User Tools

Site Tools


blog:upgrade_time

Upgrade Time...

Back in about 2015 or so, I bought two of these:

It's an HP Proliant Microserver, Generation 8. From memory, they cost me about AU$360 a piece, which seemed pretty cheap at the time for a “proper” server, capable of using up to 32GB of ECC RAM and running a 4-disk multi-Terabyte storage array, constantly and reliably. (For the record, you can now get Generation 10 equivalents, which I don't own personally; but having installed one for the local church Parish Office, I can attest that they are excellent, too).

One of the reasons I got the servers so cheap was that I bought the base model -they each shipped with a 2-core Celeron G1610T processor, for example. Since the servers just sat there running ZFS and providing bulk storage to the other PCs around the house, that level of CPU was more than sufficient.

But never one to let things run unchanged for long, I have recently been thinking that it would be quite nice to upgrade both servers with rather more capable Xeon CPUs. You can buy suitable models on Ebay for quite reasonable sums it turns out, and I therefore rapidly acquired a Xeon E3-1260L (for around £45) and a slightly more powerful Xeon E3-1230 V2 (for £60). Both are socket LGA 1155 processors, so slot in perfectly to where the Celerons used to sit!

Both new CPUs are 4-core with hyperthreading, so 8-thread parts (as compared to the Celeron's 2-cores, 2-threads). Moreover, the Celerons only ran at 2.3GHz with no turbo; the E3-1230 runs at 3.3GHz with a 3.7GHz turbo, and the E3-1260L runs at 2.40GHz with a 3.3GHz turbo. So, four times the threads and with the possibility in turbo mode of an approximately 33% boost in processor speed: what's not to like?!

Well: there is one problem. The original Celerons were 35W parts, and the passive heat sink the HP Microservers ship with are rated for 35W, so are a nice match. But the E3-1260L is a 45W part and the E3-1230 runs at a whopping 65W. More watts equals more heat, especially with a heatsink rated way below what each of the new processors can output. Space is incredibly tight within the server chassis, though, so there is no way to buy a thunking great cooling fan and sticking it onto the original heatsink: there just isn't enough clearance, as this photo shows:

It doesn't help, either, that the original heatsink is attached by a rather unique rectangular arrangement of screws; your standard, square cooling fans and alternative passive heatsinks are not going to fit.

As it turns out, the 45W CPU is almost 35W, enough so anyway that when I checked the CPU temperatures under load, the temperatures were not outrageous: I forgot to take a detailed note, but from memory, I was pegging around 58°C after 30 minutes of 100% CPU utilisation for all 8 threads, which compared relatively favourably to the original Celeron's approximately 46°C. Clearly the extra wattage of the new CPU results in a hefty temperature rise of around 12°C, but the result is not anywhere near toasty temperature levels that would cause real concern. So the 45W CPU can run as it is, with no further concerns or consideration.

But the E3-1230 V2 was another matter. After its 30 minutes of 100% utilisation, that CPU was clocked at a rather scary 86°C, which is a degree or so above the point where it's automatically throttled back to lower, slower speeds (ambient temperature in my study, where I tested these things, was always around 22°C, by the way). It is true that I don't ever expect these servers to be continually stressed like that in day-to-day operation; nevertheless such a high operating temperature under load seemed to me a problem I'd like to avoid if at all possible.

So I visited Amazon and bought two of these things:

It's a tiny -and I mean that's it's only about as big as the first joint of your thumb, so really very petite- 12 volt fan. It is fairly quiet in operation and pushes a reasonable amount of air around. I glued both of them to the passive heatsink, like so:

…and wired them into the existing tangle of molex cables, making sure that they blew air towards the back of the case, where the giant built-in fan is waiting to expell it all into the great outdoors! After another 30 minutes at 100% utilisation, I was recording temperatures of around 80°C, which wasn't a spectacular improvement, but 6°C means the CPU wasn't being throttled any more and was a temperature I could (just about) live with.

But whilst the HP BIOS on this server is configured to run the main system fan in 'optimal mode' by default (which means it just purrs quietly in the background!), there is an option to bump that up to 'Increased' or 'Maximum Cooling'. I therefore decided to go all-in and switched on 'Maximum Cooling' mode and re-ran my temperature tests. The gentle purr of optimal cooling gave way to a server-room-like roar in maximum mode, but I ended up with temperatures of around 71°C, which is an excellent result. Presumably, without the little fans mounted on the CPU, I might have been able to achieve 77°C or so anyway, just by bumping up the built-in fan capabilities, but the extra 6°C from the purchased micro-fans is appreciated anyway.

Happily, as this screenshot shows, with 100% CPU utilisation on all four cores (and all 8 threads), the CPU temperatures not only topped off at around the 70°C mark, but the CPU frequency easily maintained an un-throttled 3.5GHz:

Incidentially, the new roar of the built-in fan won't be a problem for me once these servers are returned to their normal hiding place: in the loft, where ambient temperatures are usually significantly cooler than those found in my study, which can only help matters further!

My two old servers are now both more than capable of running as my ZFS storage servers and as virtual machine hosts, increasing their utility to me significantly. For a relatively small outlay on new CPUs and an enjoyable day or two of tinkering with thermal paste application and micro-fan gluing, I've extended the useful life of these workhorses with minimal effort. Recommended!

blog/upgrade_time.txt · Last modified: 2019/05/18 15:08 by dizwell