News VMware Caps Per-CPU Fees at 32 Cores, AMD's EPYC Rome Impacted

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Honestly, this makes sense for anything that licenses by physical CPU. It will affect people who have migrated to EPYC thinking they were getting around VMWare licensing, but it does not change the value proposition from a CPU perspective when comparing EPYC to Xeon. 1-64 core EPYC still equals 2+ 28 core Xeons. VMWare will charge more now for the EPYC, but it is the same they would charge to get the same number of cores using Xeon. If you are considering either of these CPUs, you probably have a high core number you are targeting, in which case the licensing for EPYC vs Xeon comes out about the same anyway, and over large enough numbers works out better for EPYC because you get 32 cores per license instead of 28 (256 cores with EPYC is 4 CPUs and 8 licenses, whereas Xeon is 10 CPUs [9.14] and 10 licenses).
 
  • Like
Reactions: Olle P
What is Microsoft's licensing structure currently? I would love to blame VMware for these licensing woes, as I don't see why VMware didn't go straight to per core, or even per thread licensing. If you're gonna change models, might as well change all the way. As they always say, "Never attribute to malice that which is adequately explained by stupidity", although the opposite of that saying is becoming more true by the year.

https://community.hpe.com/t5/Allian...ing-explained-VIDEO/ba-p/6995818#.WnM_566nFdE

Core based licensing, not processor based licensing.

https://www.aventissystems.com/Windows-2019-Datacenter-32-Core-p/800261.htm

https://www.aventissystems.com/9EM-00653-2-Core-Pack-p/800252.htm

They did it in around 2017. Back then best EPYC was 32 core so 64 cores on a dual socket system or if you went Intel you could get quad socket with 18 cores for a total of 72 cores so this does affect Intel as well probably more so considering Intel has the majority of the server and HPC market which means more VMWare systems (we use VMWare).
 
Using a slightly different logic, they could have decided to charge per contiguous piece of compute silicon, so the big EPYCs would require 8 licenses, one for each chiplet. The justification would come from the claim that AMD uses multiple chiplets in one package to skirt multiple socket licensing.

I recommend to assume that will happen, and move to free solutions ASAP, if the potential is a concern.
 
Look out! Chemtrails!!!1!

Someone hasn't seen newer intel products with many, many cores.


56 cores at 400+ watts

Xeon 9200 is outrageously expensive, is soldered to the board and so far I haven't seen any big OEMs like Dell or HP pick up a design with it.

That means no nice extra features for management like iLO or iDRAC and god help you if you have a board or CPU failure. You're throwing the baby out with the bathwater on the replacement.

Sure, Cooper Lake MIGHT fix that, but at 14nm++, i'm doubtful they'll be able to drop wattages on the socketed part significantly without also tanking clock speeds.

AMD is significantly advantaged on any workload that isn't exceptionally clock speed sensitive like HFT.

Even with the increased licensing cost, AMD still might make more sense as some vendors are offering single socket solutions which cost quite a bit less than the two socket variants.

A 64 core Epyc 7742 is $7500 list vs a Xeon Platinum 8280M at 28 cores and $13000 list

128 lanes of PCIe gen 4 provides more than ample IO for NVMe, 100GbE, GPUs or all of the above.

Even with VMware licensing being the same between a single 7742 and a pair of 8280m's, you're losing 8 cores and paying 3.5x as much in CPUs which is going to significantly inflate your per-node cost.

A real fast comparison of list prices of the R6515 with a 7742 and 12 x 32gb dimms vs a R640 with a pair of 8280m with all other bits being identical: Network card, PERC, drive config, drive cage config.

The R640 was $38k and the R6515 was $18k

Yes, volume customer DOL will bring both prices down significantly, but that's a huge price disadvantage for Intel



For virtualization, I fail to see any compelling reason NOT to go AMD here.
 
Not like Intel has a known history of making backdoor deals for keep competitors out of the market with 3rd party vendors and made a settlement admitting as much.
That settlement was also about AMD giving GloFo a licence to make x86 CPUs without legally having any right to do so...if intel did not agree to that settlement GloFo would not have been able to make any CPUs for AMD until the final decision of the courts,intel actively saved AMDs butt by agreeing to that settlement.
 
Or, you know, VMWare might just not want to lose profits from companies switching to servers with fewer physical processors running the same workloads. The alternate option would be to raise prices across the board, effectively punishing companies not upgrading their processors to higher core-count models. ...
one does so using half the sockets, why should the other arbitrarily get charged twice as much...?
That's my thought also. They have to draw the line somewhere.

Why the step change at 32-cores?
Probably because it's 2^5. Would have been much worse if the limit was 30 cores.

Honestly, this makes sense ... It will affect people who have migrated to EPYC thinking they were getting around VMWare licensing, but it does not change the value proposition ... when comparing EPYC to Xeon.
My thought also. The main difference is that the cost of licenses no longer will be halved when going from Xeon to Epyc, but the total (running) costs will still be reduced.
 
A 64 core Epyc 7742 is $7500 list vs a Xeon Platinum 8280M at 28 cores and $13000 list

For virtualization, I fail to see any compelling reason NOT to go AMD here.
Meh,you can save about $4000 per 128gb of ram when going with intel let alone being able to have much more ram than going with conventional dimms, and we all know how much ram VMs can gobble up...
It's not all that cut and dried, there are many things to consider about both camps.
https://www.tomshardware.com/news/intel-optane-dimm-pricing-performance,39007.html
qPOpvvB.jpg
 
Not like Intel has a known history of making backdoor deals for keep competitors out of the market with 3rd party vendors and made a settlement admitting as much.

Settlements are not admissions to guilt. Companies tend to settle more often than draw it out in more expensive court costs.

Not with our SQL Server licencing agreement, it's still per-core, full stop.

That was sort of my point but SQL Server went that way with 2012 and Windows server with 2017. Instead of going based on physical processor which would be cheaper for both Intel and AMD.

Point being is I doubt this has anything to do with Intel and more to do with VMWare trying to make more money.
 
That was sort of my point but SQL Server went that way with 2012 and Windows server with 2017. Instead of going based on physical processor which would be cheaper for both Intel and AMD.

Point being is I doubt this has anything to do with Intel and more to do with VMWare trying to make more money.
Looking at MSSQL licensing it is based on vCPUs allocated in a VM environment, not based on total physical CPU cores like Server 2016/2019 is for a VM environment.

Meh,you can save about $4000 per 128gb of ram when going with intel let alone being able to have much more ram than going with conventional dimms, and we all know how much ram VMs can gobble up...
It's not all that cut and dried, there are many things to consider about both camps.
https://www.tomshardware.com/news/intel-optane-dimm-pricing-performance,39007.html
qPOpvvB.jpg
That actual savings in quite a bit less than $4k 128GB DIMMs. Right now consumer pricing from Crucial for 128GB DIMM is $1600 https://www.crucial.com/usa/en/memory/lrdimm so the savings is still a lot but if you don't need the NV ability of Optane you are better off performance wise with DRAM.
 
Looking at MSSQL licensing it is based on vCPUs allocated in a VM environment, not based on total physical CPU cores like Server 2016/2019 is for a VM environment.


That actual savings in quite a bit less than $4k 128GB DIMMs. Right now consumer pricing from Crucial for 128GB DIMM is $1600 https://www.crucial.com/usa/en/memory/lrdimm so the savings is still a lot but if you don't need the NV ability of Optane you are better off performance wise with DRAM.

It depends on how you use it. It is sold in a minimum of a 2 core pack or you buy it with CALs. Either way its more expensive than it used to be which was you buy the software and it works on the hardware you put it on.

Either way my purpose for using it was to show that companies have been moving towards more expe4nsive licensing for higher core counts before AMDs EPYC 64 core chips came out and that its just a way for VMWare to make sure they make more money on larger core count systems.
 
  • Like
Reactions: jeremyj_83
You known, VMware did this a few years ago in 2016/2017 and changed their costing models and upset many of their customers. They conducted Mandatory audits some got hit with two audits even after EA agreements were signed. Example on consumer saw 75% increase in there compute EA. WOW,! Now 3 years later, your doing it all over again . I understand that VMware what to make money, stock increases, but when your product is now costing 15% more then the hardware it runs on. HOUSTON ! Should saw Palo Alto. There is a problem, and this is no business model or solution to these goals in the end. "Just raise the costs". Customer will be looking to dump VMware on the next financial breakdown or just run other technologies on back to Bare metal. There are many other products that can cover this compute space even in Open Source; Xenserver, KVM, Proxmox, Kubernetes, Hyper-V (If licensed in your EA, then your paying for two Hypervisors) and others. Many at little costs and Yes can still get enterprises support and help.
 
You known, VMware did this a few years ago in 2016/2017 and changed their costing models and upset many of their customers. They conducted Mandatory audits some got hit with two audits even after EA agreements were signed. Example on consumer saw 75% increase in there compute EA. WOW,! Now 3 years later, your doing it all over again . I understand that VMware what to make money, stock increases, but when your product is now costing 15% more then the hardware it runs on. HOUSTON ! Should saw Palo Alto. There is a problem, and this is no business model or solution to these goals in the end. "Just raise the costs". Customer will be looking to dump VMware on the next financial breakdown or just run other technologies on back to Bare metal. There are many other products that can cover this compute space even in Open Source; Xenserver, KVM, Proxmox, Kubernetes, Hyper-V (If licensed in your EA, then your paying for two Hypervisors) and others. Many at little costs and Yes can still get enterprises support and help.

This will affect larger companies more than smaller companies. It will also be hard to move to another VM system as they would have to rebuild their entire environment across their entire network instead of just one at a time especially if they have redundancy built in.

Its similar to why some companies will stick with Intel even if AMD has better $/core. Its not easy to convince the decision makers to spend a ton of money and man hours when something is currently working just fine. On top of that there is the re-training on new hardware and software that takes time and money to become efficient on.
 
Meh,you can save about $4000 per 128gb of ram when going with intel let alone being able to have much more ram than going with conventional dimms, and we all know how much ram VMs can gobble up...
It's not all that cut and dried, there are many things to consider about both camps.
https://www.tomshardware.com/news/intel-optane-dimm-pricing-performance,39007.html
qPOpvvB.jpg

You're smoking crack if you think Optane is the end-all be-all solution to large memory spaces.

Vendor support has been slow and performance is lower than traditional DIMMS. Pretty much all of my customers have reviewed it and only 1 has bit on it.
 
You're smoking crack if you think Optane is the end-all be-all solution to large memory spaces.

Vendor support has been slow and performance is lower than traditional DIMMS. Pretty much all of my customers have reviewed it and only 1 has bit on it.
Also while VMs can take up a large amount of RAM, 1TB of DRAM isn't that expensive right now. The biggest hog of RAM are HANA DB's that for production the smallest size is 256GB RAM. SQL Server or Oracle PRD will run 32GB maybe 64GB but use a lot more CPU. For example I am running 37VMs on one of our Dual Epyc 7502's with 1TB RAM and am only using 256GB RAM right now but have 83 vCPUs allocated.
 
  • Like
Reactions: AndrewJacksonZA
You're smoking crack if you think Optane is the end-all be-all solution to large memory spaces.

Vendor support has been slow and performance is lower than traditional DIMMS. Pretty much all of my customers have reviewed it and only 1 has bit on it.
You're smoking crack if you think that I implied anything like that.
My point is just that there is more to think about than just which CPU has more cores for less money.
I just provided one scenario against this argument.
"I fail to see any compelling reason NOT to go AMD here. "

You said yourself that at least for one client it made sense to go with optane so what's your problem?!
 
SQL Server or Oracle PRD will run 32GB maybe 64GB but use a lot more CPU. For example I am running 37VMs on one of our Dual Epyc 7502's with 1TB RAM and am only using 256GB RAM right now but have 83 vCPUs allocated.
Depends on your use case. Our data warehouse running on tin only has 36C72T, but we use 512GB of RAM (and we want more, lots more. And more storage.)
 

TRENDING THREADS