Intel Core i5 2500K Linux Performance

[thorgt]

Well, I do game, but it's Nexuiz. An 8800 320Mb is perfectly enough for me. But I build KDE regularly. So the i7 930 turned out to be a nice addition to my system.

[curaga]

Michael, in the last bench (Apache) you mention how the 2500K stacks to the 970. But there's no results for the 970 in the chart?

[PGHammer]

At the page nine of the article Intel Core i5 2500K Linux Performance it's said



This is a false statement and should be corrected ASAP, as the untrue info potentially affects many Linux users. The K version is better for overclocking, but it has some features crippled compared to the non-K version. It lacks trusted execution and VT-d support. See the product details of the 2500 and 2500K, the Advanced Technologies table.

Especially the latter might be a deal-breaker. VT-d support allows a host machine to share physical PCI devices to guest hosts when running KVM based virtualization systems. With the new Sandy Bridge K processors you can't do that. Many people might want to experiment with this features as it's supported by modern Linux distributions. But you need the non-K processor for that.

I guess Intel is crippling the overclockable processor because those interested in overclocking probably aren't interested in enterprise features (though people at this forum might make an exception to this assumption). This way they also prevent cheap-ass people from building servers with "too good" power/performance/price ratio using the over clocked K processors thus leaving room for their upcoming Sandy Bridge Xeons...

The i7-2600K alone supports VT-d. Still, even this should not be a deal-breaker for *most* users of virtualization software in Linux, even in enterprises. The reality of things when it comes to virtualization is that, for most uses (and users), VT-d is not that big an item that it's a deal-breaker. (The very reason FOR VT-d is so virtualization could be used where it normally wouldn't - for I/O-intensive software and applications.) Even though most virtualization software (including KVM-style VM software, of course) supports VT-d, it's still a niche use right now.

[PGHammer]

Thanks for this information, much appreciated. Intel's market segmentation is getting worse. For me, no VT-d and/or locked clocks == no sell. AMD may lack raw speed at the high end, but if I buy a Phenom I rest assured that I'll get both VT and overclocking potential.



If you don't shed blood over your new computer, it won't work. It's been proven time and time again.

You are thinking that no VT-d = no VT at all.

Every second-generation i-series, and without exception, supports VT-x. Not ten percent (the K-subseries), but every second-gen i7, i5, and even i3. VT-d is a *superset* of VT which includes virtualization of I/O. While KVM-based virtualization software (and most desktop virtualization software nowadays) supports VT-d, it also, by and large, still supports old-school VT-x.

The difference between VT-d and VT-x, from the OS/application point of view (which is the true determinant of which is more suitable) is that VT-d includes virtualized I/O (such as SATA, USB, et. alia), while VT-x makes do with plain old *emulated* I/O. Intel had made VT-x practically standard in most LGA775 CPUs (when they replaced E1xxx with E3xxx, the newer series supports VT-x, while the older CPUs did not) - only the Pentium Dual-Core and some C2Ds and early C2Qs lacked VT-x.

One important tack taken with LGA1366 and the followups - with few exceptions, they all support VT-x. In short, hardware virtualization has become a check-box item, at least on the Intel side of things.

[Kenni]

(from the article): The new Intel chipsets required for Sandy Bridge support, which right now are the H67 and P67, are also playing well with modern Linux distributions.

Hmm, I've just upgraded one of my CentOS 5.5 systems with a Core i7 2600 + Intel DH67CL motherboard (with H67 chipset). Due to the problems with the integrated graphics mentioned in the last Sandy Bridge Phoronix article, I moved over the old and reliable Geforce 7xxx PCI-e 16x graphics card from the old system and disabled the integrated graphics permanently in the BIOS.

The system got exactly a few seconds past GRUB and then it crashed. I further disabled all the fancy stuff; USB3, onboard audio, SATA 3.0, etc. and after a few tries the system booted X11 and then it crashed after 20-30 seconds.

I still have several things to test in order to draw any conclusions, but it's not looking good so far

New BIOS, memtest86 and test of a newer Linux distribution is up next.

[thorgt]

Excuse me... how do you run Linux with Sandy Bridge? Does that motherboard have UEFI BIOS? If so, how does the CentOS system, which is effectively RedHat Linux, run off that motherboard - as far a I know, UEFI implies GPT, and GPT implies GRUB 2, which RedHat is not going to use in the near future. How do you use GPT and GRUB 1, if it is GPT and UEFI?

[TrevorPH]

Hmm, I've just upgraded one of my CentOS 5.5 systems with a Core i7 2600 + Intel DH67CL motherboard (with H67 chipset).

That's the same motherboard that I have here and I had no problems booting an old Centos 5 install on it. I did have a newer kernel than Centos supplies - I think mine was 2.6.32.5 and is now 2.6.36.2. I'd check it with memtest86+ first (and you need the beta version of that too).

[curaga]

Excuse me... how do you run Linux with Sandy Bridge? Does that motherboard have UEFI BIOS? If so, how does the CentOS system, which is effectively RedHat Linux, run off that motherboard - as far a I know, UEFI implies GPT, and GPT implies GRUB 2, which RedHat is not going to use in the near future. How do you use GPT and GRUB 1, if it is GPT and UEFI?

Like for everything else useful (ext4, zfs), there exists a patch for grub1 for GPT.

[TrevorPH]

The i7-2600K alone supports VT-d.

Not according to http://ark.intel.com/Product.aspx?id...ec-codes=SR00C

[Kenni]

That's the same motherboard that I have here and I had no problems booting an old Centos 5 install on it. I did have a newer kernel than Centos supplies - I think mine was 2.6.32.5 and is now 2.6.36.2. I'd check it with memtest86+ first (and you need the beta version of that too).

Thanks a lot for your reply - it could seem like the kernel is the key. Booting from a Ubuntu 10.10 live CD, is "mostly" stable, while Ubuntu 11.04 Alpha seems to be completely stable. I tried to stresstest the system under Ubuntu 10.10 for 5 hours, with no crashes. Then I started firefox, clicked around on some websites ...and then the system suddently crashed. I haven't had any crashes with the Ubuntu 11.04 live CD.

So I haven't yet seen a pattern between the crashes, other than "use the newest kernel as possible to avoid crashes".

Btw; do you use a dedicated graphics card or the internal one? And from which repository do you get your current kernel? Thanks..