• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Intel Hyper-Threading with ESXi

K

KapsZ28

2[H]4U
Joined
May 29, 2009
Messages
2,114
How does Hyper-Threading work with ESXi? I have a quad core with Hyper-Threading. It shows 8 logical processors, but the CPU Capacity only shows 4 x 3.4 GHz.

 
Hyper threading gives you about a 10% boost in the virtual world when it comes to consolidation ratios. The CPU can better handle all the different threads being thrown at it from vCPUs.
 
So with ESXi it would be better to get an AMD 8 core CPU at 2.8GHz rather than an Intel quad core with hyper-threading at 2.8GHz?
 
The per cor performance of Intel has been far superior to AMD for quite some time now. But it really depends on what you're doing...not a lot of things are CPU bound these days. With virtualization it really comes down to IOPS in most cases. Looking at your picture of CPU load I'd say go the cheap route lol
 
If you want a greater consolidation ratio and expect a lack of cores to be a bottleneck, then the 8 core AMD would be better.

If you want better performance per vCPU then the Intel would be a better choice.
 
I always push Intel when picking Performance but you need to pick the right ratios.
An intel 8 core 16 thread @ 3.2GHz plus turbo will smoke an AMD 16 core @ 2.6Ghz.

If a VM can get processor time at 3.5Ghz turbo it will process more in less time, resulting in less needed cores.
But, as always, it depends on your workload.

Nick
 
Take this for example. Our cluster usually averages around 70% CPU usage during business hours. See picture below.

Those servers all have two Intel Xeon X5660 6 core CPUs with Hyper-Threading. So the CPU capacity is 12 x 2.8 GHz for a total of 33600 MHz. Those CPUs used to cost over $1,000.

Now if you look at an AMD Opteron 6378 that costs less than $1,000 and comes with 16 cores per CPU. Get two of those and you have 32 x 2.4 GHz for a total of 76800 MHz. That is more than double the capacity of the Intel. Part of what we do is sell Resource Pools to our clients. So if they are buying 24 vCPU, they will get 67200 MHz allocated to them.

If we compare 10 servers, all with the Intel CPUs we have verse the AMD I mentioned, this is what you get for resources.

Intel = 336000 MHz
AMD = 768000 MHz

That is a pretty significant difference. So are you saying that Intel is so much better that it will still outperform AMD even with less than half the resources?


 
Selling resources should be much more than just a GHz reading. It should be based on the number of cores needed (physical or logical, take your pick), RAM needed and the availability of those resources. Using the metrics you are using is like comparing only horsepower numbers of a car and stating the one is better than the other. Bottom line; there is more to the equation than what I being factored in.
 
peanuthead said:
Selling resources should be much more than just a GHz reading. It should be based on the number of cores needed (physical or logical, take your pick), RAM needed and the availability of those resources. Using the metrics you are using is like comparing only horsepower numbers of a car and stating the one is better than the other. Bottom line; there is more to the equation than what I being factored in.
Click to expand...

Resource Pools are done by MHz, not cores. That is why we multiply how many cores the client wants, by the MHz of the CPU. They can use more cores than they purchased in their Resource Pool, but the reservation will only allow so many MHz to be utilized at once.

Based on what everyone is saying, it sounds like Intel CPUs will utilize less resources to do the same job as AMD.
 
KapsZ28 said:
Resource Pools are done by MHz, not cores. That is why we multiply how many cores the client wants, by the MHz of the CPU. They can use more cores than they purchased in their Resource Pool, but the reservation will only allow so many MHz to be utilized at once.

Based on what everyone is saying, it sounds like Intel CPUs will utilize less resources to do the same job as AMD.
Click to expand...

I understand that resource pools use Mhz as the share value, etc. I was thinking more of the SLA standpoint. My fault for not completely understanding the question. I agree, Intel would be the better option.
 
Just a quick example of what I just played with:

I have a dual socket blade with two Intel E5-2680 Cpus @ 2.7Ghz
(2 X 8 core cpu = 32 threads)

I replaced the two CPUs with two E5-2697 v2 @ 2.7Ghz
(2 X 12 core cpu = 48 threads)

Same Ghz, more cores, but new 22nm arch.
All of the VMs are performing 50% better.
All of the applications are reporting that jobs complete 30-60% faster.


This isn't because CPU resources were low or high overcommit ratio.
Granted, you could argue that each VM could be scheduled more often but that would only be a factor if we had more then 32 vcpus allocated on the host, (16vcpus if you want to argue the hyperthread core).

I have yet to see an AMD CPU swap work this well.
AMD is catching up with their Memory and Bus speed but the way their turbo works still needs some tweaking..


As for the charge back discussion. The more processing you can pack into your server/rack/datacenter, the more resources you can sell. Dealing with how you block those resources and sell them is a whole different story.

Nick
 
I am in the vCenter 5.5 class so I can get my VCP and this instructor seems a bit off. He made a blanket statement that it is recommended that Hyper-Threading should not be enabled. Funny that VMware recommends enabling it for single and dual socket boards and says it may or may not provide an improvement on quad socket boards and you should test before enabling in production.

He also states that Ghz is Ghz. Doesn't matter if it is AMD or Intel.

Then started saying that a quad core CPU at 3 Ghz is 3 Ghz divided by 4 for your overall clock that can be used rather than 4 times 3 Ghz.
 
KapsZ28 said:
I am in the vCenter 5.5 class so I can get my VCP and this instructor seems a bit off. He made a blanket statement that it is recommended that Hyper-Threading should not be enabled. Funny that VMware recommends enabling it for single and dual socket boards and says it may or may not provide an improvement on quad socket boards and you should test before enabling in production.

He also states that Ghz is Ghz. Doesn't matter if it is AMD or Intel.

Then started saying that a quad core CPU at 3 Ghz is 3 Ghz divided by 4 for your overall clock that can be used rather than 4 times 3 Ghz.
Click to expand...

Well a true quad core (no HT) at 3Ghz (assuming no turbo boost for a moment) is 4x 3GHz. The Summary and Resource Allocation tabs in the C# client state differently. In a simplified form, Ghz is Ghz to a VM. The execution of those instruction sets is where the difference is. I personally ignore Ghz as a metric most of the time. In regard to HT, there are cases were it's better and others it's not. I've heard to disable it too. However, until I find that there is finite proof it's better over and over again I will continue to leave it on. In the best practices in vSphere 5.0 from VMware they recommend you turn it on (page 20).
 
KapsZ28 said:
I am in the vCenter 5.5 class so I can get my VCP and this instructor seems a bit off. He made a blanket statement that it is recommended that Hyper-Threading should not be enabled. Funny that VMware recommends enabling it for single and dual socket boards and says it may or may not provide an improvement on quad socket boards and you should test before enabling in production.

He also states that Ghz is Ghz. Doesn't matter if it is AMD or Intel.

Then started saying that a quad core CPU at 3 Ghz is 3 Ghz divided by 4 for your overall clock that can be used rather than 4 times 3 Ghz.
Click to expand...

In the past there were some issues with VMware and hyper-threading. I haven't noticed as much lately. Also his assertion that Ghz is Ghz is completely incorrect. While both AMD and Intel use similar architecture, the instruction sets are not completely the same. They both have their own little variations. Intel has proven lately to be far more effective and quicker in how it handles its processing. That being said...you are dealing with an abstraction layer between the processor and your OS's. That abstraction layer does affect performance, and nullifies some of the advantages that Intel has.

In the end, the question when you are doing a virtual farm is not so much which processor is the absolute best, but rather how are you going to achieve the most reliable resources for your money. And a lot of that also depends on what exactly you are running in your virtual appliances. Many are finding these days that AMD is a better value for their farms, but its not always the best performing or the most reliable. I would say for the average home grown farm, it would probably be more economical to go with AMD. Remember on top of this is also the availability and variety of motherboards and what options they provide. And if you aren't building your own servers, you have to look at what is available by resellers.

Apart from all that, listening to the comments on what your teacher is saying, they sound a little whacky. I would almost question if they have a true handle on some of the technologies they are talking about. Especially that comment about the quad core CPU dividing by 4... It doesn't quite work like that, especially with the latest chips.
 
You must log in or register to reply here.
Back
Top