Update: I have added comparisons to the Xeon L5630 which, at 40 Watts TDP, is an even more power efficient option.
After having somewhat recently acquired two servers, I have been spending some time both setting up a useable home network setup. Part of this was to also take a look at power consumption, as rack servers aren't generally known to be very power efficient, which can be a bit of an issue for your power bill if you're doing some kind of home lab type thing.
I have two near-identical Intel branded servers, with some minor variations between. One has a pair of Intel Xeon X5570 quad core CPUs, clocked at 2.93Ghz and with 8Mb cache. The other have the slightly lower specced Intel Xeon X5550 quad core CPUs, clocked at 2.66Ghz. Both are rated for 95 watts TDP, meaning they're quite the power hungry buggers.
Not too long ago I stumbled upon a data sheet listing out CPUs of this era, along with their power consumption. I noticed the L5640 a they seem to be the right mixture of providing enough power whilst at the same time consumping less, having a TDP listed at 60 watts. With two dual-CPU servers this could bring quite a noticeable difference in energy consumption, so I was curious to give it a try. I found a local seller who had 25 of these at a decent price, so ordered 4.
Swapping the CPUs
Replacing the CPUs was, as you might expect, quite straight-forward. I'm pretty sure the most challenging part was to pull the server out from my "stack," for lack of a better word, as I have yet to stumble upon a small yet affordable rack.
After opening up the case, I loosened all four screws of the first heat sink, after which it simply lifts off. I immediately cleaned off the old thermal paste as that stuff really does get everywhere if you don't. After this it's as simple as plopping down the new CPU, applying some new thermal paste (don't go overboard; a little bit goes a long way). I used some Arctic Blue as I still had it from a previous project. After this you can screw the heat sink on again, and you're all set. Be sure to put it on the right way, follow the path the air takes through your server.
At first boot, your server might complain and throw some beeps. This is to be expected, as it's simply noticing stuff has changed. It'll be good to double check everything looks alright in the BIOS, but that's about it. Unless you're running a particularly picky operating system (ie. Windows), everything else should just work fine.
So, how do they perform?
While the L5640 is a generation newer, it is a lower powered version, so differences should exist. I was curious to see how they compare, so ran a Geekbench test both before and after the CPU swap. The results were both interesting and not all that surprising. Across the board they're basically very comparable, if not almost identical. This is good, as it means you're getting about the same while at the same time they consum less power.
First, here are some of the key specifications of these CPUs:
|# of Cores
|# of Threads
Geekbench' single core results are, as can be expected, in favor of the more powerful X5570 with a score of 2575. The L5640 came in second with 2007 points. The L5630's single core results are the lowest, coming in at 1436, or about 55% of the X5570, and 71% compared to the L5640.
Multi-core scores are more favorable to the L5640, giving them a total score of 16,509, compared to the X5570's 14,737. The L5630 comes in third with 10,382.
For a point of comparison, I also ran the same tests on several other machines. While the specific use-cases of each machine is, of course, different, I thought it might help put things in perspective.
|Debian 6.3 (Proxmox 5.3)
|12" MacBook (2015)
|macOS 10.14.x Mojave
|12" MacBook (2016)
|macOS 10.14.x Mojave
|macOS 10.14.x Mojave
If you are interested in a more detailed comparison between the CPUs, I recommend you check out these Geekbench comparison results:
- X5570 vs. L5640: You'll notice that with the exception of the AES test, both CPUs perform very similarly, which is fantastic as the L5640 comes with a 35 Watt TDP drop per CPU, so a considerable power savings.
- L5640 vs. L5630: As can be expected, the L5640 outperforms the L5630 pretty much everywhere, but if power consumption is key the additional combined 40 Watts TDP drop is great, especially if you're not usually running CPU intensive or otherwise time sensitive tasks.
What started all of this was, of course, power consumption. With each server (usually) sporting two CPUs, an almost 60% drop from 190 Watts TDP (X5570 x2) down to 80 Watts TDP (L5630 x2) is certainly not unwelcome.
I have a basic, decidedly consumer grade wattage meter, so while the results are by no means accurate, I did —obviously— notice an immediate drop in power consumption. Whereas Dengdengi would normally idle at around 170-180 Watts, it now sits comfortably around 130 Watts or so. Kokkiri now runs on average around 120 Watts, but note that this is including 14 spinning hard drives, and with both PSUs plugged in.