VMware Communities
sapeur3873
Contributor
Contributor

Slow performance guest vs host?

hello,

i buy a new laptop (DELL XPS 13, I7-1065G, 16Giga of memory, 1T of SSD), windows 10 1909

before i had an XPS I7-8660U, 16giga, 512SSD

i m very disapointed about the performance of this system with the vmware..

i do a benchmark online on this site,

https://silver.urih.com/

one on my guest i had a result of 13800

one on my VM i had a result on only 5200

the parameter for the vm is 4CPU, 8192G of memory, hard disk NVMe,accelerate 3d graphics.

i run also on win 10 1909 on the vmware.

somebody can explain me a so much difference on result?

i use some automation software on my vm, and they are very slow,

i don't have anymore my old pc for can compare benchmark, but it seems it was at least same, maybe better before.

i try a lot of parameter, also on my bios... but i don't know what to do more..

thank for your support

Reply
0 Kudos
17 Replies
scott28tt
VMware Employee
VMware Employee

If that’s a quad-core pCPU, giving the VM 4 vCPUs could be part of the reason (even if the pCPU supports hyperthreading)

Tried the VM with fewer vCPUs?


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

Although I am a VMware employee I contribute to VMware Communities voluntarily (ie. not in any official capacity)
VMware Training & Certification blog
Reply
0 Kudos
sapeur3873
Contributor
Contributor

hanks for your answer, i try different cofniguration but the result seems not so good

i try different configuration,


(Removed by moderator in respect to 2.4 of https://www.vmware.com/content/dam/digitalmarketing/vmware/en/pdf/downloads/eula/universal_eula.pdf )

i just don't understand why vmware performance is less than half of the guest performance, i know that using a vmware i will loose some performance, but not so much.

i suppose to can have a score around 9000/10000 when my guest is 15300.

i found when i check my task manager, the test is not using core time when i'm on the guest only cpu is going 100%

but when i do the test on my vm the task manager, show me that is used 100% of all processor but nearly 0% of core time, but on same time on the guest is it 100% of core time too?

maybe we can have a explanation???

Capture.JPG

Reply
0 Kudos
Mits2020
Hot Shot
Hot Shot

I am sorry that you lose so much performance, especially given that you use NVME and such high-end hardware. All I can think is that a) both guest and host may be busy doing other things (are you sure CPU usage is 0% before you start benchmarking?) or b) the spectre/meltdown/whatever_new_vulnerabilities mitigations are running two times (one on the host and one on guest) and this definitely affects performance. To my knowledge there are no published benchmarks for guest performance in VMware, but my bet is that from all desktop virtualisation solutions VM is the fastest.

If you have access to a similar linux host machine (or if it's possible to boot from a live linux CD/USB on your existing laptop and installing VMware for linux) you could try running the same guest and see if there are any significant differences.

Reply
0 Kudos
sapeur3873
Contributor
Contributor

hello, thanks for your suggestion

yes i'm sure that before doing test the cpu usage is 0% before start benchmarking.

before doing the test i look on both task manager (VM and guest).

what do you mean about "the spectre/meltdown/whatever_new_vulnerabilities mitigations are running two times "???

i can try yes to do an usb bootable linux system , and try to test the vmware!

but i would like to have my vm working on my computer directly...

Reply
0 Kudos
sapeur3873
Contributor
Contributor

hello,

i try to install a linux (unbuntu distribution) on a sdcard,

but i m not very familiar with this one, and i was not able to start a vm (i was able to install vmware, but it seems not fully.

and i'm not sure to have it on a sd card, is very fast enough for can do a comparaison test.

Reply
0 Kudos
RGarrison5
Contributor
Contributor

I've experienced similar problems, and some of the things I have done that made the biggest improvement:
Don't try to set too much memory for graphics. My Win10 client suggests 1 GB; I find better performance at 768 MB.

Don't try to set too much RAM for system; I have a 16 GB host, my clients have 2 to 3 GB, depending on their purpose.

Try toggling on/off 3D graphic acceleration; I find that makes a difference.

Make certain  VMWare Tools is current.

I run the host on a wireless network, and I found that I get quite a performance jump by having a separate USB network adapter for my one client that needs more bandwidth; it connects directly to the client, so I'm not sharing bandwidth with the host.

Reply
0 Kudos
sapeur3873
Contributor
Contributor

thanks for your feedback,

i try to do what you propose to me.

i reduce to 4 cpu,

3104MB for memory

and i uncheck accelarating 3D

the result is : 4177

i do the same test with only change accelarate 3D graphics

and video memory at 512mb

the result is : 5142

is realy far away of my guest benchmark at 15000

i don't know what to do anymore...

anybody can tell me what is due the performance between host and guest...

anybody can do the same test than me?

try this site benchmark

https://silver.urih.com/

and test it one time on is guest and one time on his VMware, and give me result?

and configuration???

thanks

Reply
0 Kudos
Mits2020
Hot Shot
Hot Shot

What a nice way to spend a Saturday morning, doing VMware benchmarks that I wanted to do for ages! Here's a small essay for your Sunday breakfast:

Well, from what I understand Silverbench is a CPU-only a java benchmark using the 'photon mapping rendering engine' and they say they don't use the GPU. Therefore all it measures is the CPU speed and at first sight it should not be very different within a VM guest, but it also depends on the java implementations, the browser, and I guess several other parameters.

Perhaps it's time to start a discussion and ask VMware experts to suggest a proper real-world performance benchmark to use. It would be nice if someone started a website collecting VMware (and perhaps other virtualization solutions') benchmarks.

Regarding your question about Spectre/Meltdown, you can read summaries like https://spectreattack.com/ and search for benchmarks demonstrating the effect the mitigation has on sites like phoronix.com. It's a hot and polarizing debate and most users probably have no idea what's going on and why their performance deteriorates over time. My point was, since both the host

machine and the guest(s) have updates/patches applied for the mitigation of the vulnerabilities, these affect CPU performance two times. So if you have e.g. 5% loss in the host (compared to a non-patched system) and 5% in the guest, total is about 10% (some guru please correct me if I'm wrong).

For what it's worth, and to satisfy my own curiosity, here are some benchmarks from my humble PC:

Workstation 15.5.0 build-14665864 on Win7 x64 Ult, 8GB RAM, i3-4130 2 cores 4 threads @3.4 GHz

All OSs are x64, use 4GB RAM, 4 threads and are fully updated as of 16 May 2020. All tests run in Firefox 76.0.1 or the latest applicable version (see details).

(Removed by moderator in respect to 2.4 of https://www.vmware.com/content/dam/digitalmarketing/vmware/en/pdf/downloads/eula/universal_eula.pdf )

So it seems that for this particular case there is no significant difference between 4 and 2 cores, which seems rather counterintuitive for a CPU-only benchmark... I am impressed, I did not expect that outcome.

Finally, I made an extensive search to see what CPU performance loss should be expected in a virtual environment. Some articles from 8-10 years ago claimed that 90% of the physical performance In VMware Workstation old versions (e.g. v6, 6.5) is easily achievable. I was not able to find more recent figures. Unfortunately, all recent articles focus on ESXi and cloud performance on big machines and not on us simple users trying to do some simple stuff.

However, after this search I just discovered a very interesting resource called VMware OS Optimization Tool (https://flings.vmware.com/vmware-os-optimization-tool): "The OS Optimization Tool includes customizable templates to enable or disable Windows system services and features, per VMware recommendations and best practices". My intuition tells me that it may help in the direction of making simpler and more effective virtual machines (and stop the endless hunt for semi-legal "debloated" or "lite" operating systems!).

So have a nice Sunday breakfast and send in your comments and insights!

Reply
0 Kudos
sapeur3873
Contributor
Contributor

THANKS for your long reply,

i hope the weather is very bad at your home, like this your are not missing any outdoor party during you do all this test!!!

your are lucky on your badder test your are at 54% difference..

me i am at 33%...

luckyly my guest is working very well!

i didn't spoke about that, but i already try vmware optimization tool.. but no difference at all.

i also try to downloaded a fresh windows 10 from microsoft..  the result is nearly the same...

most of my vm, are debloated by some tools before use...

i install fresh windows 10, i debloat it and after i install my need!!!

90% of performance, will be a dream.. but it seems so far!!!

thank you for your comment, and test..

my sunday will be anyway outside, maybe i do during the nigth!

i hope somebody can find any explanation bout this loss of performance...

Reply
0 Kudos
sapeur3873
Contributor
Contributor

what i explain on my first post is:

when i do the benchmark on the guest, i see in task manager (you can select in performance/ processeur/show the core time)

when i do the benchmark on my guest, i see that my computer don't use any core time, the processeur is going at 100%, full speed, but no or very poor core time!

when i do the benchmark on my vmware, and i look the task manager of the guest, i see the processeur going at full speed, and using 100% of core times also, is it the image of the first post.

i imagine if the core is use or not, the result or speed will not be really the same???

you can try and check if you want!

Reply
0 Kudos
Mits2020
Hot Shot
Hot Shot

Some more ideas:

1. Did you try to install any Linux guests yet? It would be interesting to see whether this is a Windows thing or not (e.g. bad/buggy windows CPU scheduling, 10th generation is still a very new CPU hardware).

2. Also, are you 100% sure your CPU is not throttling due to high temperature (are you checking the temperatures using e.g. Core Temp)? If the CPU is throttling it will show 100% even if the real speed is 0.1 GHz. After all this is a very thin 13 inch laptop and all thin laptops have very bad thermal behaviour. I remember reading a Macbook review this year where the CPU was throttling so badly, that people were talking about fraud...

3. Ask any of your friends/colleagues to lend you their laptop (preferably a gaming one, e.g. a 17" Alienware) and make speed tests to see if this an XPS thing.

4. Finally, there is also the case of bad hardware (defective motherboard, bad thermals etc.). In that case you could try to benchmark another sample of XPS13 in some shop where they have live PCs running for demonstration. This is a $2150 laptop and if I were you I'd ask for a replacement or refund if it does not work for my purposes. I'd buy a bigger, thicker laptop with proper airflow.

Reply
0 Kudos
Mits2020
Hot Shot
Hot Shot

On second thoughts, the more I think about it, the more I am convinced this seems like a usual thermal throttling issue (and maybe I was wrong that CPU would still show 100%). Searching the net for "Dell XPS throttling" one can find a lot of articles in the Dell forums like

https://www.dell.com/community/XPS/XPS-15-9570-i9-thermals-throttling-and-solutions/td-p/6116619

https://www.dell.com/community/XPS/XPS-13-9370-CPU-throttling/td-p/6061871

There is even a tool to monitor/control/disable throttling called ThrottleStop (see http://forum.notebookreview.com/threads/the-throttlestop-guide.531329/) but I am not sure whether this would a) not void your guarantee or b) damage your laptop. So, I wouldn't recommend playing with it to increase performance.

Finally, I think that using benchmarks that utilise the CPU for 10 minutes continuously like Silverbench does, would inevitably lead to throttling on ANY laptop. From my observations, when running a VM machine on my desktop, my PC uses a lot of CPU to start the VM and then it settles at 0%. Then the virtual machine runs very responsively, unless it is running CPU- or disk-intensive stuff. So it's OK to use VMware in a laptop, as long as you don't need CPU-intensive workloads.

Conclusion: If your particular application for which you need to use virtual machines does require continuous 100% usage for a long time (e.g computations, simulations etc), then clearly using a laptop is not the appropriate solution. If your application is 'normal', I'd suggest you just forget these benchmark results (which after all do not correspond to a realistic use scenario) and enjoy your high-end machine!

I hope all the above will put some peace on your mind and that you won't hit your head on the wall regretting spending all that money for a laptop that ends up not being suitable for your purpose  🙂

Reply
0 Kudos
sapeur3873
Contributor
Contributor

thanks mits,

i try to install a unbuntu, but on a SDCard, and i'm sure the result is not the same than on a hard disk.

i know about the cpu throtlling on the XPS, (is it my third XPS 13). the software throlstop is not working on new generation of CPU intel (i7-1065G7).

i check also this information.

i'm not complaining about my XPS directly, when i do the benchmark on the XPS itself, i have pretty good result.

i don't understand, why the benchmark result  on guest and host are so different??? only 35% of result!

i know that the cpu throll, and also i know that if i run in battery the power of cpu is very less than plugged on AC.

maybe i have only to forget the benchmark, but i didn't any performance upgrade on my VM betwwen my old XPS i7-8560 and my new I7-1065 (and i say on the VM only), i saw better performance on my XPS itself.

but i have somany enginneering software, that sometime can 't work together, i need use VMware . it also more convenient, when you change your laptotp, you don't need to reinstall all software (that can take some days).

maybe i should take a XPS15, that is more powerful. but the format of ths XPS 13 is very good, i like it...

anyway, i'm pretty sure, that something is wrong somewhere, and i would like really to can have comparaison data with nearly same type of CPU/MACHINE/VMWARE.

i hope u understand that!

Reply
0 Kudos
Mits2020
Hot Shot
Hot Shot

I do understand, but what I don't understand is how you manage engineering software on a 13 inch screen... Maybe I'm an old fart (62, although I don't wear glasses for the screen) but what I have learned is that 'the better is the enemy of the good' and you should sometime draw a red line and enjoy your life and your existing gadgets... If you really want to finish your job earlier and have spare time for the beach, better invest the money in a (much cheaper for the same specs) desktop to perform the hard work and use a remote desktop connection from a cheap laptop to just supervise the job. Guaranteed success (in your place I'd take the opportunity to try an AMD desktop solution, the only thing I still dream of doing!)

Reply
0 Kudos
sapeur3873
Contributor
Contributor

on my job, usually i travel a lot,

when i work on site, i have not all time power around me,

i need to move for test some industrial machine, sometime i use the computer with one hand, and maintain on the other, and i found that the XPS is a very robust laptop, good battery life, and good performance

i like really the XPS13, is nearly one of the best pc , i used, and i really not complain about it.

i'm complain only about the loss of performance when i used the vmware, and i would like to understand...

it really not a question to finish earlier my job, but only i like when i start something it start fast, like it should do...

after that i really enjoy my life, my gadget and all around me... (thank for the advice, i take it, and at 62 you are not too old, but maybe more wiser!!!) 🙂

sur i should take an XPS15 maybe better performance, but i loose in compacity, and traveling...

thank to you for your assist

Reply
0 Kudos
sapeur3873
Contributor
Contributor

and my sunday wwas very good, outside, to do some wood stuff, have a run, have a good times with my son!

Reply
0 Kudos
louyo
Virtuoso
Virtuoso

Great post. Not sure why but you got my interest. I have little use for benchmarks but I installed Geekbench on my ancient Linux hosted Dell and 2 Linux VM's. Varied the number of cores and ran the cpu test. Running bare not in a Browser. CPU benchmark in a browser? Really?
On the base machine, 3 VM's (2 Linux and 1 W10) are running but little to no activity.

Dell Inc. Precision Tower 5810
964 Single-Core Score

5726 Multi-Core Score

Geekbench 5.1.1 Tryout for Linux x86 (64-bit)
Result Information
Upload Date     May 17 2020 06:55 PM
Views     1
System Information
System Information
Operating System     Linux Mint 19.3 Tricia 5.3.0-51-generic x86_64
Model     Dell Inc. Precision Tower 5810
Motherboard     Dell Inc. 0K240Y
Processor Information
Name     Intel Xeon E5-1650 v3
Topology     1 Processor, 6 Cores, 12 Threads


(Removed by moderator in respect to 2.4 of https://www.vmware.com/content/dam/digitalmarketing/vmware/en/pdf/downloads/eula/universal_eula.pdf )

IMHO, if performance is that important, one should use a bare metal virtual host (we use ESXi) or run on the hardware. I do program testing and run W10 and W7 VM's on a MacBook using either USB 3.0 or Thunderbolt. Performance is OK, good enough for most testing.
The problem with laptops, it is difficult to run the VM on an SSD drive other than the one the OS is using. My VM's above run on a separate SSD drives, one is a NVMe.


-- Give a man a fish; you have fed him for today. Teach a man to use the Internet and he won't bother you for weeks, or even months unless you give him your email address.  ¯\_(ツ)_/¯
Reply
0 Kudos