I/O and network I/O.

The two servers in this test were configured with the same clock speed and memory size. They were tested with the same disk drives and operating systems which had been loaded from the same media. The machines were set up with the same system tunables and ran the same application code which had been compiled by the same compiler with the same compiler options.

Nelson's firm has a long history of performance consulting to some of the world's largest computer customers including the U.S. Army, U.S. Navy, Internal Revenue Service, McDonalds, WalMart and Federal Express. His performance measurement expertise has assisted with purchasing decisions for billions of dollars worth of computer equipment.

The Neal Nelson benchmarking laboratory is available to commercial and government users for independent computer performance tests.

[p8ntslinger676]

Good news for AMD in the sever market :D

[[XC] gomeler]

I'm not going to doubt that the Opteron's run cooler than the Clovertons however what about the differences in load-handling abilities? I know for a fact that servers in datacenters never sit idle, that's wasted money in hardware that isn't needed. While it is great that the Opteron performs better than the Xeon in these tests I think that the Xeon could potentially handle more users than the Opteron due to the faster architecture. The test should have been done with the total processor load at 10%, 20%, 30%, etc and measured the amount of power consumed. Even if a Xeon consumes 15 more watts at full load, what if it can serve 30% more customers?

[VulgarHandle]

consumed less power, ok, but which one got the job done faster?

[duploxxx]

This is 100% because of FBDIMMs.... but that's not to say that there aren't benefits that outweigh the costs. AMD doesn't have the memory capacity or the RAS features that Intel has.... those things are just as important as power consumption to the server community. I'm not bashing AMD's server product... they do have a decent, well deserved grip on that market... but that doesn't mean that Intel isn't competitive with the latest platform.

It is only since the woodcrest that they are able to support more memory then AMD, don't forget the original Opt's are 4 years old..... RAS features are the same, that's added with socket F. SO next architecture due in a few months adds lots more memory.

BTW since you are soo keen on memory, you are also aware that due to the old bus architecture, the more memory you add the slower it gets? not to mention the power consumption.

gomeler;2195005']I'm not going to doubt that the Opteron's run cooler than the Clovertons however what about the differences in load-handling abilities? I know for a fact that servers in datacenters never sit idle, that's wasted money in hardware that isn't needed. While it is great that the Opteron performs better than the Xeon in these tests I think that the Xeon could potentially handle more users than the Opteron due to the faster architecture. The test should have been done with the total processor load at 10%, 20%, 30%, etc and measured the amount of power consumed. Even if a Xeon consumes 15 more watts at full load, what if it can serve 30% more customers?

performance of the Wood isn't that much better.... it is really depending on types of application. But that offcourse is irrelevant to most people. VMWare is one example, equal clock is k8 10-25% faster with io/cpu, can handle more Vm's and response is way better... still people buy Xeon's for that....

[SunFlowerSeeds]

AMD Opteron Processor Faster than Xeon in SQL Database Test! But 0-6%. Not too shabby if it consume less power.

http://www.worlds-fastest.com/wfz995.html

"Chicago, IL, May 4, 2007 -- An AMD Opteron processor-based server delivered up to 6 percent higher throughput when compared to an Intel Xeon (Woodcrest) based server while processing SQL Transactions against a Relational Database, according to Neal Nelson & Associates, a Chicago area computer performance consulting firm.

"These test results seem to support recent speculation that the AMD Opteron processor architecture is superior to the Xeon architecture for applications with a large code footprint and complex execution profile" observed Neal Nelson, president of the independent consulting group.

The test was run with the Neal Nelson SQL Transaction Benchmark which generates traffic from up to 500 simultaneous World Wide Web clients and submits the transactions to a server system. It stresses all major sub-systems in a computer's architecture including memory access, inter-process communication, context switching, disk I/O and network I/O.

The two servers in this test were configured with the same clock speed and memory size. They were tested with the same disk drives and operating systems which had been loaded from the same media. The machines were set up with the same system tunables and ran the same application code which had been compiled by the same compiler with the same compiler options.

Nelson's firm has a long history of performance consulting to some of the world's largest computer customers including the U.S. Army, U.S. Navy, Internal Revenue Service, McDonalds, WalMart and Federal Express. His performance measurement expertise has assisted with purchasing decisions for billions of dollars worth of computer equipment.

The Neal Nelson benchmarking laboratory is available to commercial and government users for independent computer performance tests."

[savantu]

It is only since the woodcrest that they are able to support more memory then AMD, don't forget the original Opt's are 4 years old..... RAS features are the same, that's added with socket F. SO next architecture due in a few months adds lots more memory.

RAS features are the same ? This has to be the most ignorant statement I've seen in years.Go check the reliability features of Xeon servers and compare them to Opteron.See if the numbers of ticks is the same.


BTW since you are soo keen on memory, you are also aware that due to the old bus architecture, the more memory you add the slower it gets? not to mention the power consumption.
...

Actually that is true only of Opteron w/ DDR.The higher you go the lower the speed.

[Shintai]

People should wakeup....

First of all, check his sites. They basicly contains nothing.
And http://www.worlds-fastest.com seems to be pretty biased.

And a 12000 hits? lol...

And in january 2007. He compares a 2.4Ghz P4 Xeon with a 2.4Ghz Opteron..LOL!

That site is just too homemade...and only something theinq would link about.

[xlink]

People should wakeup....

First of all, check his sites. They basicly contains nothing.
And http://www.worlds-fastest.com seems to be pretty biased.

And a 12000 hits? lol...

And in january 2007. He compares a 2.4Ghz P4 Xeon with a 2.4Ghz Opteron..LOL!

That site is just too homemade...and only something theinq would link about.

would you think it atypical typical of a server environment to use those parts? Both are relatively "equal"

last time I checked the 2.4GHz Xeon costs more, as does the platform on a whole.

again nothing is absolute and it's subjective to the use of the systems.
Remember though, the K8 CPUs do have an IMC and as such they have one less chip to host on the platform

but yes there is a point where power frugality does not equate to power efficiency.

[FUGGER]

500 simultaneous World Wide Web clients

We were doing that on socket 478 with ease three years ago with XS...

Right now on XS we are at 2% CPU utilization on each of the cores with 1200 people online. They consider that a heavy load, they better rethink those numbers and boast 500k transactions per second or something siginificant.

500 clients, lol

To give you better numbers, this server services 80~100 requests per second, taking the low number 80*60*60*24 = 6.912 million reqs per day.

http://www.xtremesystems.org/forums/day.TIFF

[VulgarHandle]

yes FUGGER, but it depends on what those clients are doing, doesn't it?

[Shintai]

would you think it atypical typical of a server environment to use those parts? Both are relatively "equal"

last time I checked the 2.4GHz Xeon costs more, as does the platform on a whole.

again nothing is absolute and it's subjective to the use of the systems.
Remember though, the K8 CPUs do have an IMC and as such they have one less chip to host on the platform

but yes there is a point where power frugality does not equate to power efficiency.

Did a P4 Xeon cost more? WHERE! He used 2.4Ghz NETBURST in 2007 to compare. I dont even think you can buy it anymore.

[xlink]

Did a P4 Xeon cost more? WHERE! He used 2.4Ghz NETBURST in 2007 to compare. I dont even think you can buy it anymore.

Chicago, IL, May 16, 2007 -- An AMD Opteron processor-based server consumed from 10 to 63 percent less power when compared to an Intel Xeon (Woodcrest) processor-based server during some recent tests at the Neal Nelson & Associates computer testing laboratory.


woodcrest, the last time I checked was Core based.
http://en.wikipedia.org/wiki/Xeon#5100-series_.22Woodcrest.22

[Shintai]

woodcrest, the last time I checked was Core based.
http://en.wikipedia.org/wiki/Xeon#5100-series_.22Woodcrest.22

Check his site, I was talking about his JANUARY benchmark. The entire site simply looks too biased and too small.

[VulgarHandle]

doesn't matter, the site didn't do the test, and independent company did

[accord99]

Interesting to see how the Woodcrest system's power consumption only goes up 12% when working as compared to idle. Guess somebody "forgot" to enable SpeedStep?

[freeloader]

Interesting to see how the Woodcrest system's power consumption only goes up 12% when working as compared to idle. Guess somebody "forgot" to enable SpeedStep?

Why would you enable SpeedStep on a server test? That's like bolting an open parachute on a Porsche and expecting it to beat a Corvette.

[red]

The URL itself just screams viral FUD.

May 16, 2007: AMD Opteron is More "Power Thrifty" than Intel Xeon in SQL Database Test

May 4, 2007: AMD Opteron Processor Faster than Intel Xeon (Woodcrest) in SQL Database Test

January 5, 2007: AMD Opteron Delivers 34 to 115% More Throughput than Intel Xeon (NetBurst)

September 22, 2006: 32-Bit Version of Linux up to 37% Higher Throughput When Compared to 64-Bit Version (tested with Opteron)

September 12, 2006: Tyan/AMD/Seagate Set New Standard for Small Business Transaction Processing

[accord99]

Why would you enable SpeedStep on a server test? That's like bolting an open parachute on a Porsche and expecting it to beat a Corvette.
Because one of the conclusions of the "test" was on the idle power usage of the two servers.

[Ego]

Well im goin for ... 8 cores w i5310 Xeons.
Still..
Well i goin to wait for cpu price drops, if

[Stuperman]

Just One small thing to consider, if these are 4 socket (NOT 4 core) servers, Intel's FSB would hold them back Despite intels' IPC advantage. This has been the case for a long time now, and will until CSI shows up.

[savantu]

Just One small thing to consider, if these are 4 socket (NOT 4 core) servers, Intel's FSB would hold them back Despite intels' IPC advantage. This has been the case for a long time now, and will until CSI shows up.

Right now , there are plenty of benchmarks that reflect commercial performance in which the Netburst based Xeon MP Tulsa 71xx beats equivalent Opteron servers.

Opteron shines in HPC ( FP+memory BW ) , but that is a niche market afterall.

[Warship]

Just One small thing to consider, if these are 4 socket (NOT 4 core) servers, Intel's FSB would hold them back Despite intels' IPC advantage. This has been the case for a long time now, and will until CSI shows up.

And when CSI arrive, they'll seal everything shut and start processing the crime scene.

[Ego]

Haha, lol

[xlink]

Interesting to see how the Woodcrest system's power consumption only goes up 12% when working as compared to idle. Guess somebody "forgot" to enable SpeedStep?

Intel's speedstep doesn't scale as well as powernow.
Also note that FBDIMM uses a lot of power idle or load.

[Ego]

As i remembah since old 586 days...
Cpus got hlt instruction to put cpu into halt mode to cool down cpus.

Software cooling my A-Xp2000Mhz cpu ran on 18C on web browsin.
Thats low power usage!!
But in games, abaut 45-55C :P:

Now days no one made sofrware coolin for athlon 64, and i think no one for new intels.
But hlt-(cpuice.exe) works on older p4 2.2ghz-2.4ghz prob more