Used E5-2650 on X9 or New on X10-X11 or ...

[kckopp]

I'm looking for some feedback on which direction I should head with upgrades mainly to improve Plex transcoding. My current setup can barely get away with one transcode at time, if anything else wants the proc while a transcode is happening it chokes up.

I currently am running on a consumer setup, I3-550, 16GB RAM, Fractal Designs R4, 6x WD Red 3TB raidz2, Corsair CX430, USB thumb drive boot.

I am thinking about swapping to;

• Supermicro X9SRL-F - new (haven't found used X9 for much cheaper) - ebay - $270
• Xeon E5-2650 - used - ebay - $40
• 4x 8GB 1.35v DDR-1600R - used - ebay - $80 (on Supermicro's approved list)

and reusing everything else.

I'm guessing that this setup should be able to handle 4-5 simultaneous transcodes easily, which is my goal, but want to know what y'all think about my choice in hardware. I'm trying to get this done as cheaply as possible but want to be sure that I'm not gonna pay more than the difference in energy cost or headaches.

Since most of the time the server is idle, I'm guessing that the energy usage would be similar to what I am dealing with now. At idle (non-transcoding workloads) I am drawing ~75 watts, loaded (transcoding) ~110 watts. Is my assumption that the setup above would be similar on base? Am I better off coughing up some more money upfront for something new and more efficient?

I'm hoping to be able to run the above hardware for ~5 years. Is that expectation reasonable given that the proc and memory are server pulls that I am guessing would have already seen a few years service?

I've read that USB boot can be flaky and am thinking I may have a drive, SSD or spinning, laying around that I could use and if not I may look at either used SuperDOM or cheap new small SSD whichever I can find cheaper. I also know that my PS is on the lower end, but it has served me well. Given that the PS has been working, do y'all think it should fly as is?

Beside the questions above, does anyone see any other pitfalls with my proposed setup or any way to save some more money?

Thanks for looking and any feedback. I appreciate it!

edit: corrected current power consumption

 

[BigDave]

I'm hoping to be able to run the above hardware for ~5 years. Is that expectation reasonable given that the proc and memory are server pulls that I am guessing would have already seen a few years service?

It will run for that length of time with luck, but some things to consider. The CPU is 2.0GHz which will mean poor performance, it's 32nm technology so will mean increased power consumption and higher levels of heat you must get rid of.
Over the five year period, a less power hungry/faster core speed CPU would have you come out ahead at the end of 5yrs.

My system (shown in my signature below, idles @ 30 watts)

 

[kckopp]

The CPU is 2.0GHz which will mean poor performance

I chose the 2.0GHZ 8 core 16 thread part, thinking that it may be better than a higher core speed new 4 core 8 thread part, given my (very possibly poor) assumption that with transcoding I'd be better off with more slower threads than fewer faster threads.

it's 32nm technology so will mean increased power consumption and higher levels of heat you must get rid of.
Over the five year period, a less power hungry/faster core speed CPU would have you come out ahead at the end of 5yrs.

This. Exactly my thought and fear, but I can't/haven't figured the math to see about just how much difference there'd be. I think I need to take the time to wrap my head around energy cost over time vs cost of new(er) hardware. If anyone has a thread or other write up they could point me to to help me figure out the math that'd be great. Whenever I think about looking into it my head begins to hurt. I think that figuring the raw difference of watts used at the system would be easy enough but then I start to think about my AC bill and I give up on trying.

My system (shown in my signature below, idles @ 30 watts)

With my curr rig at idling ~90 watts, I was guessing that it was mostly because I do have other processes that are keeping the disks spinning while not really eating too much from the proc. Is it more likely that my old I3 32nm part is just that much more inefficient?

 

[kckopp]

Also, FWIW, my numbers above for current power usage are likely a little inflated in that I am measuring from a smart outlet on the mains side of my UPS, and have a few other devices on the same UPS as my FreeNAS rig (the UPS itself, four port gigabit switch, daughters PC powered off but drawing something, strand of LED x-mas lights attached to another smart outlet powered off but drawing something). So maybe take another 5 watts off, ~85 watts idle(ish) and ~120 watts under load...

 

[BigDave]

I chose the 2.0GHZ 8 core 16 thread part, thinking that it may be better than a higher core speed new 4 core 8 thread part, given my (very possibly poor) assumption that with transcoding I'd be better off with more slower threads than fewer faster threads.

The faster core gets done with the task in less time, thus returning to idle sooner= power saved

This. Exactly my thought and fear, but I can't/haven't figured the math to see about just how much difference there'd be. I think I need to take the time to wrap my head around energy cost over time vs cost of new(er) hardware. If anyone has a thread or other write up they could point me to to help me figure out the math that'd be great. Whenever I think about looking into it my head begins to hurt. I think that figuring the raw difference of watts used at the system would be easy enough but then I start to think about my AC bill and I give up on trying.

In order to figure exact savings, you would have to come up with a simulated usage pattern and the inflated cost of energy over five years. In short, over five years if you figure a break even on power vs. cost, your still saving wear and tear on the equipment involved. You're right, it's a headache to figure and because it must be based on conjecture, the value is reduced. It's a rock and a hard place for sure;)

With my curr rig at idling ~90 watts, I was guessing that it was mostly because I do have other processes that are keeping the disks spinning while not really eating too much from the proc. Is it more likely that my old I3 32nm part is just that much more inefficient?

My rig's disks spin 24/7 so being that our pools are made up the same, the difference has to be CPU/RAM and maybe PSU efficiency.

 

[BigDave]

Also, FWIW, my numbers above for current power usage are likely a little inflated in that I am measuring from a smart outlet on the mains side of my UPS, and have a few other devices on the same UPS as my FreeNAS rig (the UPS itself, four port gigabit switch, daughters PC powered off but drawing something, strand of LED x-mas lights attached to another smart outlet powered off but drawing something). So maybe take another 5 watts off, ~85 watts idle(ish) and ~120 watts under load...

My number is taken from the UPS itself, with just the Server being the only load plugged into it. Your 5 watt estimate may be a bit low...

 

[kckopp]

My number is taken from the UPS itself, with just the Server being the only load plugged into it. Your 5 watt estimate may be a bit low...

Yeah, I think I should either unplug the other stuff and look at what my UPS is reporting (would have to read up on NUT again) or take my server offline for a sec while I move my smart outlet/power meter between the UPS and server.

edit: good call! decided to take the server down for two reasons, first, I don't think my UPS reports through NUT actual usage only ups.realpower.nominal which was reading 865 (maybe 86.5 watts idk), second, I can't take the switch off of the UPS because I have no where else to plug it in at. With the server off the smart outlet was reading ~15 watts for the UPS and other junk plugged into it. Server connected to smart outlet directly at idle now reads ~73 watts and with load ~106.

 

[Stux]

For transcoding a good estimate of total CPU power available in the same generation is literally cores x gigahertz, x 1.6 if hyperthreaded.

Newer CPUs tend to be 5-10% faster per clock than the previous generation, although this is reducing.

Newer CPUs are far more efficient at idle than the older generations.

 

[kckopp]

In order to figure exact savings, you would have to come up with a simulated usage pattern and the inflated cost of energy over five years. In short, over five years if you figure a break even on power vs. cost, your still saving wear and tear on the equipment involved. You're right, it's a headache to figure and because it must be based on conjecture, the value is reduced. It's a rock and a hard place for sure;)

Here's a stab at the math, given a few assumptions;

• Over the last 12 months I have paid $0.077 per kWh
• Cost of energy will rise by 5% annually
• The amount of electricity it takes to cool the equipment is equal to the amount of needed to power it
• I'm only worried about running my AC six months of the year
• Proposed rig will use about the same amount of energy as my current << this is where I am especially hoping for some feedback, because I really don't know
• Newer 14nm hardware would run at idle at 30 watts
• I will be pushing 147% (based on current rig) of idle watts for transcoding an estimated three hours a day << I have no idea if this scales across generations

As best I've been able to figure to purchase newer hardware at near passmark scores I'd be out $700-$800 upfront (vs the $390 for older in OP) if purchased new and not that much less if used as the newer hardware appears to hold value on the used market.

5 year energy cost of current/proposed rig $466, more efficient newer hardware $219, difference $247. In this case I would think older would work out better.

If I change the assumption about the E5-2650 system from 75 to 100 watts idle and from 110 to 147 at load, over five years energy would be $728, or a potential difference of $509 from a newer setup. Here it seems to make more sense to go with newer hardware.

Of course this all goes out the window if the older stuff takes a dive before the five years are up. Also, I'm sure there are variables I'm missing here and as you stated a lot of it is based on conjecture.

If I had a better idea of whether or not the 32nm E5-2650 system would be similar, again given that mostly idle, in energy consumption to my current 32nm i3-550 system this might be a pretty easy choice for me. Though, if similar, the cost savings over time to go old is maybe a couple bucks a month.

For transcoding a good estimate of total CPU power available in the same generation is literally cores x gigahertz, x 1.6 if hyperthreaded.

Newer CPUs tend to be 5-10% faster per clock than the previous generation, although this is reducing.

Newer CPUs are far more efficient at idle than the older generations.

So something like;

• E5-2650 v1 32nm - passmark 10,420 - 8 cores * 2.0 GHz * 1.6 = 25.6
• E3-1241 v3 22nm - passmark 10,003 -4 cores * 3.5 GHz * 1.6 * Ivy 1.075 * Haswell 1.075 = 25.9
  
• E3-1230 v5 14nm - passmark 9,603 - 4 cores * 3.4 GHz * 1.6 * Ivy 1.075 * Haswell 1.075 * Broadwell 1.075 * Skylake 1.075 = 29.1

I had been running with the rough rule of thumb that is in Plex's faq, that you want ~2000 passmark per 1080p transcode and it looks like the product of cores, frequency, hyperthreading and gains seems to line up pretty well with passmark scores.

Is there anything to what I had read elsewhere about greater threads at lower clock speed being better than greater clock speed fewer threads for transcoding? What BigDave had said about faster cores getting the job done faster makes sense especially in single threaded jobs but with as multi-threaded as transcoding is and that simultaneous transcoding is my goal it also seems to make sense that more cores may win in this scenario. If it doesn't really impact much one way or the other I am thinking now that going the newer more efficient route may be best especially given that if the E5-2650 is a bit more power hungry than what I've got currently, over five years the older hardware would be a wash if not a loss.

 

[pschatz100]

Back to the original post...

If you are buying used, I would not suggest a processor prior to IVY Bridge for what you have in mind. You can find Supermicro X9-SCM and X9-SCL, plus many X10 boards for considerably less than $270. The biggest issue with the older boards is likely to be the maximum memory they support. Many older boards will be limited to 32 Gb. For the amount of disk you mentioned, 16Gb would probably be enough memory. You can start with two 8Gb sticks then add two more if you need it.

Regarding your Plex usage, are you really transcoding 4-5 streams at the same time? That is a lot. If so, you might want to review your movie storage strategy and save movies in formats that are friendlier with your playback devices.

 

[kckopp]

Regarding your Plex usage, are you really transcoding 4-5 streams at the same time? That is a lot. If so, you might want to review your movie storage strategy and save movies in formats that are friendlier with your playback devices.

Currently, no, I'm not transcoding more than 1 1080p stream at once since the server can't. In my house all playback devices are direct play, PMP on a NUC, Nexus Player, Chromecast, FireTV. I have shared Plex with a few family/friends and they have to transcode due to bandwidth. They're receiving the transcode error "the server is not powerful enough to convert this video for smooth playback." In testing on my side transcoding one stream from 1080p pegs the processor but works fine until another stream begins transcoding or other processes on the server want the proc. You're right that I could store things in friendlier formats, in fact the past when I stored most everything in 720p I'd find two people streaming at once and no complaints. The idea that I'd have 4-5 transcodes running at once though not necessarily likely is possible and allows for some future proofing.

If you are buying used, I would not suggest a processor prior to IVY Bridge for what you have in mind. You can find Supermicro X9-SCM and X9-SCL, plus many X10 boards for considerably less than $270.

Thanks I will take a look at what I could put together used Ivy Bridge forward. The only reason i considered the X9SRL-F @ $270 was that with the E5-2650 and 32GB memory for only another $120, $390 as a package seemed a good deal.

Seems the consensus is 22nm or less, anything older will eat too much electricity.

 

[kckopp]

How about something like this;

• Supermicro X9SCL-F ~$85 - ebay
• Xeon E3-1270 v2 ~$170 - ebay
• RAM 16GB 2x8GB ~$125 - Crucial

Should it be a lot better about power consumption than the setup in OP? There are two concerns about this setup I have, 6x SATA means no DOM or SSD for boot without adding an HBA (not too worried, I am usb boot now, thinking of adding a usb boot mirror) and though I'm sure 16GB RAM is good now will RAM recommendations go up with the upcoming move to docker containers vs. jails?