Performance difference between bare metal and virtual server

Ask questions about dedicated servers here and we and other users will do our best to answer them. Please also refer to the self-help section for tutorials and answers to the most commonly asked questions.
Post Reply
Jake
A regular
A regular
Posts: 54
https://www.youtube.com/channel/UC40BgXanDqOYoVCYFDSTfHA
Joined: Wed May 28, 2014 9:20 pm

Performance difference between bare metal and virtual server

Post by Jake »

HI, I am currently on 8 virtual cores of a E5-2697 v3. I am considering getting a dedicated E3-1270v3. Are the virtual cores here mapped to 1 physical core? I'm trying to understand if having a dedicated E3-1270v3 will net me any additional cpu performance.
BrendanG
Staff
Staff
Posts: 566
Joined: Mon Nov 06, 2006 2:55 pm

Re: Performance difference between bare metal and virtual server

Post by BrendanG »

Hey thanks for asking this question. So the physical cores are on 1 CPU if that is what you are asking. The overhead will also be less as well as it does not run the software aspect. So in you would be able to get more performance on a dedicated machine.

If you have further questions, feel free to open up a request from the support tab on our website. We would be glad to help there.
User avatar
Edge100x
Founder
Founder
Posts: 13154
Joined: Thu Apr 18, 2002 11:04 pm
Location: Seattle
Contact:

Re: Performance difference between bare metal and virtual server

Post by Edge100x »

Usually, virtual cores on a VDS will end up being mapped to the same physical CPU because Xen tries to keep them on the same NUMA node for performance reasons (that allows it to execute them on different cores without completely emptying the cache, for instance). In most cases, this does help, though if the workload involves separate processes that don't need to share cores, then it may not. It is also not guaranteed.

In general, bare hardware will offer much higher performance because of the lower overhead -- less cacheline bouncing, no virtualization layer, network events firing on the same NUMA node, etc. An E-2186G would likely give you a significant bump versus 8 virtual cores, for instance. Bare hardware here also comes with much more RAM, which can help with many applications.
Jake
A regular
A regular
Posts: 54
Joined: Wed May 28, 2014 9:20 pm

Re: Performance difference between bare metal and virtual server

Post by Jake »

Thank you both for your input. I am looking to stay in the Atlanta location at this time. My application is about 10 Plutonium T6 (CoD Black Ops II) game servers. The applications are running through wine on Ubuntu 24.04. The primary bottleneck now seems to be RAM, which will be easily solved by any of your dedicated servers. I also noticed in multiplayer a small amount of stuttering when the server was nearly full with 14-18 players, but was an excellent experience otherwise.

The question I was trying to ask is if a virtual core is equivalent to a full HT physical core (ie. the core + the HT counterpart). I assume this is not the case because only 8 cores are accessible within the virtual machine. The follow-up question would be are the HT counterpart cores allocated to other customers on the same machine?

While an E-2186G is tempting it doesn't appear available in Atlanta and the 6369P would be significantly overkill. Something more modest with a couple of extra cores would be awesome, like the E5-1660 you offer at your Seatle location.
User avatar
Edge100x
Founder
Founder
Posts: 13154
Joined: Thu Apr 18, 2002 11:04 pm
Location: Seattle
Contact:

Re: Performance difference between bare metal and virtual server

Post by Edge100x »

It's difficult to say exactly how the cores would compare because it depends so much on the workload (both from your VDS and other customers on the machine). Sometimes 8 VDS cores will end up being 8 physical cores of a beefy CPU and sometimes they will be four physical cores that are split into 8 virtual ones, and in either case, they will could be cores that other customers are at least lightly being run on (that context switching is part of the overhead of virtualization). We highly underload our VDS machines to try to maximize performance as much as possible, but the simple fact is that it's hard to beat the pure performance of bare metal for many workloads (including single-threaded workloads).

In general, though, I'd recommend trying a full machine with 4 cores if you are using 8 on a VDS, as a good starting point. That might be an E3-1270v3 / E3-1271v3. If you need to quickly upgrade beyond that, such as in the first month, we can help you to do it for the difference in price (we know right-sizing can be difficult and starting small makes sense). In Atlanta, options are more limited on the lower end, though, as you say.

Maybe we can work out some sort of trial with a machine. Give us a ring through the help system and we can talk more there.
Post Reply