Need Cinema 4D- 3D rendering and After Effects Machine

[PG37]

Haven't built anything since my workstation in 2013, which i'm writing this post on and still works great!

I have a graphics artist that is using After Effects (CPU clock speed biased), Cinema 4D (CPU&GPU), and Octane Renderer (GPU biased).

She needs a new build that hopefully has the flexibility to add additional GPUs in the future. We currently have a GTX 1070ti, GTX 770, and GTX 670 that we could use for rendering, although I suspect we mainly would keep the 1070ti and just get another 1070ti or better.

I have monitors and keyboards and what not, so just need the following:

-Case that ideally can house a motherboard that can hold 4 GPUs or have a connection for an external GPU housing
-PSU that can run 4 GTX1070ti's or better
-Highest possible CPU clock speed, keeping in mind bang for the buck.
-Most threads available while keeping the highest clock speed (Threadripper 1950x for example)
-GPUs that offer the best cost/performance ratio. (More cards would probably be more efficient compared to higher end models at the same total cost)
-At least 32GB of RAM
-2 SSDs (one reliable one for OS and one for cache)
-One spinner for larger file storage
-32GB of ram minimum

I live in Seattle, WA.
Budget would ideally be less than $2k with a single GPU1070ti or lower (thinking about 3x GTX 1060)
No Overclocking or RAID needed
Monitor resolution will be 1920x1080 maybe up to 1440p

 

[sinisterDei]

I have some questions.

since my workstation in 2013, which i'm writing this post on and still works great!

I obviously don't have context for what this workstation contains aside from possibly a GTX 1070 Ti and then circa-2013 hardware. That's IVB-E/Haswell type generation kit, so presumably you've got either 4 or 6 CPU cores now.

Your baseline is important, I think, because you said your system "still works great" which I would interpret as "is capable of performing the work acceptably". If that's the case, you could obviously go nuts and try for the best possible system you can build for $2k and wipe the old system off the map. OR, given that your current 4-6C 1-GPU system is acceptable, you could target a more modest upgrade.

In either case, I think the AMD platforms are currently going to be your winner.

If you stay away from the HEDT space (no-Threadripper) you tend to get fast cores and a much less expensive platform over all. Here is a build based on a 2700x with 32 GB of RAM, dual SSDs (which is likely overkill), a spinning disk, and a mobo that can handle 3x PCIe x16 slots. Note, the third slot is PCIe 2.0 x4 electrically, so if your work is GPU interface intensive then they might struggle on this slot. The first two slots are 3.0 X8 when you use two of them. Obviously that build has 3x GTX 1070 Tis in it, and comes to ~$2400, of which ~$500 is CPU and motherboard. You get 8 very fast CPU cores, 16 threads, and an inexpensive platform. Also, you could swap the RAM out for ECC stuff if you care. As an added bonus, you avoid any NUMA weirdness that would come with the HEDT platform.

On the other hand, if you went with something like a 1950x, the CPU + mobo cost will, at a minimum, jump to ~$990. Buying the newer 2950x with mobo moves that cost to closer to $1300. Whether that uptick in costs is worth the tradeoff depends on your specific use. The 1950x/2950x obviously buys you 8 more physical cores, and the X399 platform gets you 8 memory slots (install memory in quads) and four PCIe x16 slots that all run at either 3.0@x16 or 3.0@x8 electrically. It's nice, but it's more expensive overall. Here is the same build from above, but with a 2950x in it and double the RAM. Please note, on the 2950x build I didn't check the RAM vs the QVL for that motherboard, I just doubled the sticks to 64 GB. If you picked that system up, you would also likely want to move up a power supply tier into the 1000w range if you planned on four GPUs. PCPartpicker doesn't like you adding >2 GPUs to a system so it's not calculating in the third GPU's power use (around 175W) and the system is at 590W with just the two GPUs in it. The system would be at ~950W with 4x 1070 Ti GPUs in it.

 

[PG37]

.

Your baseline is important, I think, because you said your system "still works great" which I would interpret as "is capable of performing the work acceptably". If that's the case, you could obviously go nuts and try for the best possible system you can build for $2k and wipe the old system off the map. OR, given that your current 4-6C 1-GPU system is acceptable, you could target a more modest upgrade.

In either case, I think the AMD platforms are currently going to be your winner.

Yeah I was mainly pointing out how helpful Hard OCP has been. My old 3930k system with 64GB of ram still works well for after effects and editing but my graphic artist needs are much greater.

That being said I obviously need brush up on some of the issues with threadripper. I think your advice of doing the modest upgrade is worthwhile. The higher clock speed will likely prove more useful than the extra cores. After effects has diminishing returns after 6 cores.

Thanks for he build. I’m going to read up on this HEDT space.

 

[sinisterDei]

Specifically for Threadripper, you just want to make sure your applications are "NUMA-aware". NUMA stands for non-uniform memory access. The Threadripper CPUs are based on the Epyc server CPUs, and internally you can think of them as 4x Ryzen 8-core CPU nodes (called core complexes or CCX) that are glued together via AMD's Infinity Fabric. Each CCX brings a 2-channel DDR4 memory controller to the table. In a Threadripper CPU, 2 of the 4 CCXs have their memory controllers disabled, and on CPUs with less than 32 CPU cores then some of the cores on each of those 4x CCXs are disabled to get down to the correct number of cores for the specific SKU.

The reason this is important is that the system will function slightly more like a 4-socket system with 4x 8-core CPUs in it than a 1-socket system with a 32-core CPU in it, depending on the application. If CPU 1 needs to access some memory and that memory is attached to CPU1's memory controller, no big deal everything is kosher and fast. If CPU 1 needs some memory, but it happens to be attached to CPU 2's memory controller, then there is a performance penalty to reach across the fabric. Applications that are NUMA-aware are able to figure out the architecture of the CPU cores and which memory is local to them versus remote and allocate tasks and processes accordingly, at least partially negating the NUMA penalty by simply avoiding it. Apps that are NOT NUMA-aware may simply blunder over the lines of memory and be really inefficient. And some apps that aren't particularly memory bandwidth intensive may not care one way or another.

Here is a good read on the topic.