I’m after some help please. I’m speccing up a new dual-xeon workstation and have come to a point where I need to decide between these two configs for VRay rendering:
Now, the 12-core runs at a slower per-core speed, so will be worse for unthreaded/poorly-threaded applications, but has 2 more cores (4 in total across the machine), so 8 extra threads in total. It’s also almost £1000 cheaper than the 10-core.
According to many benchmarks they appear to be very close in terms of speed performance, but the 12-core is almost £1000 cheaper in this system, and perhaps might be faster and more efficient overall? It seems to be the best value-for-money option here.
Which would you go for?
Also, I am aware of Autodesk’s NUMA issue with >64-core systems so these ones should be fine, right? It would either be 40-core or 48-core in total.
If you simply multiply the frequencies by the number of cores the E5-2687 V3 delivers a bit more : 31 GHz vs 27.6 GHz
Now how much difference does that make in real life, I have no idea. I guess the <4 GHz difference might make a great difference for a machine rendering 24/7
This is kind of what I was thinking, given that these processors are the same architecture, so it should be reasonable to compare them in terms of raw frequencies. However I’m wondering about efficiencies with Vray. Perhaps it works more efficiently and therefore more quickly with slower threads but more of them, than fewer threads that are faster. I’m wondering if Vlado or anyone from Chaos has any direct experience with this and can chime in.
What I have just noticed… according to Anandtech, the 2670 can turbo boost up to 2.6ghz across *all* 12 cores, whereas the 2687W can only boost up to 3.1 across all of its cores… so actually the 2670 could have a raw frequency of 31.ghz V the 2687W of only 31.0ghz… so if they are equal or almost equal there, but it’s £1000 cheaper…
So it’s clear from this Intel table that the 2670 12-core runs at 2.6ghz per core when all 12 cores are loaded and that the 2680W runs at 3.2ghz per core when all 10 cores are loaded. This creates a general comparison of 31.2ghz v 32ghz - so almost identical. Plus it’s likely that more cores = better when talking about highly threaded applications like vray, plus £1000 cheaper. So the 12 core is the way to go.
Apparently their multipliers are locked, so any overclocking would have to be done using BCLK and would be extremely mild, which is not ideal, rendering it more or less a “no”, or so I’m told.
a small bclk adjustment over 12 cores should make a fairly meaty difference wether you can get a dual mobo for them with bclk adjustment in the bios is another matter.
All of the builds that I’ve been quoted from various places today are using the same Asus Z10PE-D8 WS 2x S2011 motherboard, and I have no idea if it allows for overclocking the bclk at all. I’m asking one of the vendors now, and will report back here with my findings. Even a tiny boost, as you say, could be worthwhile, as long as stability is not sacrificed. Maybe 2.6ghz stock or so…
i did a quick google.. seems “modest” adjustment is about right.. from what i could see you can change the bclk by about 3mhz. even over 12 cores thats not gonna do diddly squat.
Unfortunately I don’t have any experience comparing such machines. I would generally prefer a smaller number of faster cores, but if the final performance is the same then I guess I doesn’t matter.
Performance will be very close. I would go with more cores for less $$. Keep in mind that ECC DDR4 ram is not cheap..
You won’t get any NUMA or 64 thread limit problems with 2670v3. I have 14 and 18 core xeons. 14 core work well, with 18 you need workaround because pair of those give you 72 threads. Hopefully Vlado and his team will fix it soon.
This is what I’ve been doing for years - going with the top end i7 machine and using older ones as rendernodes before they’re retired. For me I now need maximum single-PC performance, so it makes more sense to go with a dual xeon system. The overhead of running a system over DBR means you lose quite a lot of that machine’s native performance as well. It’s done me fine for many years but I’ve just pulled the trigger on a very nice workstation and should have it set up and running on Wednesday hopefully.
I think if you’re looking for inexpensive rendernodes in a smaller form factor then 5960X could be a good option. Much cheaper RAM and obviously only 1 processor. But as I said, the performance will not scale as well as a dual CPU machine.
Keep in mind that the 5960x will be significantly faster for pretty much everything except the actual act of rendering. Loading scenes, saving scenes, simulating, turbosmooth, displacement, mass fx, animating, viewport playback, all of pflow, etc. are all single threaded. These will all run much slower on the Xeon systems, especially when you overclock the 5960x, which can get a real, serious overclock. The Xeon overclock is almost non-existent. We just used e5-2697v3 (14 real cores), and were able to get a 5.4% overclock, which is a lot compared to what most people get (2-3% seems normal). The Xeon memory runs much slower as well (even with no overclocking on the 5960x).
As someone with both, honestly, I felt like it was kind of a waste as a workstation, and absurdly overpriced for render power, even if you figure the extra 300 or whatever in for the VRay license, and even with another 150 for deadline. We have a good sized farm so I just throw every test on the farm anyway, and continue working on other things. Everything but the render runs faster on the 5960 boxes, and it is no where near close to comparing core count * clock speed vs. core count * clock speed. In the end Megahertz matters so much more than Intel wants you to think. High core counts are really engineered for running multiple virtual machines with its own memory segment, not for all the cores trying to hit the same memory through a relatively slow bus.
I think you hit the nail on the head when you mentioned that the 5960X is probably a better option when you’ve got a farm behind you to throw your test renders and final renders on. In that instance it would be best to have better single-core performance and save the cash. But if, like me, you don’t have a farm behind you, just a few older machines that you’re using in DBR, then you really will benefit enormously from having better rendering performance from the dual xeon setup in a single workstation. Furthermore, these high-end xeons are actually not too far behind the top-end i7s (overclocked or not) for single-core performance in applications like max, since they’re quite terribly optimised anyway. I would also contest that the CPU makes much of a difference when it comes to Nitrous performance, but perhaps you can cite some sources for this? In my experience that will all be down to the GPU. And I would also mention that the Xeon ECC RAM is really not slower than whatever is usable with i7s, and if it was it would be a negligible real-world render speed difference.
I would also add that Autodesk are apparently in the process of trying to multithread more of max, including the items you mentioned, so when that happens (hopefully sooner rather than later), there will be even more reason to use the dual xeon setup. Bang for buck is a delicate thing. There are other factors like power consumption, space, cooling (air con), maintenance of multiple machines etc. Again, it’s a balance.
wether you can get a dual mobo for them with bclk adjustment in the bios is another matter.