Hello. I’m considering getting an Intel CPU and I’ll be doing physics simulations occasionaly with it since CUDA can’t do it. Is Blender able to use multiple CPU cores at once on one thing (softbody physics for a pillow for example) without hyperthreading or is hyperthreading necessary? I ask because the i5 is similar to the i7 except hyperthreading isn’t on the i5 and it’s 300$ cheaper. Sorry if this question is silly, I’m really not sure how stuff on the technical side works. I searched and didn’t find an answer.
Hyperthreading should provide no benefit. Hyperthreadings works best when there are many elements of processing coming through the CPU at different times. A physics sim is just one, large block of floating point calculations that streams through the CPU and doesn’t stop until its done. The only way to speed that up is to either increase the speed or the number of floating point units in the CPU.
The only possible place where it could be useful is if the CPU is for some reason unable to keep its FPU filled, but that’s rarely a problem with today’s programs.
And which CPU are you talking about? The most expensive i7 is the 4790 and it’s only $310.
Get the i7 you will thank yourself later when it comes render time. Blender softbody physics are still basically single threaded and perform poorly on any computer so don’t think you can just throw money or CPUs at it to make it go faster. You do want CUDA for GPU rendering if you are using the Cycles render engine, this mean purchasing an nVidia based graphcis card, not AMD or Intel.
I was curious so I benchmarked myself. I have an AMD 8350 which has eight cores but only four floating point units. When I render with all eight cores, I see a time of about 1:02. Reducing it to four cores increases the time to 1:20. Certainly not insignificant.
I ran a fluid sim and saw no difference between four cores and eight cores.
Yeah, CPU shopping isn’t easy, I’m saving up for a new build as well, I had planed to get the i7 4930k extreme [socket 2011], but since the price [£435.-] to performance is to little and power consumption is quite high, I’m going for the cheaper i7 4770k haswell [socket 1150] [£238.-] which has low power consumption but high temperatures (with my gigantic heatsink this shouldn’t be a problem)
The thing you have to remember is, what are you going to use it for… rendering, baking, etc.
Build you build around that and you’l have a great build.
Cheers,
Jim Morren
I ran a fluid sim and saw no difference between four cores and eight cores.
Certain processes can only be multi-threaded so much. Simulations, in my experience, only use about 25%-40% of all your CPUs even with multi-threading. However, ray tracing can benefit the most thus rendering speed can use all the CPUs at almost 100%. This is not just a Blender problem, other popular software packages like C4D and so on suffer the same fate.
I would recommend you reconsider this. You are going to spend maybe £1200 on a new system and you don’t want to spend an extra £200 to nearly double the performance?
Have a look at the benchmark results:- https://docs.google.com/spreadsheet/pub?key=0AoxhxwLZ7VLadGJxMEtKRldKTWJBTEVnN2VheVcyR1E&output=html
Close to 3 minutes for the 4770K and just over a minute and a half for the 4930K. Yes, benchmarks are unrealistic, but it gives an indication of performance. It does depend what you are using it for though…
I’m using an OpenMP build which maxes my CPU during fluid sims. My four/eight core comparison is a pretty decent analogy to Hyperthreading since I only have four FPUs. I re-ran my rendering benchmarks with larger blocks. My CPU is fastest with 64x64 blocks or smaller, but if I increase the tile size to 256 then then the render time difference between 8 cores and 4 cores is small. The difference persisted, though, so Hyperthreading will give you more performance.
And I agree with Organic: buy the 4930k. That chip is much more than Hyperthreading. It is six, full cores with six, full FPUs. It is amazing.