i5-2500k vs dual 8 core Xeon E5-2687w vs 12GB EVGA Titan X - some test results
Thank you to everybody who weighed in with advice on my recent buying choice. It greatly helped my decision. In the end I went with a dual Xeon 8 core 2687W processors with 64GB of ECC DDR3 RAM. I felt that buying behind the leading edge gave me the best bang for my buck. Including new Asus motherboard, RAM, fans, Coolermaster Cosmos 2 case and 1000 watt power supply, it cost me just over £2000.
I'm VERY happy with my purchase. It's rare to be able to see performance increases in orders of magnitude nowadays, yet moving from Intel 3.4ghz I-5 2500k quad core processor to 3.1ghz Xeons of the same generation, I experienced an improvement of approximately 300%. You might expect goinmg from 4 cores to 16 to see a 400% improvement, but consider the slightly slower clock speed. Additionally some of the performance is lost in the bus traffic (hope I have that right) between the processors. Anyway, I did some comparative render timings, between the i5, the Xeons, and my Titan X 12Gb graphics card, which I thought you might find interesting. I kept the scenes relatively simple but rendered on Superfly various samples to reduce loading time as a factor given that short renders in my previous tests varied wildly due to this factor.
i5 2500 10 samples 638 seconds
Dual Xeon 2687 180 seconds
Titan X - 30 samples - 504.5 seconds
Dual Xeon - 2687 - 735.07 seconds
I also tested a 30 frame fluid simulation in Realflow creating a crown in 10 meter square of particles.
The original i5 took 17m 50s to complete the sim.
The Xeon took 8m 40s
Previously, I tried creating a 100 meter square at a resolution of 1, creating 1gb of data per frame and it totally locked my computer up. Now (and this is undoubtedly extra RAM as well as processors), I can actually run sims on datasets that large.
I also have a scene I am working on that contains 15 figures and scenery. My memory has increased from 16 to 64GB, but the scene only takes up 9GB in memory, so I have to assume that it is the processors that makes the scene so much more navigable. Previously, it was so juddery that it was painful. Now I navigate with the same speed and smoothness as a scene with a single figure.
I was also pleasantly surprised to discover that all of the games I play regularly (ESO, Starcraft 2, Fallout 4 and Doom) all use all 16 cores and 32 threads, leading to smooth movement and transitions during rapid turns and scene loads. Even the original Borderlands uses all 8 cores on the primary processor.
Thanks again for the buying advice. Hope someone finds these facts and figures useful.