Mercury Research Says AMD Made Gains in the x86 Market

AlphaAtlas

[H]ard|Gawd
Staff member
Joined
Mar 3, 2018
Messages
1,713
AMD claims they've been making significant gains in x86 market share, but a recent report from Mercury Research attaches some hard numbers to those claims. Their Q4 2018 report says that AMD grabbed 3.2% of the lucrative x86 server market in Q4 2018, which is a massive jump from the 1.6% share the previous quarter and the 0.8% share the previous year. Meanwhile, their desktop market share grew to 15.8%, the highest it's been since Q4 2014, while their notebook marketshare grew to 12.1%, up from 6.9% a year ago.

As a reminder, Ryzen desktop processors launched in March 2017. First gen EPYC processors launched in June that same year, while AMD's Raven Ridge APUs started shipping in laptops in October. Overall, it looks like the Zen architecture has steadily been clawing back x86 marketshare that AMD lost in recent years, and with the imminent launch of their 7nm products, the company seems to think that upwards trajectory will only continue.

A note about server share. Mercury Research captures all x86 server class processors in their server unit estimate, regardless of device (server, network or storage), whereas the estimated 1P and 2P TAM provided by IDC only includes traditional servers. We used IDC's server forecast of the 1P and 2P server TAM of roughly 5M units to compute our server market share estimates. We believe that in Q4 2018 we achieved ~5% unit share of the 1P and 2P server market addressed by our EPYC processors (as defined by IDC).
 
The server marketshare is fantastic news. Especially if they can make more gains in the AWS and Azure space.

Cisco UCS isn't putting them in their blades, which makes me sad. We're kinda locked into those.
 
I personally am not a fan of the cisco UCS blades. For mission critical systems it's far too many eggs in one 'basket' for my tastes. Plus that shared bandwidth plan is ALWAYS never enough.
 
I want to see AMD get market share. The problem going forward for AMD is executing on a timely manner. Already Intel is having stock and availability issues for their top tier CPU's. If AND can deliver on performance, stability, AND availability they will get the server market quickly. And if they can get a contract with Intel and Oracle to lower the per CPU licensing cost. Good god that would see them take the server market by STORM.

EDIT Adding: Currently for Intel SQL Enterprise Core licensing you're paying like 20k for 4 cores of licensing. So it makes more sense to get faster lower core count CPU's for that environment. This means that AMD's amazing 32 core and 64 core CPU's in the near future are scary from a licensing standpoint. Why in gods name would I want 64 cores when the cost of licensing could almost build me a new datacenter?

Where they have a win is the per socket licensing like Vmware. Problem is when organizations are putting together orders for new server hardware unless they are ONLY refreshing vm environments they generally think one or the other.
 
Last edited:
I want to see AMD get market share. The problem going forward for AMD is executing on a timely manner. Already Intel is having stock and availability issues for their top tier CPU's. If AND can deliver on performance, stability, AND availability they will get the server market quickly. And if they can get a contract with Intel and Oracle to lower the per CPU licensing cost. Good god that would see them take the server market by STORM.
What licensing costs with Intel are you referring to?
 
This is good news. Hopefully by 2020 we can start seeing some AMD laptops that are equipped with 1st class components instead of always being given garbage or previous generation remaining stock.

That will only happen when amd grows enough to afford the support required by oems ... And if their competitors stop doing shady if not illegal deals.

One can only hope
 
For Oracle at least, the issue was that the licences based on the processor formula definitely favored Intel due to Intel's higher IPC. You take the total of all cores across all sockets and multiply it by a core licencing factor and voila that's how many licences you need. AMD and Intel x86 products share the same core factor, so the 'more slow cores' approach can be much more expensive than fewer faster cores. The cost of the licence far outweighs the actual hardware cost. But if Epyc can level the field a bit.....

Scroll to the end for the definition of a processor based licence
https://www.oracle.com/assets/technology-price-list-070617.pdf

Core Factor Table used in the calculation
http://www.oracle.com/us/corporate/contracts/processor-core-factor-table-070634.pdf

I know for most of us it's just academic, but it is kinda interesting to see where all the different architectures fall in the list.
 
I personally am not a fan of the cisco UCS blades. For mission critical systems it's far too many eggs in one 'basket' for my tastes. Plus that shared bandwidth plan is ALWAYS never enough.

I hear you. We've got 2 chassis with only 9 blades between them. But with 40gbs of throughput to each for both storage and network we've yet to have a problem.I wouldn't hesitate to add a 3rd chassis once ours are fully populated but that might take a long while.
 
Going to be ordering some new 2Us at the end of the year/beginning of next. HPE offers Epyc, so here is hoping they compete well against what Intel will have out then.

Looking for a balance of most cores with most speed, so I may have suggested a speed optimized 24c part to the AMD rep at the last Ignite conference.
 
If he is anything like me than for the per CPU licensing, it is cheaper to have 1 CPU with 64 cores than 2 CPU’s with 32 each.
I think you misunderstood me. What licenses does AMD pay Intel for. Is he is referring to the x86 license? Because AMD and Intel has a cross-license agreement with no licensing or royalty fees. As Intel owns the x86 patent AMD owns the x86_64 patent.
 
I think you misunderstood me. What licenses does AMD pay Intel for. Is he is referring to the x86 license? Because AMD and Intel has a cross-license agreement with no licensing or royalty fees. As Intel owns the x86 patent AMD owns the x86_64 patent.

No I'm referring to the SQL Enterprise Core licensing cost. For performance UNLESS THREADS ARE NEEDED for simultaneous processing it makes more sense to have fewer cores at higher clock/ipc because it is more effective than throwing more and more cores at a licensing model that is going to effectively charge me 7k+ per core. (thankfully not including logical cores unless you are going to run a large VM.)

How can I explain this easily. Lets say you had a piece of software.. lets say a video game. And it was multi thread friendly. It could use a reasonable number of threads for max performance. But the price you paid for the game was based on the number of physical cores your CPU had. and lets say they were very nice and said we will charge 10 dollars per core licensing fee.

Everyone with a 16 core threadripper that game would cost them 160 dollars.

Everyone with an 8 core CPU would pay 80 dollars.. and so no. It was cores not sockets.

Now when building a PC you may need a lot of cores for multitasking. (great) But you also need to consider the license model of the games you want to run. Someone like me might be just fine with my 4 core 4.6ghz system for gaming on. But someone who's doing streaming might want that 16 core system so they can stream and game off of one system. But now it is a point of concern when building.

That is the market that AMD needs to compete in for the large server farms. Now if they are just cloud hosting/hypervisor running then you are set no worries.

And sure that 16 core system you COULD run a VM and game on the VM with a sliced up video card and everything that entails. But that's a small part of the conversation. Just a way to try and make it more easily understood.

Hope that helped.
 
No I'm referring to the SQL Enterprise Core licensing cost. For performance UNLESS THREADS ARE NEEDED for simultaneous processing it makes more sense to have fewer cores at higher clock/ipc because it is more effective than throwing more and more cores at a licensing model that is going to effectively charge me 7k+ per core. (thankfully not including logical cores unless you are going to run a large VM.)

How can I explain this easily. Lets say you had a piece of software.. lets say a video game. And it was multi thread friendly. It could use a reasonable number of threads for max performance. But the price you paid for the game was based on the number of physical cores your CPU had. and lets say they were very nice and said we will charge 10 dollars per core licensing fee.

Everyone with a 16 core threadripper that game would cost them 160 dollars.

Everyone with an 8 core CPU would pay 80 dollars.. and so no. It was cores not sockets.

Now when building a PC you may need a lot of cores for multitasking. (great) But you also need to consider the license model of the games you want to run. Someone like me might be just fine with my 4 core 4.6ghz system for gaming on. But someone who's doing streaming might want that 16 core system so they can stream and game off of one system. But now it is a point of concern when building.

That is the market that AMD needs to compete in for the large server farms. Now if they are just cloud hosting/hypervisor running then you are set no worries.

And sure that 16 core system you COULD run a VM and game on the VM with a sliced up video card and everything that entails. But that's a small part of the conversation. Just a way to try and make it more easily understood.

Hope that helped.

I understand all of that, which is exactly how Netflix utilizes their CDNs, they only pay for cores, processing power as they need it . What does that have to do with licensing fees with Intel and Oracle? Or do you mean to try and get them to license it from AMD, which i don't see happening? I guess i don't understand what roll you think Intel and Oracle play here for any licensing agreement/contract to take place with Intel or Oracle.
 
Last edited:
What does that have to do with licensing fees with Intel and Oracle? Or do you mean to try and get them to license it from AMD, which i don't see happening?

I'm saying if AMD can get sweetheart deals with Microsoft and Oracle they will be in a better position only because their core count is so much higher. For instance if I could build a 128 core Database server and only have to PAY for 64 of those physical cores in licensing I would lean toward that over intel EVERY time even if the per core performance wasn't as good.

Make sense? If I said intel before I apologize.
 
I'm saying if AMD can get sweetheart deals with Microsoft and Oracle they will be in a better position only because their core count is so much higher. For instance if I could build a 128 core Database server and only have to PAY for 64 of those physical cores in licensing I would lean toward that over intel EVERY time even if the per core performance wasn't as good.

Make sense? If I said intel before I apologize.

Now that makes more sense. Microsoft is completely different than Intel.. Intel is what confused me as having a per core license agreement with Intel didn't make any sense. :)
 
I'm saying if AMD can get sweetheart deals with Microsoft and Oracle they will be in a better position only because their core count is so much higher. For instance if I could build a 128 core Database server and only have to PAY for 64 of those physical cores in licensing I would lean toward that over intel EVERY time even if the per core performance wasn't as good.

Make sense? If I said intel before I apologize.
You could also just build a massive 128 Core Host server, virtualize the Database in that Host and only assign it 64 cores but up that VM's priority and give it the performance of that 128 core system with the costs associated with the 64 cores. Then use the left over headroom on the server for other less taxing subsystems? Performance wise you will be sitting at like 98% of running it on the host hardware itself but spending a lot less.
 
You could also just build a massive 128 Core Host server, virtualize the Database in that Host and only assign it 64 cores but up that VM's priority and give it the performance of that 128 core system with the costs associated with the 64 cores. Then use the left over headroom on the server for other less taxing subsystems? Performance wise you will be sitting at like 98% of running it on the host hardware itself but spending a lot less.

Priority only comes into play if there is a reason for contention and it doesn't give you more than the cores you've allotted to the server.
 
Priority only comes into play if there is a reason for contention and it doesn't give you more than the cores you've allotted to the server.
I was under the impression that if you taxed the assigned cores to 100% they would spill over into other unused cores on the physical host if they were under utilized, just never tested it because I am yet to get to a point where any of the virtual machines tax the CPU that much they are always GPU limited or occasionally HDD limited or Network limited but never yet CPU limited.

That being said I am told that much like there is dynamic memory on most virtual hosts we can expect to see dynamic CPU's in the near future, I would expect during that time you would be licensing the average number of cores used, how they would enforce that is beyond me but I can see it happening.
 
Last edited:
I'm curious to see how Ryzen will do in the portable computer space. AMD does not have a good track record in this space - Intel always had the upper hand. I owned an Athlon based laptop (about 10 years ago, so worth keeping in mind...) - it was a junker. Then again, I probably didn't pay as much for it as I did for the Centrino (or whatever) Intel system of the day. I'm guessing most corporations buy their employees laptops anymore. My last few jobs were exclusively laptops.
 
I was under the impression that if you taxed the assigned cores to 100% they would spill over into other unused cores on the physical host if they were under utilized, just never tested it because I am yet to get to a point where any of the virtual machines tax the CPU that much they are always GPU limited or occasionally HDD limited or Network limited but never yet CPU limited.

That being said I am told that much like there is dynamic memory on most virtual hosts we can expect to see dynamic CPU's in the near future, I would expect during that time you would be licensing the average number of cores used, how they would enforce that is beyond me but I can see it happening.

Interesting I don't recall that being brought up at VMworld but there was a heck of a lot going on so it is entirely possible I missed it. For some of our VM's I like that idea. For others with 'power users' I just know they would destroy my VDI hosts by opening up 60 gig projects and eating all the memory and just leaving them open when they go home.
 
I'm curious to see how Ryzen will do in the portable computer space. AMD does not have a good track record in this space - Intel always had the upper hand. I owned an Athlon based laptop (about 10 years ago, so worth keeping in mind...) - it was a junker. Then again, I probably didn't pay as much for it as I did for the Centrino (or whatever) Intel system of the day. I'm guessing most corporations buy their employees laptops anymore. My last few jobs were exclusively laptops.

Yep,. hard to compare. AMD aim for the cheapos, because laptop manufacturers don't even care about building quality around their flagship products.

The cheapos are what they are, their main issue has always been drivers. I would love to play with a new-ish one just to see if all the driver commitments AMD made actually trickled down to those.

Now, I will love to see a OEM fully committed to building a quality machine around a flagship AMD mobile processor. I hope that'll be easier when they get their driver situation sorted out, making it easier to force manufacturers to follow suit.
 
Yep,. hard to compare. AMD aim for the cheapos, because laptop manufacturers don't even care about building quality around their flagship products.

The cheapos are what they are, their main issue has always been drivers. I would love to play with a new-ish one just to see if all the driver commitments AMD made actually trickled down to those.

Now, I will love to see a OEM fully committed to building a quality machine around a flagship AMD mobile processor. I hope that'll be easier when they get their driver situation sorted out, making it easier to force manufacturers to follow suit.

Amd Doesn't really have a flagship mobile processor. Their Apu lineup is based on Raven Ridge which is just a excavator core. We saw the 2700u, and the 2500u which were released based on the Ryzen core, abiet first gen core, and Vega 10 graphics. These are pretty decent chips, but AMD truely doesn't have a high end offering for mobile space until they release their APU's based on Zen 3.

Laptop makers always cater to what is selling and Intel has had the mobile market since like Sandy Bridge. They outsell AMD laptops 10 to 1. Probably closer to 20:1, but i Digress. If you are going to spend engineering hours on a product you are going to focus those hours on a product which is likely to move larger volumes. Amd is coming back in the CPU space since Ryzen was released and has shown to be competitive. It will take a bit more time before they pick up mobile marketshare, aka probably when their Zen 3 apu's make it inside will people be inclined to buy the AMD powered part over the Intel powered part. It doesn't really help that AMD is Behind in development on mobile vs their Desktop and Server spaces, but is likely due to increased focus in those areas.
 
Back
Top