nMP vs iMac-VM performance

[analog guy]

hey all...
i run at least one (usually only one but always at least one) virtual machine, so VM performance is important to me. to that end, i'd evaluated a 2013 iMac i7 running Win7 via Fusion 6, and i wanted to make some comparisons with my new nMP.

i'd spent some time tweaking settings on the VM (# of virtual cores and memory allocation) to optimize performance of my apps (they are more processor- and disk-intensive and are not at all video-intensive). i confirmed the settings with some synthetic benchmark testing.

the VM resides on a 2xSSD RAID0 on a sonnet enclosure (i.e., fast). last night i ran the same VM from the same place on the nMP (no changes at all to the disc).

a quick benchmark showed performance on the nMP to be lower (~20%) but i did not tweak any settings. in that benchmark (not real-world testing...yet), CPU and disk scores were both lower. i would think disk performance would be the same, or possibly even better given the nMP's ability to access the discs in the enclosure faster than the iMac.

as mentioned above, i experimented with settings on the iMac to optimize the VM on the iMac and only had time to run one configuration on the nMP, but all configurations on the iMac ran faster--it was just a question of degree.

on the other hand, i would experience some major slowdowns (near freezes) on the VM on the iMac when running many apps at once. i suspect the nMP (esp 6-core) might handle those scenarios a bit better but don't know for sure.

i am going to test more later but figured i would start a thread for this sort of stuff.

 

[xraydoc]

The CPUs on the 2013 iMac are a bit faster than the nMP, so if you're only going to pushing 4 cores or less, the iMac will be faster.

If you're going to need for than 4, the MP is your only choice. If you run more than one VM, especially if you assign two cores to each, you'll want the MP. I sometimes use a WinXP and a Win8.1 VM, but each one only needs one core (and mostly idle at that), so the Mac side of things takes no noticeable hit.

The 27" iMac will accept 32GB of RAM, so unless you need more, you should be good there, too.

If you need more than two or three displays, the MP is a better bet.

 

[derbothaus]

The CPUs on the 2013 iMac are a bit faster than the nMP, so if you're only going to pushing 4 cores or less, the iMac will be faster.

That would not account for 20%. It would account for under 5% and only in regards to the 4 core thanks to Haswell vs. Ivy. Once you throw more cores at the VM it should run substantially faster than iMac. Unless there is something else going on.
Memory and data bandwidth in my experience have more to do with faster VM's all being equal.

 

[xraydoc]

That would not account for 20%. It would account for under 5% and only in regards to the 4 core thanks to Haswell vs. Ivy. Once you throw more cores at the VM it should run substantially faster than iMac. Unless there is something else going on.
Memory and data bandwidth in my experience have more to do with faster VM's all being equal.

Yeah, I didn't notice the 20% statement, but it's possible the virtualization features of the Haswell processor give it that much more performance (though just a wild guess) over the older gen in the nMP.

 

[analog guy]

Yeah, I didn't notice the 20% statement, but it's possible the virtualization features of the Haswell processor give it that much more performance (though just a wild guess) over the older gen in the nMP.

i could be mistaken, but i thought that the version of the Xeon in the nMP offer the same virtualization features that are offered in the Haswells. i looked this up on intel's tech specs and that is what i saw, but it could be that there are different implementations.

certainly we know these Xeons are running a bit slower.

need to do some more testing on the VMs. i had recently plugged in the sonnet to the nMP. just brainstorming here, but perhaps spotlight or something else was running. i don't *think* so, but there were some big hits.

hope to be able to test more in the next 24h.

 

[derbothaus]

certainly we know these Xeons are running a bit slower.
.

Xeon. Singular. The Quad core.
The rest are faster in things that you would have bought them for like parallel operations and renders, etc.
The hex has the same max turbo speed for the iTunes and consumer single thread operations. We are also talking about 100MHz bumps here and there which are hardly apparent in real world when you are looking at a screen. ymmv.
If you need cores get the nMP if not the iMac IS the better buy. Inless you don't like glass displays.

 

[deconstruct60]

i could be mistaken, but i thought that the version of the Xeon in the nMP offer the same virtualization features that are offered in the Haswells. i looked this up on intel's tech specs and that is what i saw, but it could be that there are different implementations.

Same feature can run at different speeds with different amounts of overhead.

For example Haswell has a few virtualization updates over Ivy Bridge ( E5 v2 )

http://www.anandtech.com/show/6355/intels-haswell-architecture/11
If running the newest VM software that is Haswell aware there can be a substantive difference.

Apps don't normally make tons of system calls, so improvements likely largely elsewhere also (e.g, there is something mismatched about the RAM or free RAM configs. )

 

[analog guy]

Xeon. Singular. The Quad core.
The rest are faster in things that you would have bought them for like parallel operations and renders, etc.
The hex has the same max turbo speed for the iTunes and consumer single thread operations. We are also talking about 100MHz bumps here and there which are hardly apparent in real world when you are looking at a screen. ymmv.
If you need cores get the nMP if not the iMac IS the better buy. Inless you don't like glass displays.

...except that the 6-core processors & up do not hit the rated fastest clock speeds in these nMPs because in order to do so the other cores need to be disabled, which is not a feature implemented by apple.

user VirtualRain was among the first to discover that his 6-core would not rev up to those rated speeds.

anyway, the point is to provide data on VM performance for nMP vs iMac -- information that i wish i had before i made my purchasing decision. my hope is to help others.

 

[derbothaus]

...except that the 6-core processors & up do not hit the rated fastest clock speeds in these nMPs because in order to do so the other cores need to be disabled, which is not a feature implemented by apple.

That is how they are supposed to work. But if on single threads VR's did not hit 3.9GHz then Apple is doing something or the trash can can't cool them enough for them to hit the highest target bin. I remember him having issues but the specifics are hazy to me.

 

[analog guy]

That is how they are supposed to work. But if on single threads VR's did not hit 3.9GHz then Apple is doing something or the trash can can't cool them enough for them to hit the highest target bin. I remember him having issues but the specifics are hazy to me.

there's another thread about this, but unless the cores are put into an idle state that max speed is not achieved.

 

[johngwheeler]

I don't think cores are ever really "idle"

That is how they are supposed to work. But if on single threads VR's did not hit 3.9GHz then Apple is doing something or the trash can can't cool them enough for them to hit the highest target bin. I remember him having issues but the specifics are hazy to me.

I would imagine that unless you find some way of actually disabling cores, or guaranteeing that no processes will use them, that all cores will have *some* level of activity.

Simply running the OS with its hundreds of threads is likely to put some work onto every core, even if very lightly used. This light use may be sufficient to disable maximum turbo speeds because the CPU will sense that all cores are in use (albeit lightly loaded).

I don't know exactly how the Xeon turbo boost works internally, but it wouldn't surprise me if the figures quoted are a "theoretical ideal maximum", which in reality will never be achieved.

I've tried running the Intel PowerGadget utility on various Macs, and never seen anywhere near the max turbo speed achieved (e.g. 2.8GHz for 3.3GHz max turbo speed), even when running no apps (apart from Mac OS, of course)

 

[AidenShaw]

I would imagine that unless you find some way of actually disabling cores, or guaranteeing that no processes will use them, that all cores will have *some* level of activity.

Yahoo! for "core parking". Intel CPUs do support basically "powering off" idle cores.

 

[derbothaus]

I've tried running the Intel PowerGadget utility on various Macs, and never seen anywhere near the max turbo speed achieved (e.g. 2.8GHz for 3.3GHz max turbo speed), even when running no apps (apart from Mac OS, of course)

I am not disputing your experience but I have seen both my Westmere hex and i7 Ivy touch the max frequency quite often using the same gadget. It is really all over the place with various sub bins. 2.8 to 2.3 to 3.4 to 3.6 to 2.0, etc.

This light use may be sufficient to disable maximum turbo speeds because the CPU will sense that all cores are in use (albeit lightly loaded).

This is probable but not what I see on Westmere and consumer Ivy.

----------

there's another thread about this, but unless the cores are put into an idle state that max speed is not achieved.

That must be a new curve ball then. See other responses.