Hi,

We currently have two servers that have AMD Opteron processors.

The older one has AMD:
Opteron 2384 C2 2.7 GHz 4x 512 KB 6 MB 1 GHz[7] 13.5x 1.35 75 W 115 W Socket F

The newer one has AMD:
Opteron 6172 D1 12 2.1 GHz 12x 512 KB 2x 6 MB 3.2 GHz 10.5x 1.1875 80 W 115 W Socket G34

And the next one that may be purchased TBD/TBP is AMD:
Opteron 6282 SE B2 16 2.6 GHz 3.0 GHz 3.3 GHz 8 × 2 MB 2 × 8 MB 3.2 GHz 13-17.5× 105 W 140 W Socket G34

My question is for the older vs. newer one because the GHz is lower on the newer at 2.1 vs. 2.7 that there is some question regarding performence.
There is a question to whether or not the older AMD is much faster in processing and performence than the newer one which has lower GHz and as well the one yet to be purchased with it only 2.6 GHz vs. 2.7 older but faster than the 2.1 GHz newer server.

In terms of use both current servers are being used as ESX host's and running a minimum of 6 to 10 VMs and the one TBD/TBP will also be a ESX host.

To satistify the questions from users and developers can someone explain this to be true or point to a document or location that does?

Any help is appreciated = Thanks...

It's going to be tough to give you a good answer on this one, as it really depends on use and loads.

How are your VM's configured. Are they each configured with only one logical CPU, or multiple logical CPU's?

If you only had 4 VM's each assigned one logical CPU, then the older Opteron 2384 would likely be faster for CPU intensive tasks (unless choked by it's slower HT speed).

Now, since you run 6 - 10 VM's, if they are each assigned one CPU, on the 2384 each physical core is handling between 1.5 and 2.5 logical cores. This will result in a much slower theoretical processing speed than the 6172 which will be able to run each of the logical CPU's on it's own core.

If you - like some do - assign 2 logical CPU's to each VM (this is not recommended by most VMWare experts by the way, but many do it anyway), now you have 12-20 logical CPU's sharing 4 cores on the 2384 and or sharing 12 cores on the 6172. That's somewhere between 3 and 5 logical CPU's per core for the 2384 and somewhere beteen 1 and 1.67 logical cores per CPU for the 6172. You should expect a HUGE advantage to go to the 6172 in this case.

There are many areas of computing where core count does not scale well, but VM's is one area where core count is very important.

Keep in mind that core sharing is a messy affair, and as such it does not scale linearly. You can NOT say "I have 2 logical CPU's per 2Ghz core, so I that means I have 1Ghz per logical CPU". There is a lot of overhead switching between tasks from different VM's on the same physical core, so you lose a lot of performance through sharing the same core.

If you are having performance issues with your current server, before you replace the CPU, first try dropping all of your VM's down to a single logical CPU. This usually results in pretty good efficiency improvements when dealing with lots of VM's. If this doesn't help, then maybe look into that 6282. That being said, it may not help as much as you think.

While the 2384 and 6174 both use very similar architecture (just different core counts and speeds) and are thus easy to compare to each other, the 6282 uses AMD's new "bulldozer" architecture. While it is an all around better, more power efficient and more modern design, in terms of raw CPU performance, this new architecture is actually slightly slower per clock. Not hugely so, but slightly. It may help, and if you are using 2 or more logical CPU's per core on your VM's the 16 cores definitely will help (but still consider dropping them down to one logical core per VM, unless you have a compelling reason not to) but the difference may not be as large as you think.

More likely than not, if you are having performance issues with the 6172, it is probably something else than the CPU... VM configuration (as mentioned above, one logical CPU per VM recommended), insufficient RAM, or maybe even a datastore array that isn't fast enough to handle all your VM's well.

The Opteron 2384 and 6172 are fairly similar, they differ mainly in numer of cores, the amount of RAM they can support, and clock speed. They are essentially equal in performance per core and per GHz. The 2384 has higher per-core performance than the 6172 because the 2384's cores are clocked at 2.7 GHz rather than 2.1 GHz. However, you can probably only run 8-10 VMs on the dual 2384 server before it starts to get bogged down because two 2384s only have 8 cores between them. You can run a couple dozen VMs on a dual 6172 server because two 6127s have 24 cores. Each VM will have a little less performance than a VM on the 2384s, but you can run a ton more of them on the 6172 server because you have a lot more cores.

As far as the 6282 SEs, the 6282 SEs are a completely different animal than the 2384/6174s and their performance depends a lot on the particular loading and workload. The 2384 cannot run any faster than 2.70 GHz no matter what, and the 6174 maxes out at 2.1 GHz. The 6282 SE is rated at 2.6 GHz but can increase its clock speed to 3.0-3.3 GHz depending on how much it is being stressed. Granted 3.0 GHz on a 6282 is more like 2.7 GHz on a 2384/6172 because the 6200s are a little slower per clock, but you get the point. If you have only 1-4 VMs stressing the CPU, the 6282 SE can run at 3.3 GHz and will give those VMs more performance than any of the other CPUs could. If all VMs are heavily loaded, you'll get the 6282 SEs to run at 2.6-3.0 GHz depending on cooling. Performance will overall be better than a 6174 and per-VM performance will be roughly on par with the 2384s, just that you can handle a lot more of them than you can handle with the 2384s.

-------------------------

Thank you mattlach and MU_Engineer

Sorry for the delayed response to the posting but I will take both posts and use the information from both.

Thanks again...

Hi guys,

Thank you for your valuable thought, In the case of AMD, the clock rate seems to be stagnating - it has not increased since the Athlon64 FX 53, the Athlon64 3800+ and the Opteron 250. The maximum is 2.4 GHz, and the Texas manufacturer of the 130 Nm version won't say more than that. For comparative purposes the AthlonXP 2800+ was clocking 2.25 GHz as early as the end of 2002.

Nice to meet you
Rodger Hooks