CPU for a medium server?

[cen]

Hello everyone

I've done a lot of "google research" but on the end of the day I cannot decide whether to go with Intel or AMD. I found very few benchmarks or suggestions from community when it comes to servers. I will be using it for web, storage, ftp, mysql, game servers and game hostbots (minecraft server, cod server, battle.net hosts etc) things like that.


Should I go with:

Intel i5 2400
-I chose SB over IB because of the price and they still have similar performance
-no need for K since I'm not overclocking. 2500 has slightly higher frequency but it make slittle difference

or

AMD FX-8120
-with latest price cuts even cheaper while it has double the cores
-I guess I could go with 4 or 6 core model which are similarly or less priced but # cores is really the only advantage over i5


Any suggestion is welcome

 

[LeftSide]

AMD can use ECC ram. If you want to protect important data, or 100% up-time is critical, then you should go with the AMD system. Intel dominates all the benchmarks and performance/watt.

I went with a C204 motherboard and an intel G620 (the low end i3's and XEONs can use ECC ram). Got the processor from microcenter for $50, the mother board was $150, and ram was $90. It runs OpenIndiana 24/7 as a ZFS file server and has 2 ubuntu VM's running teamspeak. Looks like you need quite a bit more power than me.

 

[paperwastage]

minecraft - java based, severely single-threaded last I heard... so faster single-thread performance the better

other stuff, depends on how much $ you want to spend

 

[Dravic]

if virtualization is important you'll get better hardware support with AMD in the mid to low end. Intel doesnt support vt-d in all its processors. if its single threaded performance means more, Intel is superior.

 

[T_Yamamoto]

I would personally go with amd for this. Cheaper but more cores.

 

[cen]

AMD can use ECC ram. If you want to protect important data, or 100% up-time is critical, then you should go with the AMD system. Intel dominates all the benchmarks and performance/watt.

I went with a C204 motherboard and an intel G620 (the low end i3's and XEONs can use ECC ram). Got the processor from microcenter for $50, the mother board was $150, and ram was $90. It runs OpenIndiana 24/7 as a ZFS file server and has 2 ubuntu VM's running teamspeak. Looks like you need quite a bit more power than me.

I didn't know that FX supports ECC memory so that's another + for me. The only thing I have a hard time with is finding Unbuffered ECC RAM. I'd prefer a 8GB stick so I can later expand up to 32GB if ever needed but it seems the market is very small at the moment.

I can always throw in normal non-ECC tho..

if virtualization is important you'll get better hardware support with AMD in the mid to low end. Intel doesnt support vt-d in all its processors. if its single threaded performance means more, Intel is superior.

I will be running FreeBSD so if I ever need any virtualization I'll be using jails.

 

[Cerb]

I didn't know that FX supports ECC memory so that's another + for me.

Just make sure the mobo also supports it, as many do not.

The only thing I have a hard time with is finding Unbuffered ECC RAM. I'd prefer a 8GB stick so I can later expand up to 32GB if ever needed but it seems the market is very small at the moment.

Just get 4GB sticks for now, replace when/if you upgrade. 8GB sticks aren't common enough to be cheap/available, yet, but even if prices rise a bit, RAM will still be cheap in the scheme of things.

 

[Ferzerp]

As someone who does this for a living on a much larger scale, I have to facepalm at "moar cores! must be better for a server!"

http://www.newegg.com/Product/Produc...82E16819115083

would be a decent choice for you at the price range you're talking if you absolutely must have ECC, but it's nearly useless for home tinkering.

 

[cen]

As someone who does this for a living on a much larger scale, I have to facepalm at "moar cores! must be better for a server!"

http://www.newegg.com/Product/Produc...82E16819115083

would be a decent choice for you at the price range you're talking if you absolutely must have ECC, but it's nearly useless for home tinkering.

So from your experience, how important is ECC? What's an example of a bad scenario I could face with a non-ECC memory?

 

[Ferzerp]

You might experience an unexplained crash once a year or maybe slightly more (or very likely nothing at all).

For the stuff I do, it is necessary as 96GB of ram is the minimum I order on things, and the larger the memory volume, the more likely issues would occur and uptime is key for me. For a home user though, it is of marginal benefit.

What you're about to get is a pile on of people who think that ECC memory suddenly means flawless memory. These people will think that every computer ill of the world can be solved with ECC memory in home systems. They don't understand that ECC only addresses the random bit flip that happens from time to time, and does not answer major hardware failure one bit. Nor do they understand that the problem they're trying to address is actually poor quality memory. If you buy bottom of the barrel components, you will have issues. If I buy a $20 PSU, I will have issues. The answer to that isn't to assume I need fully redundant hot swappable $300+ PSUs for my home system. The solution is to stop buying shoddy components.

 

[Vic Vega]

...

 

[cen]

Allright, thanks everyone for your opinions.

I think I'll go with ECC and an AMD. Although Xeon E3 with hyperthreading seems to have better performance and I wouldn't mind having it, it also costs like 80€ more in my local stores and I can invest this difference elsewere.

 

[Ferzerp]

I'm a SAN/NAS engineer for a large storage OEM. From my experience ECC is very important. You would be amazed at the number of CORRECTABLE memory errors production systems really have. It's more than you might think, which is why for production systems ECC is required. Most people never know the real number because their systems aren't sophisticated enough to keep track and report on it, and since they are correctable the system keeps chugging along.

UNCORRECTABLE memory errors are less common, but understand that every correctable error you get with ECC is an uncorrectable with non-ECC - and on a production system which runs 24/7 this can be quite a few. Things to ponder.

And fully redundant power will shield you against power outages, PDU and PSU failures, but they aren't appropriate for most home users.

Why not go whole hog with mirrored memory, fully redundant power (utility, UPS, genset, PDU, and PSU), etc?

He did say server, but it seems like it's just a hobby box.

RAS features are great, but they come at a price, and the price rarely justifies their use in home systems. The only sort of RAS feature implemented in my home is that my backups are parity protected.

 

[Vic Vega]

...

 

[Ferzerp]

It's about twice the dimm cost on the lower end of things.

 

[Vic Vega]

...

 

[Ferzerp]

Around $100 or so, but he's already balked at an additional $100 on the proc.

It's spending $100 for a "maybe I'll save myself 1 crash in a year" insurance. The value just isn't there in my view. Is a single crash (not likely to happen at all in all honesty) once per year a $100 inconvenience?

 

[Cerb]

Around $100 or so, but he's already balked at an additional $100 on the proc.

It's spending $100 for a "maybe I'll save myself 1 crash in a year" insurance. The value just isn't there in my view. Is a single crash (not likely to happen at all in all honesty) once per year a $100 inconvenience?

Crashing? What? Where? A crash would be the most ideal failure mode. The scary failure mode is when you don't even know anything changed on the way to or from RAM, and the computer happily keeps on running.

What you're about to get is a pile on of people who think that ECC memory suddenly means flawless memory.

No, you just like to ignore when anyone explains that being able to know if an error occurs is preferable to assuming it hasn't, which ends up going round and round for a few pages, because...

They don't understand that ECC only addresses the random bit flip that happens from time to time

...is entirely false. ECC has no knowledge or care about the nature of a bit flip, nor whether it occurred inside the DIMM, on the ay from it, or on the way to it. It only handles the case that it happened or did not. That most errors are single-bit and easily correctable, assuming no major faulty HW, is a nice bonus (a PSU or mobo VRM problem under heavy load, FI, might be an intermittent cause of an error, even if none are in the midst of dying).

Why not go whole hog with mirrored memory, fully redundant power (utility, UPS, genset, PDU, and PSU), etc?

Because that would be wasteful, relative to the system's needs. ECC is not a form of a redundancy. It is to ensure that what was intended to be written to RAM is, and if it isn't, that it can be more easily diagnosed. In a desktop computer, your RAM is the only place you can transfer data in and out of that has no means to detect or correct for hard errors.

 

[Rifter]

depends on whether the programs you are running are multi threaded or not. If they are get the BD chip, if not go SB or IB.

 

[MisterMac]

As op CLEARLY states, this is not a production system.

This is not mission critical.



Since you'll be using it for various stuff, thats NOT VERY cpu intensive, get the AMD.
If you wanna splice up better IPC - go for a E3 with HT or i7 with HT.


But all in all, with your workload in mind - I severely doubt ANY of your modern processor choices will be in ANYWAY a bottleneck.

Focus on getting good DISK's... I\O here is much more important for your mishmash of workloads.

ECC is not important, forget it.
Your workload doesn't demand it - it demands fasts disks and interconnecting.


Coming from me, i've run various game servers, wrote emulators for years for various MMO's.

Disk I\O is the major king when we're talking low mission critical stuff like this.

 

[Ferzerp]

Crashing? What? Where? A crash would be the most ideal failure mode. The scary failure mode is when you don't even know anything changed on the way to or from RAM, and the computer happily keeps on running.

No, you just like to ignore when anyone explains that being able to know if an error occurs is preferable to assuming it hasn't, which ends up going round and round for a few pages, because......is entirely false. ECC has no knowledge or care about the nature of a bit flip, nor whether it occurred inside the DIMM, on the ay from it, or on the way to it. It only handles the case that it happened or did not. That most errors are single-bit and easily correctable, assuming no major faulty HW, is a nice bonus (a PSU or mobo VRM problem under heavy load, FI, might be an intermittent cause of an error, even if none are in the midst of dying).

Because that would be wasteful, relative to the system's needs. ECC is not a form of a redundancy. It is to ensure that what was intended to be written to RAM is, and if it isn't, that it can be more easily diagnosed. In a desktop computer, your RAM is the only place you can transfer data in and out of that has no means to detect or correct for hard errors.

Cerb, it's a home hobby system. It just isn't that important.

No one claimed it's a form of redundancy. In your desire to correct me, I think you've inserted words in to my post. *However*, with that said, ECC *is* redundancy within a dimm in the same way that RAID 4 is redundancy within a storage system treated as a unit.

No one also claims it is of no benefit, so you can back that straw man train up.

I say that the extra cost is not worth it for someone trying to pinch pennies on a hobby box. I also say that the extra latency is not worth it for someone building a screaming gaming PC.


It just isn't important for a home system. And no, for home stuff, a crash is typically the worst thing. The way most people use their PCs a bit flip is probably going to happen in a currently unused section of memory anyway.

There really isn't any point in continuing to discuss it though, as, you have decided that it is, and I have decided that it isn't.

 

[blckgrffn]

I'd likely spend the $100 on a mirror drive first, OP. Nothing like losing it all.

This isn't to say you should back up things, you should. But having a drive crap itself and being down is a freaking PITA and likely to happen - and data is (almost) always lost. It also tends to be that thing you were just working on. (this is a good time to come up with that backup plan...)

The next splurge I would make is a nice, quality, efficient PSU. It may even pay for itself over time.

Once you've got other bases covered, ECC is something to consider, IMHO. It's cost vs availability and data integrity and you need to decide how to weigh these things in your own particular situation.

 

[tynopik]

Cerb, it's a home hobby system. It just isn't that important.

I'm sick of this attitude that home users have no data that is important and worthy of protecting

'Unless you're a super duper enterprise customer and can afford quadruple redundant systems, all the data on your computer is worthless shit'

that's the attitude that comes across loud and clear

he said he would be using it for storage, maybe it's for something important, maybe not, he didn't say, but you went ahead and assumed that just because it was a 'home' user, by definition it was worthless.

 

[Ferzerp]

If your data is important, you back it up.

This is what I meant about people who will pile on thinking ECC is better than sliced bread and will solve all your computing woes.

 

[Subyman]

Having been through unmirrored failures, it is worth the extra dough to mirror the important drives. I wouldn't dream of running higher RAIDs with out a dedicated card though, so stick to RAID1 if you decide to go down that route with mobo support.

For a home server that will be used for gaming, I'd get the most processing power you can. Forgo ECC and jump to Sandy IMO.