AMD EPYC Market Share Gains in 2018 Our Take

[Patrick Kennedy]

We give our take and context on the reported AMD EPYC market share gains in 2018 and discuss why 2019 is set to be an even more exciting year in the industy

The post AMD EPYC Market Share Gains in 2018 Our Take appeared first on ServeTheHome.

Continue reading...

 

[cheezehead]

Next server refresh, we are seriously looking at going to 1P EPYC configs vs existing 2P Xeon builds. Most backup and hypervisor licensing is done per socket so there's a significant advantage there for some. The issue is with vendors who do per-core licensing *coughs Microsoft*....where single-threaded IPC matters vs spreading the load over more cores.

The other issue is around power consumption, EPYC draws more at the wall so those in areas with extremely high KWh costs may think twice similarly to those who are running single-phase where certain configs push beyond what certain vendor power supplies offer.

 

[Patrick]

On March 4 we are going to have an interview with someone at a large server OEM. I specifically asked about single-socket servers @cheezehead

It may be one to read.

 

[Edu]

EPYC market share is mostly Cloud providers and hyperscalers. I assumed that these customers buy their servers directly from ODMs (or am I mistaken?). My point is that they wouldn't be using Dell or HPE servers, so the theory that EPYC sales were held up by lack of servers from big OEMs is false.

 

[Patrick]

EPYC market share is mostly Cloud providers and hyperscalers. I assumed that these customers buy their servers directly from ODMs (or am I mistaken?). My point is that they wouldn't be using Dell or HPE servers, so the theory that EPYC sales were held up by lack of servers from big OEMs is false.

The 2018 uptake in the cloud was not as big as some people would assume. Many of the major ODMs would not even announce they were working on an AMD EPYC platform.

I can tell you, a smaller cloud player like a Packet is using the PowerEdge servers for their EPYC instances because the ODMs were not stepping up. When they were picking a model, Tyan did not have what they needed in 1P and Supermicro had been focused on 2P designs early, as was HPE. They went Dell because about a year ago they could not source what they wanted and Dell was first to market with what could work.

That type of experience was not uncommon.

In 2019, cloud providers will drive unit sales. In 2018, I heard over-and-over that lack of platform ecosystem was slowing trials and deployments.

 

[Edu]

That's quite interesting. Thanks for the info.

 

[Evan]

@cheezehead that 1 socket license advantage probably won’t last that long, as Microsoft did they will find another way to extract $$

Power consumption is a topic but I don’t believe it’s nearly as bad as it first looked, it’s not that far off intel scalable.

 

[cheezehead]

@cheezehead that 1 socket license advantage probably won’t last that long, as Microsoft did they will find another way to extract $$

Was thinking on the VMware & backup software licensing. M$ core charge on datacenter is just going to hurt the next refresh.

 

[Evan]

Was thinking on the VMware & backup software licensing. M$ core charge on datacenter is just going to hurt the next refresh.

If it’s under software assurance would have been wise to pump the number or cores up before MS gave the equivalent core license free.
If your just buying the license then yeah it is going to hurt or else you tend to start again looking for say single 16-cores and high frequency.

Much like it’s a good way to license DB’s, low core count and high frequency.