Phenom II vs. Core 2 Quad vs. Core i7

[Ubuntu User]

Setup:

AMD Phenom II X4 920 @ 2.80GHz, 4 x cores
2x 2GB 1066MHz memory modules

Intel Q9450 @ 2.66GHz, 4 x cores
2x 2GB 1066MHz memory modules

Intel Core i7 920 @ 2.66GHz, 4 x cores
3x 1GB 1333MHz memory modules

NVIDIA GeForce 9800 GTX / 512MB DDR3 (irrelevant since we are not doing graphics testing)
Seagate ST3500320AS 500GB drive
Ubuntu 8.10
Desktop set to 1680 x 1050
Version 1.6.0 of the test suite
Running these tests inside a GNOME environment

I have compiled my own XML suite using various tests that use all the cores in the chip, plus some memory tests.

Full results of the benchmarks here:
http://global.phoronix-test-suite.com/?k=profile&u=root-15631-24407-1064

Summary below:

http://www.dnuk.com/files/phoronix.png


I have found the memory-bandwidth test in Phoronix very inconsistant. I think it is version 0.13. In the tests above I have downloaded version 0.15 and run it manually. It is actually a 32-bit application and won't compile on 64-bit systems.

I have also tested the Core i7 with 3 x 1GB 1600MHz modules (Uncore is 3200MHz). However, it actually performed slightly slower. Not sure why. Could be the motherboard.

With the Core i7, Intel Turbo Technology is enabled in the BIOS, which means the processor actually runs at a multiplier of 21x instead of 20x. So the processor actually runs at 2.793GHz.

[Ubuntu User]

Edit to say:

Chipsets:
Phenom: AMD 790GX (Asus M3A78-T)
Core 2: Intel P45 (Asus P5Q-E)
Core i7: Intel X58 (Gigabyte GA-EX58-DS4)

[curaga]

Thanks for the results.

They pretty much summarize the current state - Phenom II is very near C2Q, but costs 25% less, while i7 is in it's own class, but damn expensive.

[Kano]

You have to disable EIST/Powernow or set it at least to max speed. This makes a benchmark more consistent between runs.

[Ubuntu User]

Yeah. There is very little between the Phenom II and the Core 2 Quad is there?

I'm not sure how much weight to place on the Bandwidth tests. Not entirely sure how synthetical they are.

I originally started with Phoronix a couple of weeks ago. I started using it to compare the performance of Xen and KVM virtual machines to native. Fantastic benchmark.

I will be re-testing the i7 when I can get a Asus P6T board. Not the various deluxe variations but the standard one. Doesn't seem to be many around at the moment though.

[Ubuntu User]

You have to disable EIST/Powernow or set it at least to max speed. This makes a benchmark more consistent between runs.

Do you mean in the BIOS?

I've turned off the power daemon in Ubuntu but not changed any BIOS settings regarding to power saving.

service powernowd stop
update-rc.d -f powernowd remove

[Michael]

I have found the memory-bandwidth test in Phoronix very inconsistant. I think it is version 0.13. In the tests above I have downloaded version 0.15 and run it manually. It is actually a 32-bit application and won't compile on 64-bit systems.

I'll look into it and getting that test updated.

[Michael]

I originally started with Phoronix a couple of weeks ago. I started using it to compare the performance of Xen and KVM virtual machines to native. Fantastic benchmark.

Glad you like it :)

[Ubuntu User]

Glad you like it :)

Thanks Michael.

On another note, you might want to update the download resource file for parallel BZIP2. I've seen problems with version 1.0.2 crashing virtual machines.

I've edited the suite on my system to download version 1.05:

<Package>
<URL>http://compression.ca/pbzip2/pbzip2-1.0.5.tar.gz</URL>
<MD5>e2448d22ee29d1e6549ac58b98df11ab</MD5>
<FileSize>26997</FileSize>
</Package>


Plus edit the install.sh script as well.

Also md-gromacs will not download (you might have fixed this with 1.6.0, I've not tried it yet since 1.4.2) so I had to edit the download file for that as well. The version Phoronix tries to download no longer exisits on the FTP server.

If there is somewhere to post these suggestions, let me know and I'll post there from now on.

Thanks

[Michael]

Thanks Michael.

On another note, you might want to update the download resource file for parallel BZIP2. I've seen problems with version 1.0.2 crashing virtual machines.

I've edited the suite on my system to download version 1.05:

<Package>
<URL>http://compression.ca/pbzip2/pbzip2-1.0.5.tar.gz</URL>
<MD5>e2448d22ee29d1e6549ac58b98df11ab</MD5>
<FileSize>26997</FileSize>
</Package>
Plus edit the install.sh script as well.

Also md-gromacs will not download (you might have fixed this with 1.6.0, I've not tried it yet since 1.4.2) so I had to edit the download file for that as well. The version Phoronix tries to download no longer exisits on the FTP server.

Just updated memory-bandwidth to version 0.15 in Git. Also updated pbzip2 to 1.0.5 in Git. I try to keep the tests up to date, but sometimes I don't realize when new updates are out for them.

Also, I believe Phoronix Test Suite 1.6 has the fixes for md-gromacs.

If there is somewhere to post these suggestions, let me know and I'll post there from now on.

Thanks


Just creating a new thread here is the preferred way or posting it to the Phoronix Test Suite mailing list or emailing michael [at] phoronix.com.

[keithlm]

Setup:

AMD Phenom II X4 920 @ 2.80GHz, 4 x cores
2x 2GB 1066MHz memory modules


You might try running your Phenom II benchmarks with a higher NB speed if you run the benchmark again.

I did some NorthBridge (NB) tests at 3.6Ghz on my PHII 940. The first set I ran was with the NB at the stock 1.8Ghz and the second set was with it set at 2.4Ghz.

Results:
http://global.phoronix-test-suite.com/?k=profile&u=keithlm-8369-7544-1128
and
http://global.phoronix-test-suite.com/?k=profile&u=keithlm-16114-15446-24502

As you can see it didn't make a lot of difference on most of the tests. But the Compile tests and the memory tests saw some extra speed. It shaved between 7 and 28 seconds off the compile tests. I would believe there was something weird going on if many of the other tests did not have almost identical results regardless of the NB. Most of the graphics tests were within a few FPS. GLMark almost doubled in speed.

Some would argue that this is overclocking; personally I just consider it a tweak. If you wanted to do clock per clock frequency tests you would generally tweak everything you could including the memory and everything else that can be tweaked in bios. This is just another thing to tweak.