This processor offers a significant performance boost compared to the i7 5820K.
This processor offers a significant performance boost compared to the i7 5820K.
There are much quicker options than i7s for many tasks. The performance difference isn't affected by whether you're using a server or desktop—just choose what suits your workload best. Used models are now affordable too. CPUs won't last very long in this setup, as they'll likely become too slow before 99% of the time. Server CPUs tend to endure better because they run at lower voltages and are usually well-cooled.
It gets really hot when you hit around 1.3x Haswell-E, so if things are already warm then overclocking probably isn't worth it. You might want to consider a Xeon that offers a better mix of clocks and cores. A swing cooler would help, paired with a 6950X for manual tuning while maintaining a solid core count (10 or 20). Those tend to cost a lot right now (my setup was near $500). These chips could last a long time if you run them at 80s-90s speeds, especially Xeons that are designed for stability. That’s usually when the used market sees a big drop in prices. Earlier LGA1366 Xeons were cheap because they lasted longer than expected and got thrown away in bulk. The 2011 and 2011v3 models should become more affordable as they phase out. I think the disposal prices aren’t tied to replacement costs—they seem to be selling for a steal, encouraging people to upgrade instead.
They're great chips with a solid performance boost over Haswell-E. My results show similar single-core scores between my 6950X and 5960X, even at lower clock speeds. The 6950X still runs well on a single core, though power consumption increases if you overclock it. I disabled hyperthreading to manage temperatures, keeping it under 60°C during games and reaching the 80s in stress tests. With just a 360mm air cooler, it stays cool even at lower speeds. At lower frequencies, it requires less voltage and generates less heat. If you have a decent 280mm AIO or similar cooler, achieving 3.8-4GHz all-core performance should be straightforward. For the stats I have, my 6950X achieved 4984 multicore and 403 single-core scores at 4.2GHz/3.5GHz, while the stock 5960X did 3671 and 307 at those speeds. All used the same 3200MHz CL16-18-18-36 RAM setup. I’m still trying to track my 5820K numbers, but I had one before switching through 5820K, 5960X, and back to 6950X.
Generally, most of those chips found on eBay originate from old data center disposals. I can confidently say many have never operated at 100% capacity and performed well. Of course there are exceptions, but they’re typical. I’ve run a few of these as render rigs for years without issues, staying near full capacity most of the time. CPUs tend to be very reliable, so you shouldn’t worry much. ECC RAM isn’t usually necessary—it’s standard for enterprise hardware that’s built to be consistently stable, even with regular memory. For most buyers using these systems, the extra RAM rarely makes a difference since their workloads aren’t demanding.
For a regular user, it doesn’t matter at all. In data centers, it’s considered essential even when not required.