UE4 Dev Rig: Threadripper 1950X or i7-8700K?

Hey guys,

I’m currently on i5-3570K with 24 gigs of DDR3-1866.

I’m certain my performance is no sufficient for UE4 development, in sense that cpp compile times are slow and load all 4 of my cpu cores, shader building takes quite some time, as well as asset loading and other things of that nature. Also I tend to have **a lot **of chrome tabs (150+) opened simultaneously (with smart Tab Suspender, tho), so I want to be able to work on all the stuff at the same time and not observe any significant hiccups.

I want to get the fastest thing for UE4 development, esp minimizing the compile times. I’m not playing games pretty much at all, so Threadripper’s relatively bad gaming perf does not bother me.

So my choice would be either TR 1950X with 16 cores, or i7-8700K. I understand that these guys are in different categories, but I do not consider intel’s Skylake-X platform at all: if I need more cores, I’d definitely take TR cause it’s cheaper and I wanna have AMD cpu if I can (developing games on different hardware in a team is beneficial for testing), but tests show that my own 3570K is **faster **than TR 1950X in single-core tests, while 8700K is much faster than my 3570K in single-core tests, but has much less cores than 1950X.

So, I would really appreciate some knowledge drop about how UBT uses cores, as well as overall opinions about “more, but weaker cores” vs “less, but faster cores” for UE4 dev in general.

Thanks :slight_smile:

Definitely not the threadripper - there were some benchmarks scattered around in other topics and it’s actually slower at compiling and lighting builds than my 7820X, while costing more to build a system with.

Can’t tell you about the 8700k as I never had one, maybe you’ll find more benchmarks around. UBT really likes more cores, as long as they aren’t slow.
A 7820X will rebuild the engine in 12-15 minutes while reading stuff on the forum with chrome.

Depends on your budget and what you want it for tbh. If I was making a big splurge on a workstation that I expected to last a long time, I’d get the Threadripper. The extra PCI-E lanes are a bonus, meaning you have plenty of space to chuck in NVMe SSDs as well as your GPUs.

I saw the benchmarks and some dicussions
Turns out TR is pretty slow in UE, like UE was built with intel instructions in mind or smth
8700 is out of stock EVERYWHERE cause of the rushed launch
So my main option now is 7900X. I’m thinking to take it instead of 7820X and others cause of higher core count and PCIE lanes, basically it looks like a more future-proof rig.
Problem is, of course, that it’s extremely expensive. Ordering from Germany, 7900X + mid-class x299 mb + AIO + 32 gigs DDR4-3000 is like 2000 eur.
I’d wait for 8700K restock, but it seems that it might take months.

If you look at this benchmark, the difference between 7820X and 7900X is pretty negligible:…e-Ryzen-7-984/

I don’t think it is worth the extra 400€ at all.
Sure it might rebuild the engine 1 minute faster but how often do you really rebuild the whole engine?

The 8700k is considerably cheaper than the 7820X, and in most benchmarks has better single and multi-threaded performance - so I know which I’d be favouring personally.

Yes, look at this benchmark comparing the two:…/m340638vs3937

It’s really no contest! If you can wait, I would highly suggest buying the 8700k as soon as it becomes available.

I’m running on an FX 8350 right now, so I definitely feel your pain, but I’m going to hold out for a bit so I can get the better CPU for the better price.

EDIT: It also seems that even thought the 8700k has two less cores than the Threadripper, it still makes more efficient use of those cores over the Threadripper and the difference is pretty stunning. Compared to my 8350, the difference is outright ridiculous! :stuck_out_tongue:

I would not trust those userbenchmarks for unreal engine. They also claim the threadripper to be 73% faster than a 7820X when tests done by people on this forum show it is in fact slower.

Two less cores? More like ten :slight_smile:

Yea, I got ur point, will have to hold on. My 3570K has served me well over the years, so I’m sure it can keep it up a little bit longer :slight_smile:

This said, you could build two 8700k machines for the price of a single Threadripper machine.

True, but I was posting the link to show it just seemed to be that the 8700K is a better value overall. I would definitely consult with people using Unreal who have upgraded to the 8700K and ask for benchmarks. Perhaps there could be a benchmarking thread put up here so that people could actually do that? If one doesn’t already exist, of course.

Sadly this wasn’t my experience. Opera performs the best on my system currently.

Ordered 8700K with 32 gigs DDR4-3200 and AIO that may get me around 4.8, judging by what ppl write. Should come next week. I’ll share some UE4 benchmarks here and my general impressions, once the system arrives.

There is an ad-hoc benchmarking thread here, but unfortunately there are only lighting builds: Intel i7 6800k or AMD Ryzen 7 1800X - General Discussion - Unreal Engine Forums
You’ll notice a 1950X has the best results, better than Zeblote’s.

But benchmarks distract from general usage which is harder to quantify, and Ambershee’s valid point that if you do need more cpu power then UE4’s most demanding tasks are distributable.

Threadripper will perform like a intel ONLY if you overclock the ram to 3000 or 3200 mhz.So if you dont plan on going over 64gb ram,than those speeds are no problem.I my self was about to buy a 1950x but since i need 128gb and with that ammount im sure i wont be able to go over 2666,i will go for skylake x.But as I said,if you need 64gb max,that just use the XMP ram profile and you will not lose anything.At 800$ now the 1950x is a no brainer.

Alright, I got the 8700K build last Monday. Worked & played with it stock, OCed to 4.8 today. Will send screenshots as promised, but here’s some general impressions after my 4c/4t i5-3570K in UE4.

C++ compile times got much faster, especially the hotreload fast iterations and the debug project startups. Subjectively, 2-3 times faster.
I have benchmarked the full compilation of the Shooter Game on 4.16.3, the results there are not as impressive, but bear in mind that this is the full recompilation (iterative would show bigger perf gap):
3570K @ 4.4 GHz, 24 GB DDR3-1866: ** 86.5 sec (avg)**
8700K Stock (3.7 - 4.7 GHz), 32 GB DDR4-3200: 53.55 sec
8700K @ 4.8 GHz: 50 sec

Does not scale well with faster CPUs, as you may notice from these results. CPU loads during compilation did not exceed 25% with 8700K.

BUT what **does **scale well is the shader compilation. UE4 loads all 12 threads up to 100% in the process, and shaders compile **significantly **faster than on 3570K and even my friend’s 7700K. No numbers for you here, sadly :frowning:

Nice! I just tried rebuilding that project myself, and it also took 50 seconds. Probably doesn’t have enough files to build in parallel to really make use of the extra 4 threads.

Lordlink, are you using Incredibuild? You would find your compile times would become better on the 8700k. There is “free” version of Incredibuild that will install with VS2017. I would suggest taking a look at that. I am building a new machine, the plan currently is to go with a 7900x at 4.6Ghz ( maybe higher ) With Incredibuild I do see all my cores/threads ( 4c/8t sandy bridge ) at 100% while compiling. I am looking forward to my faster build times on the 10c/20t cpu.

UBT can already use all local cores, if you only have one PC there is no point to use incredibuild. Its intended usage is distributed builds in office environments where you have hundreds of idle cores that can be used.

If UBT isn’t using all cores for some reason, you can manually set it to do that by placing this in AppData/Roaming/Unreal Engine/UnrealBuildTool/BuildConfiguration.xml:

<?xml version="1.0" encoding="utf-8" ?>
<Configuration xmlns="">

“Because the fix entails severing kernel memory entirely from user processes, patched OSes could potentially see a massive performance hit of “five to 30 percent slowdown, depending on the task and processor model””