I was recently exploring what my options were for upgrading the CPUs of my main server. Currently it’s running dual E5-2670 CPUs, each housing 8 cores (16 threads), 115W max TDP, on an ASRock EP2C602 motherboard. This motherboard will support up to v2 of the LGA2011 cpu socket. The v3 and v4 versions need a new mobo, so my choices are limited to v1 or v2 CPUs.
I’ve been using the data from cpubenchmark.net to compare CPU performances. Below are a list of CPUs and their rated performance. To equalize things, I’ve created another column to compare per core performance, which is what games typically care more about.
|CPU Name||Benchmark score||No Cores||Per Core||ebay pricing||Comment|
|E5-2670||12319||8||1539||~$90||Current server (reference)|
|FX-6300||6349||6||1058||Current desktop (reference)|
|Ryzen 7 1700||13779||8||1722|
The best I can possibly do for highest performance per core, is the E5-2667 v2. Luckily it is a supported CPU by the motherboard. (The E5-2673 v2 was nowhere to be found on the mobo CPU compatibility list, so I’m not going to entertain that one.)
The performance increase over my current E5-2670 is only about 1.33 times faster (overall and on a per core basis).
For comparison, the latest generation, top of the line xeon CPUs don’t do that much better on a per core performance. Sure there’s up to 20 cores which is where the overall performance is gained, but even the 10 core, top of the line i7 and E5 CPUs put out similar per core performance to the aging E5-2667 v2 (which was released in 2011).
Yes, I realize that purely cpu benchmark numbers don’t tell the whole story, as the architecture of the newer processors will likely add speed improvements to the entire system, but in order to compare apples to apples, cpubenchmark.net is the best I can do right now. It’s very likely that the a system based around the LGA2011 v4 socket will overall be faster then just what these benchmarking numbers show, but the price difference is not justifiable for me.
Is that significant enough to warrant an upgrade? Right now, I don’t think so. Anything less then around a 2x performance boost per core is likely not going to be noticed to any great extent in day to day usage.
The E5-2670 was/is a bargain at around US$90. Since part of this machine will be used for pushing out VR games, my money is likely much better spent on GPU upgrades in the near future, which have a better return on image quality and maintaining high frame rates.
With 16 cores available (between the 2 CPUs) I can throw more cores at the games that need more CPU (obviously to a limit), so that should not be a problem. I had noticed that VR has a much higher demand on CPU then regular gaming, and more cores should help.