Server vs Consumer Hardware..

nocloud

I see a lot of providers offering consumer grade hardware, so much so that it's quite common to see Ryzen 5 /7/9, and core i9 / i7 / i5 and even atom systems in some cases.

The other day I noticed ryzen 7 5700G, which intended as a light gaming APU, I found it surprising especially as the (not home not server ) business version ryzen pro 5750G is essentially the same APU at only a very slight price premium but with mem encryption and ECC supported.

I guess my question is, is this trend increasing or is it just a symptom of the lack of silicon due to covid? Or is it done simply to cut cost in an ever more competitive business?

Also does it really make a difference, like if we had an electrical storm or some terrorist drops an EMP / E-bomb, will server hardware stay online and consumer grade be screwed? And in everyday conditions is there any advantage? Like i'm sure banks won't use consumer grade, but would Bob at Bob's Burgers see any real world advantage?

or is it a well ingrained business myth?

hadawayandplay

Consumer grade is fine for non-critical systems where data integrity isn't absolutely essential (things such as financial services). If this is essential then those seeking hosting services should be opting for true ECC memory supported systems and currently consumer grade Ryzen systems don't accomplish that. DDR5 has some capability for a rudimentary form of ECC but it's not the same standard as server grade ECC memory for example.

emgh

ChatGPT (but I managed to get a very good answer):

Feature	Ryzen 9 5950X	EPYC 7543	Relevance to Stability
Core/Thread Count	16 cores / 32 threads	32 cores / 64 threads	More cores allow for better workload distribution.
Base Clock	3.4 GHz	2.8 GHz	Lower base clock may result in less heat and better longevity.
Max Boost Clock	4.9 GHz	3.7 GHz	Lower max clock implies more conservative performance tuning.
TDP	105W	225W	Higher TDP in server CPUs accounts for more cores and is optimized for 24/7 operation.
L3 Cache	64MB	256MB	Larger cache can improve data access times, reducing wear on other components.
Memory Support	DDR4-3200	8-Channel DDR4-3200, DDR4-2666	More memory channels can provide greater stability under high-load scenarios.
ECC Support	Partial (Depends on Motherboard)	Full	Full ECC support provides better data integrity and system stability.
RAS Features	Limited	Extensive (MCA, PFA, etc.)	RAS features help in error detection and correction, enhancing reliability.
Multi-Socket Support	No	Yes (Up to 2 sockets)	Allows for more redundancy and therefore higher reliability.
PCIe Lanes	24	128	More PCIe lanes allow for more devices and therefore more balanced workloads across devices.
Power Management	Standard (P-states, C-states)	Advanced (More P-states, quality of service features)	Advanced power management can lead to better thermal efficiency.
I/O Virtualization	Basic	Advanced (SR-IOV)	Advanced I/O virtualization can improve stability in virtualized environments.
Operating Temp Range	Consumer-grade	Datacenter-grade	Server CPUs are often validated for extreme temperature ranges.

nocloud

i accidentality deleted the poll when editing EEC to ECC. So if you voted already, please re-vote!

chihcherng

If you want to handle large amounts of RAM, going with a server-grade CPU is your only option.

neverain

I host important stuff strictly on reputed providers with ent hardware, I am not sure if it really makes a difference but it doesnt fit right with me when i use Ryzen/I series VPSes.

For hobby stuff i go with ryzen because its faster, i can feel the difference when using it.

I also have had better uptime / stability with providers that use ent hardware

AXYZE

Consumer CPUs are better for CPU-bound scenarios (Ryzen, i9).
Server CPUs are better for bandwidth-bound scenarios (8x+ NVMe, 4channel DRAM), memory (256-2TB RAM) and density (one server, 128cores, 2TB RAM...).

In terms of reliability... some people will kill me for that, but they are the same. You can get nice ASRock Rack board for AM4/AM5, you can get shit Aliexpress mobos for server sockets. It all depends on your choices, cooling, physical restrictions.

Tons of people have servers on $35 Pi or NUC. It works fine for years. Just look at absolutely bottom of the barrel Kimsufi Atom servers that are 10 years old.

CPUs are least failing component of desktops and servers, I wouldnt be worried about it.

host_c

I played around for years, offering servers for Consumers from PC grade systems, and in my opinion, in the short term it works perfectly, until shit goes sideways.

I promised myself, that i will not use a consumer grade system for the rest of my natural born life.

It is not linked to performance, many Ryzen systems outperform XEON'S, it is about the stability and management.

Imagine you have 50+ servers in 3 datacenters, PC's do not have out of band management, ( BMC, ILO, IDRAC ). so if one hangs how do you gain access to it? open a ticket in the DC to hook up a KVM for remote management? that will be time-consuming and expensive. Option B is to use Rasberry like Hardware and put a KVM on int, that is not viable, it will create you more problems than the one it solves.

PC's are not built to run 24/7, the can, but not designed nor engineered that way.

Servers are engineered to run 24/7. ECC memory is vital for system stability, SAS interface opposed to SATA has many benefits.

You can use Server Grade Hardware in desktops, with adding cards like RAID cards, HBA, etc... some AMD cpu's even support ECC Memory, they will never be servers, and at this point, adding server grade hardware to it, will actually cost you as much as a SH server.

But the short answer is TCO, as desktops, will cost you much less than a 2-4 year old Server.

Second fact is trend + cost, as these systems are cheep, have very good computing performance, you can sell cheaper and give good performance, so clients jumped on the train, and it became a trend.

On the long run, if you do not own your colocation space/Data-center, it will cost you much more than Brand-Servers.

MattBGP

@chihcherng said:
If you want to handle large amounts of RAM, going with a server-grade CPU is your only option.

Providing some numbers about this.

Consumer CPUs mostly only support up to 128GB Memory with some rare exceptions that support up to 192GB.
Server CPU's support up to multiple TB of Memory.

host_c

@nocloud said: Also does it really make a difference, like if we had an electrical storm or some terrorist drops an EMP / E-bomb, will server hardware stay online and consumer grade be screwed? And in everyday conditions is there any advantage? Like i'm sure banks won't use consumer grade, but would Bob at Bob's Burgers see any real world advantage?

NO, none have EMP protection, alto Servers are better shielded than PC's, none of them have EMP protection.

EMP ptotection is done at Building Level, not hardware level.

Unless Panasonic or Getac starts making servers, they have military grade laptops that have some EMP protection depending on the model

tentor

@host_c said: PC's do not have out of band management, ( BMC, ILO, IDRAC ).

I heard there are Gigabyte MB with OOB for Ryzens as well as some providers colocating PiKVMs to access their consumer grade servers, but it is so clumsy

AXYZE

@host_c said:
Imagine you have 50+ servers in 3 datacenters, PC's do not have out of band management, ( BMC, ILO, IDRAC ).

Have you tried Intel AMT which is out of band managment for Intel Core desktops with vPro cert (ThinkCentre, HP Z-series) or Realtek DASH (some Ryzen board s), ASRock Rack BMI?

host_c

@tentor said: I heard there are Gigabyte MB with OOB for Ryzens as well as some providers colocating PiKVMs to access their consumer grade servers, but it is so clumsy

exactly, it is like tuning your 2L engine to have 500 HP, it will work for short period of time, but you cannot go LE MANS and finish with that same engine and compete with the other guys that have those cars built to out stand 24H of speeds over 200 KM/h

host_c

@AXYZE said: Have you tried Intel AMT which is out of band managment for Intel Core desktops with vPro cert (ThinkCentre, HP Z-series) or Realtek DASH (some Ryzen board s), ASRock Rack BMI?

Either way, it is a desktop, 8 ours / day, 5 days / week. if you have 50+ of thoes, you will not sleep my friend.

I tested AMT, it is a bad joke, better than nothing, bad joke compared to ILO and IDRAC

FlorinMarian

@host_c said:
I played around for years, offering servers for Consumers from PC grade systems, and in my opinion, in the short term it works perfectly, until shit goes sideways.

I promised myself, that i will not use a consumer grade system for the rest of my natural born life.

It is not linked to performance, many Ryzen systems outperform XEON'S, it is about the stability and management.

Imagine you have 50+ servers in 3 datacenters, PC's do not have out of band management, ( BMC, ILO, IDRAC ). so if one hangs how do you gain access to it? open a ticket in the DC to hook up a KVM for remote management? that will be time-consuming and expensive. Option B is to use Rasberry like Hardware and put a KVM on int, that is not viable, it will create you more problems than the one it solves.

PC's are not built to run 24/7, the can, but not designed nor engineered that way.

Servers are engineered to run 24/7. ECC memory is vital for system stability, SAS interface opposed to SATA has many benefits.

You can use Server Grade Hardware in desktops, with adding cards like RAID cards, HBA, etc... some AMD cpu's even support ECC Memory, they will never be servers, and at this point, adding server grade hardware to it, will actually cost you as much as a SH server.

But the short answer is TCO, as desktops, will cost you much less than a 2-4 year old Server.

Second fact is trend + cost, as these systems are cheep, have very good computing performance, you can sell cheaper and give good performance, so clients jumped on the train, and it became a trend.

On the long run, if you do not own your colocation space/Data-center, it will cost you much more than Brand-Servers.

@host_c said:
I played around for years, offering servers for Consumers from PC grade systems, and in my opinion, in the short term it works perfectly, until shit goes sideways.

I promised myself, that i will not use a consumer grade system for the rest of my natural born life.

It is not linked to performance, many Ryzen systems outperform XEON'S, it is about the stability and management.

Imagine you have 50+ servers in 3 datacenters, PC's do not have out of band management, ( BMC, ILO, IDRAC ). so if one hangs how do you gain access to it? open a ticket in the DC to hook up a KVM for remote management? that will be time-consuming and expensive. Option B is to use Rasberry like Hardware and put a KVM on int, that is not viable, it will create you more problems than the one it solves.

PC's are not built to run 24/7, the can, but not designed nor engineered that way.

Servers are engineered to run 24/7. ECC memory is vital for system stability, SAS interface opposed to SATA has many benefits.

You can use Server Grade Hardware in desktops, with adding cards like RAID cards, HBA, etc... some AMD cpu's even support ECC Memory, they will never be servers, and at this point, adding server grade hardware to it, will actually cost you as much as a SH server.

But the short answer is TCO, as desktops, will cost you much less than a 2-4 year old Server.

Second fact is trend + cost, as these systems are cheep, have very good computing performance, you can sell cheaper and give good performance, so clients jumped on the train, and it became a trend.

On the long run, if you do not own your colocation space/Data-center, it will cost you much more than Brand-Servers.

I know two Romanians who had the courage to set up a datacenter at home that would satisfy the minimum conditions for LowEnd services, but I don't remember their exact names here on the forum so I can ask them why they don't have PCs instead of servers. Is it because the servers are cheaper and more sustainable? May be.

darkimmortal

Server grade or at least 24/7-rated workstation motherboard is critical. The key difference is all the minor components are rated for 24/7 at high temperatures, and this is an actual physical difference, not just a longer warranty

Also ECC if you want to keep your data intact when a ram stick fails

host_c

@host_c said: AMT

@FlorinMarian said: I know two Romanians who had the courage to set up a datacenter at home that would satisfy the minimum conditions for LowEnd services, but I don't remember their exact names here on the forum so I can ask them why they don't have PCs instead of servers. Is it because the servers are cheaper and more sustainable? May be.

Lol, If we known each-other 8 years back, there would have been 3 Romanians. But I moved on. I wait for the other 2 to do the same, at some point.

FlorinMarian

@host_c said:

@host_c said: AMT

@FlorinMarian said: I know two Romanians who had the courage to set up a datacenter at home that would satisfy the minimum conditions for LowEnd services, but I don't remember their exact names here on the forum so I can ask them why they don't have PCs instead of servers. Is it because the servers are cheaper and more sustainable? May be.

Lol, If we known each-other 8 years back, there would have been 3 Romanians. But I moved on. I wait for the other 2 to do the same, at some point.

If in 2 years when my contracts expire, someone will co-locate me 6 servers and a Switch (4U) that consumes 1.2MWh monthly together with 1Gbps guaranteed at < 500 EUR/month, possible.
Otherwise, I doubt it.

host_c

Either way , we are derailing from the purpose of the topic.

So again, as all said before, Servers A+

SGraf

Server-Grade is generally the way to go.

There are some interesting"grey" areas around that. Where server-grade components and consumer grade parts go hand in hand. Such as the ryzen microcloud systems. Those are you basically have a server grade mainboard/... and then have a desktop cpu in it.

hostdare

I prefer server grade hardware to prevent headache to prevent data corruption . This has worked for us till now

FlorinMarian

@hostdare said:
I prefer server grade hardware to prevent headache to prevent data corruption . This has worked for us till now

Even with "theoretically 0 chances of your data being corrupted", you will still make backups if the data is very important, right?
I think what makes us buy servers instead of gaming PCs is the simple fact that we can find refurbished servers on the street corner at a very good price, while not many companies sell refurbished gaming PCs.

nocloud

I've never been inside a datacenter but always wondered...

If the machine is faulty and you can't connect to the machine via the network, and the motherboard, cpu has no gpu available and no expansion slots or too difficult to install a GPU temporarily . how do they fix the issue?

are there any usb gpus, or adaptors with basic output?

jar

I'm not going to overly elaborate, so feel free to take my words as having whatever value you choose to assign to them.

I was consistently of the opinion that consumer grade systems are fine, and that enterprise grade hardware was a scam. I continued to believe this until my business reached a scale high enough that now I can even see the benefit of ECC vs non-ECC memory. I was wrong. If my business never grew as large as it did, I probably wouldn't have ever known I was wrong.

MrRadic

@nocloud said:
i accidentality deleted the poll when editing EEC to ECC. So if you voted already, please re-vote!

All Ryzen support ECC.

host_c

Most of the time they do not.

and for the rest, short answer is no, data-centers offer space + net + cooling. Most of them will not open up a server. Usually most that you are going to get from basic support is change a hard drive, in a server, that has a red/orange/amber led flashing or PSU, this is why using PC grade hardware is a no go. If you want, they can schedule a tehnician visit, but it will cost you a few hundred bucks / our.

Even basic support is like 35-50 usd minimum. Basic meaning changing a failed drive or PSU, a yes, if you sent them one.

This is why using good hardare and RAID for drives in a datacenter is crucial.

Also, network setup is extremely expensive. Most when opening up a colocation, will travel on-site to do the initial Hardware installation, adding devices later can be done with the datacenter personel, but you have to think of a plug-and-play setup.

There are exceptions from this, but 99.9% is the above.

A yes, everything is labeled !!!

davide

@jar said:
I'm not going to overly elaborate, so feel free to take my words as having whatever value you choose to assign to them.

I was consistently of the opinion that consumer grade systems are fine, and that enterprise grade hardware was a scam. I continued to believe this until my business reached a scale high enough that now I can even see the benefit of ECC vs non-ECC memory. I was wrong. If my business never grew as large as it did, I probably wouldn't have ever known I was wrong.

So non-ECC is fine for games, reddit and porn?

Gotta be smart about these!

What happens if a bit flips on the privates of Sasha Grey, she flips into a trans dude? I wouldn't want a full byte ahah.

But I don't see the point of ECC a posteriori of all the bugs bundled in today's crapware anyway.

nocloud

@MrRadic said:

@nocloud said:
i accidentality deleted the poll when editing EEC to ECC. So if you voted already, please re-vote!

All Ryzen support ECC.

Not according to wikichip, or chat GPT..

As of my last knowledge update in September 2021, not all AMD Ryzen CPUs officially support ECC (Error-Correcting Code) memory.

AMD has traditionally differentiated between its consumer-oriented processors and its professional/server processors when it comes to ECC support. CPUs in the Ryzen series, which are aimed at consumers, often lack official support for ECC memory. However, some of these consumer CPUs may still support ECC in practice, but it's not guaranteed or officially validated by AMD.

https://en.wikichip.org/wiki/amd/ryzen_7/5800

https://en.wikichip.org/wiki/amd/ryzen_9/5900

The X series might but the Cézanne non-pro APUs in my example don't. or the non-X cpus

Swiftnode

We do a mix, server boards with consumer chips. Most of our customers are gaming related and require high single thread performance/high clock rates.

Server/workstation boards primarily for IPMI/OOB management. At the end of the day, all of this is subject to fail. I've seen entreprise equipment die long before it's rated MTBF, and I've seen consumer boards/cpus last over a decade.

davide

@Swiftnode said:
We do a mix, server boards with consumer chips.

Such as?

I know the opposite was possible with the intel 771. Flipping two pads on the 771 turned it into a 775 compatible with junk motherboards.

Swiftnode

@davide said:

@Swiftnode said:
We do a mix, server boards with consumer chips.

Such as?

I know the opposite was possible with the intel 771. Flipping two pads on the 771 turned it into a 775 compatible with junk motherboards.

I mean, there are near endless possibilities, plenty of Supermicro "server" boards support i3/i5/i7/i9 chips. AsRockRack workstation boards have IPMI and support Ryzen and Core series chips, same with Gigabyte, etc.

Current use case for us is W680D4U or Z690D4U boards with 12900K/13900Ks. (With some X13SAE-F motherboards in the rotation too when we couldn't get ASRR boards)