Why are server parts so expensive

Like wtf, the TR 1950X cost's 718 yuros right now. That is 16c/32t @ stock 3.4Ghz base to 4 Ghz turbo. The same processor but branded "Epyc" costs nearly the same but can only go from 2.1 GHz to 2.7GHz turbo. Not only that but the cheapest server mobo costs twice as much as the desktop one. Same thing for the heatsinks, a shitty piece of copper that is supposed to be the server cpu cooler is 40€. And a Cooler Master with 6 heat pipes and being much more capable of cooling a TR with two fans included is 46€.
>Inb4 is not the same cpu
It literally is, same cache, both can do ECC, same PCIe lanes, same everything. The only different thing is that the server cpu has slighlty less tdp with at half the speed.

Attached: komi-san-is-my-waifu.png (2373x563, 392K)

>It literally is, same cache, both can do ECC, same PCIe lanes, same everything. The only different thing is that the server cpu has slighlty less tdp with at half the speed.
Source: Your ass.

Generally because servers are purchased by businesses with credit.

Because corporations will pay those prices for servers.

Btw I want a virtualization server. Is 16c/32t enough?

Wait till you see RAM prices.

Threadripper 1950X:
16 Cores 32 Threads
Operating frequency: 3.4GHz
Max Turbo Frequency: 4.0 GHz
L1 cache: 1.5MB
L2 cache: 8MB
L3 cache: 32MB
Memory types: DDR4 DDR4-ECC
Memory channel: 4
Virtualization Technology Support: Yes
PCI Express Revision: 3.0
PCI Express Lanes: 128
Thermal Design Power: 180W

EPYC OEM 7281:
16 Cores 32 Threads
Operating frequency: 2.1GHz
Max Turbo Frequency: 2.7 GHz
L1 cache: 1.5MB
L2 cache: 8MB
L3 cache: 32MB
Memory types: DDR4 DDR4-ECC
Memory channel: 4
Virtualization Technology Support: Yes
PCI Express Revision: 3.0
PCI Express Lanes: 128
Thermal Design Power: 155-170W

Source: Amd webpage

But I understand why server ram is more expensive. They've got ECC and are usually more reliable and work at lower voltages. But this cpu thing is stupid, and Intel is way worse. A shitty 16c/32t server cpu costs 1k min,

Epycs are 4 chips with 8-channel registered memory
>Amd webpage
Bullshit. It's correctly shows as 8-channel on amd website.

>Threadripper 1950X
>PCI Express Lanes: 128
>EPYC OEM 7281
>Memory channel: 4
WRONG!

Attached: 1504438409155.jpg (640x617, 49K)

>Why are server parts so expensive?
Because businesses can pay more. Businesses also demand 24/7 uptime which means parts have to be validated for reliability.

To make it clear, epyc supports 8 memory channels up to 2TB (4 channels at 128GB on TR) and 128 pcie lanes and is able to be used in 2 socket motherboards. There are cheaper 1 socket EPYC P chips available that have a more palpable price tag. Plus the 1950X has had a price reduction from 999dollars.

There's also expected service times as well. An enterprise part will be on the market and supported for far longer. Manufacturing and maintaining stock is a bigger issue than you think and that's recuperated through higher costs as well.

Just buy old Xeon processors if you want to save money...Not everything has to be new.

Server grade silicon =/= consumer grade silicon.

When you buy server grade, you get parts built, tested and verified for 100% uptime.

Muh businesses can spend that money is not the reason.

apart from a lot more validation and cost associated with it.. willingness to pay more is probably a factor. AMD can charge more while still being able to charge less than Intel's offerings so corporations will buy them. Keep in mind that a lot of the things such servers are used for are critical in a way that will cost money. 1000 EUR isn't very expensive if that's less than what one hour of downtime would cost you.

1.) Market segmentation : People buying server-tier hardware are making money out of it for real world work

2.) Server-tier hardware undergoes far more rigorous validation and platform is built for fault tolerance

It's mainly validation. Enterprise grade shit goes through lots of testing to ensure it works in . Just because something is supported does not mean it's been validated to work in all workloads/scenarios with 100% uptime.

Get a R7 w/ 32gb ram for the price of a TR instead

Short answer : yes
Longer answer : if we don't know how many fucking VMs you'll be running, nor what the fuck those VMs will be doing, nor how much fucking responsiveness you want from your VMs how the actual are we supposed to answer that question ?

Because aside from businesses, only a terrorist would want to host his own server instead of renting one from a reputable company that opens its doors widely each time the gov asks

In general, you want 1 core per VM and save a core or 2 for the host and HDD access. But RAM and HDD throughput is often a greater limit to VMs then # of cores. You want 4 GB per VM? Then you really should get 4 GB of physical RAM per VM plus a large amount for the host, which adds up quickly.

Things a server needs but a desktop doesn't :
- 6 or more SATA ports
- 4 or more RAM slots
- 2 or more gigabit ethernet
- 2 or more CPU sockets

Also, # of cores is important, speed of each core less so.

Also :
- 2 or more PSUs