I have a general question about CAD performance and Rendering.... I know this is a gaming forum, but your youtube videos are great and led me here.
1) If i were to spend the same amount of money on a Dual Xeon vs a faster Singe Xeon, which would be preferred for day to day CAD use? Faster Clock Speed is generally better for daily CAD tasks, but I dont know if it is a bottle neck at all anymore.
2) I know I'll eventually be using 2 Quadro's in SLI mode together, but I also know that I cant use 3 or 4, SLI only scales to 2. Thus, am I wasting money by going with a Asus Z9PE like in your Dual Xeon video??? It has 4 full x16 slots and that seems like overkill.
3) Can anyone link to me an explanation of where the Dual vs Single xeon really benefits? I know its great for rendering but I am not thoroughly understanding how exactly it impacts rendering.
Thanks for you're help, its really hard to find info in detail about how CAD specifically uses various hardware things. For example, from what I understand - and correct me if I'm wrong - neither Solidworks, Bunkspeed iRay, or Keyshot do not benefit from ECC ram, and i found some people saying to turn off ECC, which just seems silly to me (why save 2% of time if you could spend that time and guarantee not having a crash), and I also read those applications do not really use hyper-threading either. Both of things lead me believe that I'm probably better off not spending more money on ECC ram and only getting a hyperthreaded processor if there are other non-hyperthreading reasons to get it.
Note: there are no gaming considerations or preferences for this build - CAD & Render only. Thanks :)
1) One stronger CPU is better for Solidworks CAD 2) Quad GPU solutions are not that good. The best way would be one strong AMD card as Solidworks benefits from OpenCL more then from CUDA. 3) A video like that might exist somewhere, the problem is the minimal audience for such a video. Having only one CPU eleminates some problems you can run into with multi-anything setups. Best bet is one strong CPU and one strong GPU.
As Streetguru said, you want a cpu with fewer more powerful cores unless you are working with complex fluid or material/thermal stress simulations. In case of the 5820K try turning hyperthreading off as it seems to slow down rendering as Solidworks is kinda... bad with high core counts.
So, what applications are you going to be using? Solidworks? Maya? AutoCAD?
Anyway, I'm gonna recommend you to look at Puget Systems hardware articles. They've done plenty of articles on this kind of stuff.
For example, here's their Solidworks 2016 CPU Performance review with Skylake-S vs Haswell-E/EP
And the multicore performance of Solidworks
Basically some things don't scale at all 1core vs 20cores, some things scale well up to 4 cores and dive after that, some things scale up to 8-10 cores while some things scale nicely all the way up to 20 cores (like rendering).
It will primarily be used for Solidworks, but I want to get into Rendering now that Solidworks bought Bunkspeed, they call it Solidworks Visualize.
I'm thinking of the Xeon E5-2643, which has turbo frequency up to 3.5ghz, so it should be plenty fine for day to day use. The big question is do I get it as a Dual Xeon or a single? I dont really need the dual for most applications, but at those times when I am rendering, there is a huge amount of data that is to be rendered on the host computer, even before it gets farmed out.
That is relevant because my current computer, an i7, I will try to keep around and put in dual quadro with SLI in the future, and use it as a local render farm. Since the i7 has gigabit ethernet it only looses 4% to network loss, and Bunkspeed is optimized for both processor cores and CUDA so running 2 quadro will theoretically be really efficient. BUT, the big But, is that between 15-40% of the total render must be done on the host computer, before the remaining percent can be distributed between the host and networked computers. So I'm wondering how the Dual Xeon will affect it all, and just how much better it will do. My heart wants dual, my wallet wants single, the struggle continues.
More CPU is better, in general. Rendering will also go faster unless you are doing GPU rendering. Photoview is a CPU based rendering solution. More CPU = better.
I don't think there are any benefits to having multiple GPUs other than raw GPU rendering power and display support. I will defer to @CrossCarbon for the OpenCL/CUDA support as I have only ran Quadros for Solidworks. I personally CAD on an Intel iGPU and gaming hardware, so yeah, pleb level cad hardware.
More CPU cores/threads means that more of an image can be rendered at once. With photoview and things like Cinebench or w/e after the prelim framework is made (or something like that) the image is split up into hundreds or thousands of sections, each one a certain size and containing a varying amount of info to be rendered. Each thread that you have you will be able to work on one of the "sections", so if you have 8 threads, then 8 will be worked on at once, if you have 32 threads, then 32 of them will be worked on at once. This parallel work will end up working faster as the more complex sections take more time and the less complex go faster. So the image will be filled even as the complex sections are worked on.
With ECC RAM, it will really only affect rendering. If you plan on massive renders, it may make an impact on the renders that you put out. I have had renders on normal gaming hardware that ends up screwed up, and the only thing that I can work it out to be is errors that could not be corrected at some point of them rendering process. General work will not be affected by ECC, or even the lack of it.
Puget is a great place to look. They make me sad though since it'll be a while before I can work with the hardware they are talking about. XD
So are you looking at used CPU's or is there a specific reason you're looking at Sandy Bridge Xeon-E5's? At least the MSRP of that E5-2643 is 884$ which is a bit ahem these days, for a 4c8t CPU. Amazon.com they're almost a thousand dollars. If you look at modern or rather current Xeon E5-26xx v3 (Haswell): 900 USD gets you a E5-2640V3 (8c16t, 2.6Ghz/3.4Ghz base/turbo). In Germany that thing starts at around 960€ or UK about 775£..
Or if you want dual then ~800 USD get's you 2x E5-2620 V3 (6c12t, 2.4Ghz base/3.2Ghz turbo) for example.
But, the most important question. What country and what's the budget we're looking at for the whole thing? And what kind of a slice of the whole budget do you wish to allocate for CPU/GPU.
Since if I'd have the money, I'd probably go for a single E5-2687W v3. But it's prices start at ~2000$/2260€/1760£. That's the 10c20t variant with 3.1Ghz base/3.5Ghz turbo. Nice high clock speed and ten cores.
I will look into that, thank you for the recommendation! I am determining how much i want to spend on the Xeon vs the Quadro, i figure i should decide long term which CPU combo is truly in my best interest then just get a faster quadro in time when the budget allows, depending on funds then. I have a k2000 starter card that will hold me over for the moment.
Pcpartpicker though includes E5-1xxx V3's which are 1 socket only, E5-2xxx is the dual socket variant. Geizhals has only the E5-2600 V3 filter selected.
Broadwell-EP Cpus should be coming out in Q1/Q2 2016 and will work on the X99 chipset used by Haswell-e. If your going to be rendering traditionally through Solidworks then a dual socket X99 should be good, get one cpu now then add another if you would require parallel rendering or increased performance. Another bonus would be the increased RAM speed with DDR4 support. ECC memory isn't a requirement of or supported by Solidworks, a cheaper i7 variation may be a better choice than a Xeon. I'm not familiar with Bunkspeed or now SW Visualization but it seems that it unloads much of the rendering process to the GPU unlike PhotoView 360. In that case strong or Multiple gpus would be more beneficial. Like @lagittaja links stated most of the SW processes are single core except for rendering and multiple drawing view generation.