Performance difference between Intel 5160 Dual-core and E5345 Quad-core

[Fogel]

Hey,

We need to buy 2 new processors for one of our HP DL380G5 and we were considering quad-core since they are actually cheaper than dual-core. Let me recap the specs :

Intel 5160 --> 3.0Ghz, 4MB Cache, 1333Mhz --> 6Ghz Total

Intel E5345 --> 2.33Ghz, 8MB Cache, 1333Mhz --> 9.33Ghz Total

There's actually a 50% premium on the 5160, which is the processor we curently use on the other 7 ESX servers. I thought it might be a good idea to use the quads in the new server since total usable Mhz are higher on the quads. I'm just worried about whenever or not it is an actual performance upgrade and if i'm gonna have any processor problem using VMotion.

Anyone knows? Thanks in advance.

Mario Lavigne

[RParker]

Core to core is what you look at, since this is what a VM will see.

A Quad Core is more efficient and faster than a Dual core, even if the dual is rated higher.

Quad (2.33) vs Dual (3.0). Quad wins, hands down.

[Fogel]

Hi RParker,

What do you mean by "more efficient"? I thought it was the same cores, just bundled as 4 instead of 2 with a speed decrease. I'm guessing the VMs will indeed each see a 2.33Ghz processor but there's 4 for them to share instead of 2 of 3Ghz. That's how I'd could think it would be more efficient. Since I haven't tested it, it's hard to guess though which is why I wrote here

And what about the VMotion? Should I have any problem with the server or should it be transparent flag-wise?

Thanks!

Mario Lavigne

[RParker]

Look at the specs on your own post. 8 meg for quad vs 6 meg for dual. That's more efficient. You would think Intel just compacted more cores in a die, but that's just on the surface. Quad is their leading technology, so dual core really doesn't have any advancement, they focus their energy on quads. so it gets the best technology, faster, smaller die form, and more work per cycle.

So pound for pound a quad is faster.

Forget the number. 1 quad core processor is faster than 1 dual core processor. It's just like a Pentium D processor. You can buy a 3.8Ghz Pentium D. how fast is that? Well it has like 512K of cache, vs 2 Meg or 4 meg Dual core cache. Dual core when they were first introduced were faster, even though they were only 1.6Ghz. That's not combined speed, thats PER core, so sure 2 of them comprise 3.2Ghz, but a SINGLE processor on a dual processor can out perform even a Pentium D 3.8Ghz processor. Why? becuase they were designed to be more efficient. More instructions per second, more L2 cache, faster FSB, and the processor itself is more powerful (even though it's less than half the speed). NOW you put two of them on a socket, and you have a really fast processor.. this is taking into account that programs today STILL are not multithreaded/multiprocessor aware... once that happens you will see even better performance.

Move ahead to Quad core, and the same increases, better performance, and larger cache accounts for EACH core being faster. They can do more, at slower rated speed, and still use less power. That's the difference.

Vmotion cares about processor type. A dual core is comparable to quad core in terms of capability (VT, flexclone, etc..). If you have a processor that doesn't have some of these type of functions, you may have trouble.

One reason why AMD and Intel are not compatible, how do you compare AMD technology with Intel when they use totally different methods? It's difficult, same with different chipsets/processors.. If you stay in the same family or close like 5000 or 7000 you are fine. If you have Dual processor Xeon hyperthreaded vs a dual or quad core, their capabilities are vastly different.

[Fogel]

Thanks for your answer.

With all due respect though, the quad core Xeon E5345 is indeed 2x Xeon 5160 downclocked and with lower voltage (reference Anandtech : http://www.anandtech.com/showdoc.aspx?i=2897&p=1). It's the same architecture, unlike the Pentium D and the Core Duo 2 for example.

You do have more cache but it's still the same 2MB per core, dual core or quad core, so the VM should still use only 2MB of cache, presuming there's only one vCPU assigned to the VM.

Let's say we have 8 VMs on an ESX server. With 2 dual cores (5160), you should have 2 VMs per core running, while with 2 quad cores (5345), you would have 1 VM per core, which should technically be better unless ESX can deal extremely well with threads (unlikely).

Now, I was more worried about the Mhz per core since that gives each VM a little bit less power to play with. When you get above 8 VMs and more than 2 machines on each core, that could get problematic. Maybe it wouldn't and maybe truly this quad-core is really better on ESX.

Thanks on the VMotion answer, that what I wanted to hear

Thanks!

Mario Lavigne

[RParker]

> Thanks on the VMotion answer, that what I wanted to hear

Thanks!

What.. no POINTS!?!?!?! sniff

[khughes]

We run 4 ESX servers, 2 with 2 x Duel Core and 2 with 2 x Quad Core, comparable speeds, with the duel core's being slightly faster in clock speed. Like the person in the post above mine said just about everything you could say on the issue. But hands down quad cores are the way to go, Especially if they are cheaper! With that being said, our quad core boxes handle VM's way better than our hosts running the duel cores.

Quad > Duel

[jhanekom · Accepted solution]

Just to advance some more theory on the subject: Dual-core CPUs with higher clock speeds should deal better with a smaller number of single-threaded, single-vCPU VMs. Quad-core CPUs with lower clock speeds should deal better with larger numbers of VMs, or multi-vCPU VMs with applications that thread well.

In general, quad-cores are generally seen as well-suited to virtualisation due to the fact that competition for resources between VMs is less likely. However, as per above, this benefit tends to fall away if you're running a smaller number of virtual machines, or if only one or two VMs have high CPU requirements.

In practice, you need to tailor your decision based on the workload you'll be running. Ideally you will want to measure performance between the two platforms and choose the right one for your particular situation. Since this isn't always possible, having any other data (such as performance monitor logs over a couple of days) will be most useful.

[Fogel]

Thanks everyone, we will go with the quad cores on our server. We usually run more than 8 VMs per ESX so the workload should be more friendly to quad cores.