AMD Phenom 9500 Linux Performance

[phoronix]

Phoronix: AMD Phenom 9500 Linux Performance

Have you recently upgraded to AMD's Spider platform with their quad-core Phenom processor and are running Linux? If so, and are experiencing kernel panics, stability problems, and even a psychedelic Ubuntu logo, you're not alone. Earlier this week we had looked at AMD's new 790FX Chipset under Linux and now it's time to deliver the world's first Linux benchmarks of AMD's Spider platform. However, getting to the point of delivering these Linux benchmarks wasn't exactly smooth sailing. In this article, we'll be looking at the AMD Phenom 9500 performance under Ubuntu 7.10 as well as sharing our experiences with this new AMD platform.

http://www.phoronix.com/vr.php?view=11661

[f33l]

I have to say that you guys are doing an outstanding job in providing the latest hardware benchmarks relevant to linux users.

I've recently become a regular reader and am very pleased with all the information relevant to linux users, which isn't available anywhere else in this form.

Thanks and keep up the great work!

[d2kx]

Why haven't you disabled the TLB fix? The TLB bug isn't doing anything against the stability, because it simply won't happen. Everyone should disable this "fix" as it's causing a noticeably performance loss.

[pirast]

Thanks for the review. I'd be curious to know if you tried the 2.6.23 kernel or 2.6.24 rc4 and if you experienced kernel crashes using it.

[remm]

Hum, the overall stability of the Phenom looks really bad right now Hopefuly the B3 stepping will improve it, otherwise there's no way it will succeed.

[Dragoran]

Looking at the kernel panic screen it seems that booting with:

nohz=off highres=off

should have avoided the crashes. Can you either try a 2.6.23 (or better 2.6.24-rcX) and or boot with this options?

[Kano]

Well if you want a very new kernel you could try Thorhammer RC7 with Ubuntu 2.6.24 git kernel patched to 2.6.24-rc6-git7 - without CONFIG_IDE set (thats needed for some boards/laptops). You don't get this kernel config by Ubuntu itself. Also it is really funny that only kernel compilation was faster, but no Q6600 was in the test which is basically in the same price range.

[d2kx]

Well if you want a very new kernel you could try Thorhammer RC7 with Ubuntu 2.6.24 git kernel patched to 2.6.24-rc6-git7 - without CONFIG_IDE set (thats needed for some boards/laptops). You don't get this kernel config by Ubuntu itself. Also it is really funny that only kernel compilation was faster, but no Q6600 was in the test which is basically in the same price range.

The Phenom is faster with the useless TLB fix turned off. Also, the Phenom gets faster when more cores are used. When only 1 core is used, then the Intel Core 2 is leading the way (but which applications that use only 1 core do need that speed anyway?), but when using 4 cores, the Intel Core 2 doesn't have anything to laugh about anymore.

Testing the Spider platform with the KANOTIX Thorhammer RC7 release would be really interesting as it uses the Kernel 2.6.24, while the latest Ubuntu release uses the outdated Kernel 2.6.22 one. I am wondering if the upcoming AMD SB700 will be supported when the motherboards arrive in a month or so.

[teroedni]

so that was with the tlb fix enabled.
Thats good to know since the cpu will be faster then

As d2kx said the workaround is not needed as the bug requires alot of hard virtualization to show up at all.

For more information regarding how to get this bug look here

http://www.xtremesystems.org/forums/...d.php?t=172026

He had to run 4 vm to produce the bug


Edit:Meant tlb fix enabled. sorry for that

[d2kx]

so that was with the tlb bug disabled.
Thats good to know since the cpu will be faster then

As d2kx said the workaround is not needed as the bug requires alot of hard virtualization to show up at all.

For more information regarding how to get this bug look here

http://www.xtremesystems.org/forums/...d.php?t=172026

He had to run 4 vm to produce the bug

No, in this article the TLB Fix was enabled, so the Phenom was slower than normally in the benchmarks.