Jump to content

Virtualisation - Performance

Christ1992
Go to solution Solved by leadeater,
1 hour ago, Christ1992 said:

Do you think this difference in performance could be purely due to the CPU?

Yes you are comparing a much older architecture CPU to a much newer one then equally or more importantly the Xeon Gold 6142 has a higher power allowance and a lot more CPU cores so should be much faster. The Xeon Gold 6142 is also nearly twice as expensive as the Xeon E5-2643v2 and is not a direct replacement for it, all be it like 4 generations newer Xeon.

 

Xeon E5-2643 v2:

  • ~$1550 MSRP
  • 6 Cores / 12 Threads
  • 3.5 GHz Base (more important than boost typically) / 3.8GHz Boost
  • 130 TDP

Xeon Gold 6142:

  • ~$2950 MSRP
  • 16 Cores / 32 Threads
  • 2.6GHz Base / 3.7GHz Boost
  • 150W TDP

Not considering any other VMs running on the host the 6 vCPU VM running on the Xeon Gold 6142 will sustain a higher average frequency across the 6 allocated cores, top that off with a much more performant architecture and the VM on this host should perform significantly better.

 

Then you also have to consider any other VMs running on each host, if you are sharing physical cores, typically the case, a CPU with a higher core count has better ability to schedule CPU time to a VM and achieve higher effective performance. You need to look at VM CPU metrics like  CPU %wait times and CPU Ready times, these will tell you if the host has too many VMs demanding CPU time and thus cannot service the VMs adequately.

 

Also when you say VMware are you running ESXi or VMware Workstation on Windows? Hyper-V is actually very good and should not give any performance difference between it and ESXi, personally I choose ESXi but Hyper-V is great too.

Good afternoon,

We have 2 setups that we are comparing.

 

Setup 1:
Hyper-V
Intel Xeon E5-2643 v2 (3.5GHz)

With a VM:
6vCPU & 32RAM
Host OS: Windows 2019

 

Setup 2:
VMware
Intel Gold 6142 (2.59GHz)

With a VM:
6vCPU & 32RAM
Host OS: Windows 2016

 

We notice in these 2 setups that the VM with the intel gold (vmware) has a much better overall performance.

(Even though the clock speed is higher from the old CPU.)
The 2 big differences in my opinion are the CPU & Virtualization software.
Do you think this difference in performance could be purely due to the CPU?

I know there isn't much info to work with.
But I believe it could be just these 2 differences.

 

Kind regards,

Link to comment
Share on other sites

Link to post
Share on other sites

1 hour ago, Christ1992 said:

Do you think this difference in performance could be purely due to the CPU?

Yes you are comparing a much older architecture CPU to a much newer one then equally or more importantly the Xeon Gold 6142 has a higher power allowance and a lot more CPU cores so should be much faster. The Xeon Gold 6142 is also nearly twice as expensive as the Xeon E5-2643v2 and is not a direct replacement for it, all be it like 4 generations newer Xeon.

 

Xeon E5-2643 v2:

  • ~$1550 MSRP
  • 6 Cores / 12 Threads
  • 3.5 GHz Base (more important than boost typically) / 3.8GHz Boost
  • 130 TDP

Xeon Gold 6142:

  • ~$2950 MSRP
  • 16 Cores / 32 Threads
  • 2.6GHz Base / 3.7GHz Boost
  • 150W TDP

Not considering any other VMs running on the host the 6 vCPU VM running on the Xeon Gold 6142 will sustain a higher average frequency across the 6 allocated cores, top that off with a much more performant architecture and the VM on this host should perform significantly better.

 

Then you also have to consider any other VMs running on each host, if you are sharing physical cores, typically the case, a CPU with a higher core count has better ability to schedule CPU time to a VM and achieve higher effective performance. You need to look at VM CPU metrics like  CPU %wait times and CPU Ready times, these will tell you if the host has too many VMs demanding CPU time and thus cannot service the VMs adequately.

 

Also when you say VMware are you running ESXi or VMware Workstation on Windows? Hyper-V is actually very good and should not give any performance difference between it and ESXi, personally I choose ESXi but Hyper-V is great too.

Link to comment
Share on other sites

Link to post
Share on other sites

Also I have servers with the slightly newer E5-2643v4, these are frequency optimized CPUs for database type workloads or anything that benefits from stronger single thread performance. Currently these servers with these CPUs run our infrastructure backup software and are used to ingest and deduplicate data and then store on backend storage.

 

These frequency optimized CPUs aren't really for virtualization unless the VMs you will be running will be doing these types of workloads and you specifically don't over provision virtual CPUs too much. Nothing really makes them unsuited to virtualization though, you just pay a lot more for the higher frequency targets (higher TDP) when you could spend the same and get a Xeon option with many more cores instead which is more suited to this.

 

The Xeon Gold 6134 from that generation is more a direct newer generation replacement for the E5-2643v2.

Link to comment
Share on other sites

Link to post
Share on other sites

Hi @leadeater
 

Thanks for the comprehensive answer!

Why will the gold keep an average higher GHz if the base GHz is a lot lower?
and the number of vCPU is the same?

 

-

 

Which elements within a CPU should i pay attention to in order to use the CPU for the things below?

 

1. Hyper-V (VMs)
2. Remote desktop
3.Office Suite
4. Browsing
5. SQL Database (CRM)

 

What we do is place a refurbished server in a datacenter to provide small SMEs with a cheap cloud solution.
A small SME (1 - 2 - 3 people) in Azure is simply too expensive and unpredictable.

That is why we have put our own servers in a datacenter that helps small SMEs.

 

But of course we want to keep improving and I'm now looking for a way to find suitable CPUs.

Which is better high core count or high GHz?
What is the difference when a old and new CPU have the same GHz and same core count?
Is there a kind of guideline i can use?

Link to comment
Share on other sites

Link to post
Share on other sites

8 minutes ago, Christ1992 said:

Which is better high core count or high GHz?

What is the difference when a old and new CPU have the same GHz and same core count?
Is there a kind of guideline i can use?

CPU clock speed (GHz) means nothing without taking IPC (Instructions Per Clock) into account. Basically this is how much work a CPU core/thread can do per clock cycle.

 

I'm not familiar with these types of CPUs but you can test single-core performance in all versions of Cinebench. This can give you a better comparison on how these two chips compare to each other in respect to IPC and single-core performance.

8 minutes ago, Christ1992 said:

Why will the gold keep an average higher GHz if the base GHz is a lot lower?

Like @leadeater explained they are very different architectures. I don't know what scale of changes were made between these generations but it's safe to say that changes were definitely made to how the chip behaves in regard to frequency and boosting.

 

Have you also taken into account the memory differences? A Quick search shows that the E5-2643 uses DDR3, with the Xeon Gold being DDR4.

Link to comment
Share on other sites

Link to post
Share on other sites

38 minutes ago, Christ1992 said:

Why will the gold keep an average higher GHz if the base GHz is a lot lower?
and the number of vCPU is the same?

Intel CPUs have a boost table and this is based off the number of active cores and the allowed power draw, along with workload/instruction type. My assumption was wrong however but there's only a 100MHz to 200MHz difference anyway, I doubt you are running anything AVX-512 (regardless doing so gives massive performance increase).

 

Quote
Mode Base Turbo Frequency/Active Cores
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Normal 2,600 MHz 3,700 MHz 3,700 MHz 3,500 MHz 3,500 MHz 3,400 MHz 3,400 MHz 3,400 MHz 3,400 MHz 3,400 MHz 3,400 MHz 3,400 MHz 3,400 MHz 3,300 MHz 3,300 MHz 3,300 MHz 3,300 MHz
AVX2 2,200 MHz 3,600 MHz 3,600 MHz 3,400 MHz 3,400 MHz 3,300 MHz 3,300 MHz 3,300 MHz 3,300 MHz 3,200 MHz 3,200 MHz 3,200 MHz 3,200 MHz 2,900 MHz 2,900 MHz 2,900 MHz 2,900 MHz
AVX512 1,600 MHz 3,500 MHz 3,500 MHz 3,300 MHz 3,300 MHz 2,800 MHz 2,800 MHz 2,800 MHz 2,800 MHz 2,400 MHz 2,400 MHz 2,400 MHz 2,400 MHz 2,200 MHz 2,200 MHz 2,200 MHz 2,200 MHz

https://en.wikichip.org/wiki/intel/xeon_gold/6142

 

Apologies for the horrible table layout above, just go to the link lol.

 

Have a look at this review, mainly this page/section of it. The 8176 is the same architecture and core performance as the 6142 and operate at similar frequency for a single core/boost.

https://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/14

 

Intel Skylake Xeon vs Intel Xeon Ivy Bridge is likely roughly around 30% faster clock for clock (IPC) so you're going to be at least 30% faster on the 6142 CPU host not taking in to account any other benefits like memory bandwidth improvements or utilizing newer modern instructions that offer better performance.

 

38 minutes ago, Christ1992 said:

Which elements within a CPU should i pay attention to in order to use the CPU for the things below?

 

1. Hyper-V (VMs)
2. Remote desktop
3.Office Suite
4. Browsing
5. SQL Database (CRM)

Quite honestly most CPUs offer excellent performance so the bigger factor is number of users required to be supported, or in other words active connected sessions to the server. SQL like frequency the most rather than a lot of cores but you have to keep in mind the number of vCPUs to pCPU ratio.

 

Personally I favor any Xeons with the higher TDP then look at the rough number of cores I want. That way I can balance the frequency and core performance with number of cores to get a good balance.

 

For example a lot of our Intel ESXi hosts use Intel Xeon 6242R as these have the highest allowed TDP in this generation at 205W but still have a decent number of cores at 20 and can run these cores at higher frequencies than other options.

 

Compare the 6242R to a 6238 or 6252

 

6242R:

  • $2700 MSRP
  • 20 Cores / 40 Threads
  • 3.1 GHz Base / 4.1 GHz Boost
  • 205W TDP

6238:

  • $2800 MSRP
  • 22 Cores / 44 Threads
  • 2.1 GHz Base / 3.7 GHz Boost
  • 140W TDP

6252:

  • $3900 MSRP
  • 24 Cores / 48 Threads
  • 2.1 GHz Base / 3.7 GHz Boost
  • 150W TDP

The 6242R will out perform the other two at the same or cheaper cost even with less cores, and by a decent amount I might add. The 6242R was a mid product cycle revision release to compete better against AMD EPYC, it's got some other details and features removed from it but that's really getting in to minor details that don't matter.

 

Basically it's hard to go wrong unless you go too old like that Intel Xeon Ivy Bridge which is starting to get a bit long in the tooth. Newer generation Intel Xeon product ranges are also way more complicated with sooo many more options so my hot tip is look at the TDP as it's a huge factor in performance.

 

38 minutes ago, Christ1992 said:

What we do is place a refurbished server in a datacenter to provide small SMEs with a cheap cloud solution.
A small SME (1 - 2 - 3 people) in Azure is simply too expensive and unpredictable.

That is why we have put our own servers in a datacenter that helps small SMEs.

haha yes Azure/AWS are anything but "cheap".

 

Hope the above helps, feel free to ask any other questions.

Link to comment
Share on other sites

Link to post
Share on other sites

1 hour ago, Christ1992 said:

We go from E5620 to E5-2643 v2.

Surely this should be an improvement?

Yep, that's quite a big improvement. I have a lot of E5520, E5640 and X5690 CPUs in my home lab plus used these many years ago in our actual production ESXi clusters. Going from LGA1366 (Xeon 5500/5600 series) to LGA2011 (Xeon E5/E7) is quite a decent performance upgrade.

 

As with anything if it works and does the required task then it doesn't really matter too much if there is something better etc, end result is what's most important. For what you are doing I would focus on parts availability and being able to standardize a bit, that way you can be sure you can deliver good service and reliability and also not spend a lot of time trying to figure out performance differences.

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×