Download

Light Building Benchmark

Hi,
I am considering buying a workstation to help reduce light building times, but I realized the cost of Xeon processors and other parts is a bit high. Therefore, I wanted to know what performance gains I will be getting and whether it’s worth that cost. I googled light building benchmarks and didn’t get useful results.

So would you guys be interested in doing a light build benchmark using your PCs so that we have a better idea of assessing the need for hardware?

If there are enough submissions, I’ll try to make a wiki page and post a results table for everyone to see. We’d need to agree on a free archvis scene so anyone can download it, and what information to include in each benchmark (CPU, motherboard… etc).

You dont need a “workstation” with crappy Xeon processor.
Every desktop with slightly overclocked 6700k and 32gb 3200 ddr4 is way faster for building light :slight_smile:

I have a 6700k with 32g on ram, and until know, it’s been a smooth experience overall. If you need more horsepower, use more PC’s with swarm “farm style”. Only thing i don’t know how to speed up it’s compiling for mobile.

I don’t think Swarm can be used to compile a project… maybe I’m wrong but I think it’s just for lightmass.

Lightmass is all about having a solid processor and enough ram, no other specs really matter.

If you want the best performance without spending a ton of money, get an 4 core 8 thread i7.

If you have a ton of money and don’t want to build a render farm, get a Intel Core i7-6950X. Or 2 CPUs on one motherboard if you want to get fancy.

If you have a lot of money and want to build a render farm, just get a bunch of i7 processors.

There’s no real need for a complicated database.

I take it there isn’t enough interest for creating a benchmark table.

In any case, thanks for the replies, I didn’t know the i7 can rival some xeon processors.

I wanted to make my own benchmark table soon but my 2nd i7 pc has problem at the moment. Will try later I guess.

Some of these comments are ridiculous. There’s nothing crappy about a Xeon whatsover. You can’t overclock a quadcore to get remotely close to a 12 core + rig. You can’t even take a 4.5Ghz i7-6950x and expect it to compete on the same field as a dual socket 64 logical core Xeon rig.

I’m also interested in a standard benchmark and may end up making my own so we can have some objective measurements to work with. I was having issues using swarm agent to get more than 8 cores for lightmass and a production build was over half an hour. Now that the farm is rocking in full swing, my build time is less than 2 minutes. Try to do that within the same power envelope with a bunch of i7’s networked together. It’s just not happening.

Guess I don’t have to mention they are all rack-mounted at my residence and ready to roll-out to the datacenter for remote operation once my business gets going. i7s make cheap workstations, but they are no substitute for the real thing.

I totally agree with webindustry. My new workstation with dual Xeons ( 24 Logical with 48 Cores) is way too fast (almost 8 times) then my other two PCs with i7 5930K combined. In my opinion, if you are doing commercial works in UE4 - Going for Xeon is the most logical choice (with few projects, you can easily recover the cost). Lightmass is totally CPU based so other hardware does not affect the rendering performance. Go for as much RAM as possible.

On the other hand, if you are only starting with lightmass and UE4, going with a i7 processor is safest way (either 6 core or 8), so you can at least test few scenes and settings in UE4 without waiting for days to complete the light build. Calling a machine with i7 a workstation, is a misnomer.

Yep, I will chime in that this is utter horseshit. We have swarm running on local machines here(i7-6800ks oc’ed) and we also have Swarm running on Xeon 32core machines. The bake times on the 6 core 6800 is about 32-40 hours for very complex maps, it is between 4-7 hours on the xeon. We also use swarm coordinator locally over the network to bring those times down even more. IT’s not a 1:1 performance gain networking machines but it definitely makes a difference. We have been struggling to get external machines outside of the office connected, they all show up in Coordinator but the external machines will not receive job assignments and we’re at a loss as to why

Id love to help test a Arch viz scene out :smiley:

On a more amusing note, there are 4-core Xeons, with no better stats than an i7, and some even without hyperthreading. E.g. The E3-1220 v6.

All spot on with our experience too. We got LM working submitted from a local workstation to an Azure VM (a super-simple always-on coordinator, and a super-high memory LM cruncher VM) … but the “transfer” time was insane – more than the processing time. We needed to implement a Source Control server that would be on or sync with the Coordinator because it was always-on, and then sync to the LM cruncher VM’s.

@ZacD - have said it, CPU and RAM that’s all you need to build huge lightmass with reasonable time.

But on other hand, it all depends on how much complex work you are going to do in foreseeable future (like 2-3 years). I myself have dual Xeon configuration with total 48 cores with 64 GB ECC Memory (people tends to forget how life saving ECC is), it’s been 5 years and I never felt a need of upgrading the setup. It served me well and I was able to recuperate the cost within 8 months of investing. If you are planning for commercial work, I highly suggest you go with AMD Threadrippers, awesome price/performance ratio.

Intel Xeons are not all feasible in current scenario of AMD and their really bad price/performance ratios.

Consumer level CPUs i.e i9 and i7 are also good choice if you are only beginning in Unreal Engine and not planning to meet deadlines on huge projects. but they lack ECC support, which a huge drawback.