Server Upgrade

[bungee91]

I get it, and I want to agree (and I do, to a point...  )

 

I have 4 DIMM's installed, and do understand the degradation the signalling can have.

However I also feel that Intel made this chipset with quad channel memory support specifically to support the use of 4 DIMM's at a time.

I have 8 DIMM slots, but don't plan to use more than 4, and 32GB for me should be good enough for quite a while.

 

I find it pretty hard to believe (but didn't research) that Intel releases and recommends a memory controller/product that has a high % of failure and doesn't disclose it.

Again this could be a .02% occurence, and that falls into "acceptable" territory, no clue.

However a lot of "gamers" use these fancy quad channel memory setups, passing gobs of data in/out of memory with hours and hours of playing games with little to no issues/stability problems.

 

I of course tested for 24+ hours with Memtest with 0 errors, now that doesn't mean they won't occur, but I feel confident enough.

JonP does primary testing on his 4790k, which is of course not ECC.

 

Again, just my perspective, having a friendly (hopefully) productive chat.

 

[garycase]

I should have noted that with unbuffered RAM you should limit your modules to one module per channel -- clearly in the case of a quad channel board that means 4 modules, not the 2 I typically suggest as max (since most folks use dual channel boards).

 

I agree the board makers shouldn't provide the additional slots they do ... but if you're using single sided modules the loading is lower per module; and even with higher density modules you can adjust the voltage and timing so they'll be pretty reliable => in fact many BIOS implementations automatically make these adjustments if the extra modules are installed.    But by far the best way to keep an unbuffered system reliable is to limit the installed modules to one per channel.    [Note that even then, the bus loading and consequent signal degradation is far worse than a fully populated buffered memory system with 8, 10, or even 12 memory modules.]

 

 

[smakovits]

I know we are talking about cost on the E5v3, but it is worth noting microcenter has

 

Xeon E5 2620V3 2.4GHz LGA 2011-V3

 

For 339 right now,  5$ coupon also available via search,  but yeah,  that's the cheapest on the net and same as the v2.

So if you have a microcenter,  that is 60 less than the 5930.

 

[garycase]

However, the 2620v3 is appreciably slower than the 5930.    I think it's worth spending the extra $$ to go with the E5-1650v3 and get both the performance of the 5930 and the benefits of the Xeon.

 

[smakovits]

but, I am assuming the 2620 is still really good and if I am not spending the extra money I should keep it as I will still be really satisfied with the upgrade from a sempron 140...

[garycase]

Absolutely -- the 2620 is a superb choice ... I was just noting that it doesn't match the performance of the i7-5930 that was discussed above.

 

However ... compared to your current Sempron 140 ==>  the 2620 has a PassMark score of 9981, compared to your Sempron's 745 !!    i.e. it's got over 13 times the performance AND that "horsepower" is spread among 6 individual cores, so individual threads will run VERY well.

 

At the price you're able to get it for, it's probably not worth nearly doubling the CPU cost (for a 1650) to bump the performance by an additional 35%  (PassMark for a 1650v3 is 13493).

 

[smakovits]

OK,  great. Gonna press on and get those boards.  Not sure if it still is,  but the other day,  the one with ipmi was cheaper... 270 vs over 300.

 

One reason I know the sempron is outmatched is that even during file copies,  the cpu will spike to 100%.  When one of my docker containers is doing something,  I run 4 beets, emby server, nzbget and sonar, the cpu sits at 100%. While I don't transcode or anything,  the system struggles just to complete an import,  etc. 

 

After the upgrades I will add 2 vms.  One will be openelec and the other will be pfsense.  So,  as noted,  cost to performance wise,  this should be a happy medium.

 

To note,  there was no chance of doubling price to the 1650, I cannot justify it.

[garycase]

Your new setup will easily run circles around the old Sempron => each of the cores on a 2620 will more than double the performance of your old Sempron ... and you'll have SIX of those cores 

 

I agree the price difference isn't really justified to bump up to the 1650 ... the 2620 is already well beyond "enough" for what you want to do, so there's no need for even more "horsepower."

 

[smakovits]

dumb question, but with hyperthreading to I have access to all 12 cores for cpu locking or just the 6 physical ones?

[garycase]

You'll have 12 logical cores you can assign.    Remember that you actually only have 6 processing units ... hyperthreading just provides additional register sets to task switching is virtually instantaneous between the two threads associated with each physical processing unit.

 

[smakovits]

well, I feel smart.  Got my motherboard, got my CPU, but, I got no heatsink...

 

Quick question.  I have a server case with the shroud over the mobo.  does this mean I just need the heatsink without the fan or is it still best to get the one with a fan?

[garycase]

With a fan.

 

[smakovits]

With a fan.

OK,  will order one up,  maybe this weekend it comes or just next week.  And here I thought I would get to play tonight...

 

 

This one right,  http://www.amazon.com/Supermicro-Heatsink-Cooling-LGA2011-SNK-P0048AP4/dp/B0090YPQJ2

[CHBMB]

With a fan.

OK,  will order one up,  maybe this weekend it comes or just next week.  And here I thought I would get to play tonight...

 

I reckon you'll be lucky to get it this weekend with Christmas and all....

 

But... if it makes you feel any better, I've made the exact same error myself before...... twice! 

[smakovits]

Yeah,  Christmas... But I need something to do at night.

[smakovits]

Does "fan speed 1x8400rpm" mean the fan speed is always maxed and the mobo cannot slow it down?

 

 

 

[garycase]

Note that the fan you selected is VERY loud => it will definitely keep the CPU cool ... but 52 dBA is LOUD.

 

Fan Speed: 8400 RPM

• Noise:52.0 dBA

 

I presume the fan is controllable via the motherboard to run slower -- but it's not a PWM unit (at least the specs say nothing about that).    Note that while this is undoubtedly a VERY effective cooler, it may indeed be louder than you like.    With the CPU you're using you could use a much lower rpm (and quieter) unit, like one of the Noctua's.  [Your CPU has an 85w TDP, which is MUCH easier to keep cool than the 140w TDP Xeons.]

 

I presume you've already ordered the unit ... so I'd just install it and see how controllable the noise level is.  You can always either (a) replace it; or (b) add a FanMate (or other controller) to mitigate the noise.

 

 

 

 

 

[smakovits]

I did order it,  but I can always get something better,  like a no fan one.

 

It is going into a supermicro case,  that thing is already super loud.  I have the power flipped to all 5 fan to make them like 5v (lower than normal) to take them from full blast to half blast.  It is still super loud,  but much better than if they had full voltage.  But this is also because my motherboard does not have fan control to slow them down.

[smakovits]

Just realized I have 3 options already,  don't plug it in,  deal with it or 3, I looked up what a fan mate is and realized I already have one from a Zalman cooler I have from some other old system.

[cpthook]

The E3 series support unbuffered ECC memory ... you have to go up to the E5 series Xeons to get registered RAM support.    The E3's are fine ... even though the memory is unbuffered, you still get ECC protection.

 

My view is simply that for a fault-tolerant server I want fault-tolerant memory ... and the motherboard/CPU/RAM is only a small % of the total cost, so I don't mind spending more for a relatively high-end system. If I was building a new system today and wanted to keep the cost modest, I'd probably use a SuperMicro X10SLL-F-O board ($168) with ECC memory and an E3 series Xeon

 

Great information! Been searching for a thread like this for recommendations on the best performing hardware for unRAID-6 and virtualization.  I went with your Xeon E3-1241 v3B & SM X10SLL-F-O board choice to migrate my existing system, detailed in my sig, from 5.0.6 to 6.1.x.  I'm running a test rig with 6.1.4 and I'm having a great time learning all the new features and have stood up one VM successfully and about 6 other apps (maraschino, HB, MakeMKV, CP, Transmission, and SickRage).

 

What I gained most from this post read was some valuable knowledge on ECC memory correction so I invested the few extra bucks on these:  Crucial 16GB Kit (8GBx2) DDR3/DDR3L-1600MT/s  I haven't seen these recommended for this board in the forums yet but vendor site reviews are reporting no issues.

 

Not quite sure if the extra investment was overkill for the things I'm doing currently with unRAID, but hopefully your recommendation(s) will future proof my new rig for a few years to come along with some added protection with the addition of ECC sticks.  I also went with this case: Thermaltake Suppressor F51, R5 Fractal Design is sooooo yesterday.  Let the debate begin

 

Thanks again garycase!!!

 

 

 

 

[smakovits]

gary, in the process of getting this mobo installed.  should I use the fan headers as they can control fan speed or no?

 

edit: nm, the fans only have the 3 pins, not 4.

 

How many fans can I connect to a fan mate?

[garycase]

You can connect 3-pin fans to the 4-pin headers.    The difference is HOW the fan speed is regulated ... a 4-pin connection is controlled by PWM (pulse width modulation) ... which is more efficient than just varying the voltage, which is how it's done with only 3 pins.

 

As long as the motherboard headers provide for speed control, it's fine to use them -- in fact that's better than using a fan mate, since the fan mate doesn't provide automatic speed control (you have to manually adjust it).

 

A Fan Mate is designed to control one fan ... although I've seen folks split it and control a pair with no problem.

 

[smakovits]

Thanks,  turns out the cpu cooler I got had the 4 pins,  so it is controlled by the mobo,  noise is fine when compared with the rest of my fans running lower voltage.

 

I must say,  things are way faster.  From moving files to extracting via nzbget,  it's pretty sweet.  Might be overkill,  but it's nice.

[garycase]

... Might be overkill,

 

  No such thing !! 

[smakovits]

... Might be overkill,

 

  No such thing !! 

+1, so true