DevOps Workstation 3900xt/3950x

Hey folks,

This is my current PCPP list. Any feedback is welcome and appreciated.

New: https://pcpartpicker.com/list/FRK9zN
Old: https://pcpartpicker.com/list/dWwqV6

I feel like I’ve drained the last little bit of usefulness out of my 5930k, and the itch has become unbearable. I’m probably gonna hand a lot of my old rig down to my kid so she has her own PC, but there will be some part scavenging occuring.

My only real price concern is, I’m trying to stay under $1500 for the Mobo and CPU (I already upgraded my RAM with 64GB of wide QVL supported sticks, but still have the old 32gb kit).

My primary OS is windows, mainly because I’m too lazy to dual boot/setup my games with proton. I generally only have a single CentOS VM (4 core 16gb) running on a normal day, but can have up to 5 (open shift cluster), or an additional windows pro vm with IIS and SQL Express.

My day to day is usually a lot of docker builds, with python, angular and java. Occasionally it’ll be a kernel or some other OSS project. I also fill in as a dev a few times a week, and that usually means I’m running mongo or MySQL, + flask/Django/webkit/springboot. JMeter tests on the host OS while the app is running on the VMs is is also a pretty common workload.

If it were you, would you rather have the all core frequency potential for the majority of your workload, or the core count for mass VMs and Host OS stability?

Right now 6/12 core/threads definitely ain’t cutting it, and I feel the blockage every couple days with all cores pegged. I do still play games on this machine pretty regularly, so single core frequency is important, as is rock solid stability, so any overclocking is usually very mild.

P.S. I am pretty write heavy on my current SATA SSDs for VM disks, and haven’t really noticed anything being slow. The NVMe in the build will likely do a lot of lifting, for Steam and VMs (docker volumes), and I’ve heard mixed feelings on it. I think as long as I’m not expecting performance any better than my 850 pros I shouldn’t have a problem, but feel free to correct me if I’m wrong.

I’d consider swapping the noctua for a CLC. 280mm Arctic liquid freezer II costs more or less the same, but is more effective at same low noise levels, or rather it can be less noisy for same heat dissipation.

3 Likes

I guess that would preclude shrinking my current case footprint and retaining my current Define R5. 280mm rad doesnt fit in the define C from what I’ve read.

I’d take core count over single thread every single day, unless all you’re doing is gaming.

I think you’ll be impressed with the OOTB experience on the 3950x. It’s just a solid CPU. And since you’re keeping the 980 Ti for now, you’ll be fine.


Given your day-to-day, I think you’ll see a lot of benefit from faster storage. Are you set on the 660p? I suspect you might see better numbers out of a higher quality disk.

The concern to me is the QLC. If you’re doing a lot of writes, you should probably consider going MLC or TLC, to try and extend the lifespan, and provide better performance.

The 970 Evo would be a drop-in replacement with better performance. The 970 Pro would be an even better alternative, but it comes with a maximum capacity of 1TB, so you’d have to get 2 and RAID them together or do some magic with the partitions.


Also, I’m getting the impression you’re using Linux, but I just wanted to confirm your operating system, before I go making wildly inaccurate recommendations for you.

1 Like

I dislike closed loop watercooling for a number of reasons. Primarily, I find them to be unreliable in the long term, and considering that OP seems to keep computers for a longer period of time, I suspect the likely point of failure in this next machine would be a closed loop cooler.

Additionally, it’s important to note that almost all closed loop watercooling solutions are more expensive than air.

The idea that a AIO watercooler is more capable of cooling your PC has been thoroughly debunked by multiple sources for a few months now.

Now custom loop, that’s another story.

Simply put, air cooling is more reliable, and if noise is an issue, we can look to the Greg KH build to see how air cooling can be done with noise in mind.

GKH Threadripper 3970X Setup Notes

3 Likes

yeah I think this is really only gonna help me on GCC compiles, which are, admittedly infrequent. Single core boost clock is what I care about 90% of the time given the “single threaded nature” of python, which is the majority of my work.

I’m only set on the 660p because of the price. I try to be frugal where it’s appropriate, and use my money on the parts that’ll actually make noticible a difference.

So my initial thought is, keep the 850 pro (maybe both) in service for the hypervisor/host os (windows, to answer another question), And then have VM volumes and steam library on the 2TB NVMe. I dont FEEL like disk throughput has ever really been an issue except when "dive"ing docker images to make sure I’m cleaning them up correctly.

All my images are in version control and/or remote registry, only my pre-commit compiling is done locally, so if the drive were to die, it wouldnt be too big of a deal, and at most, I’d lose a day of work. It’d just be a checkout from version control and pull’s from upstream.

This is right on the mark. I generally take air cooling over water cooling for my own builds because of my wife’s history with asetek pumps dieing or getting unreasonably loud, where I’ve been on the same NH-D15 forever, and it’s just as silent today as the day i bought it. It also can cool the VRMs without a small fan built into the pump :wink:

My upgrades come few and far between, generally only when my hosted jenkins instances complete builds faster than my local machine does.

1 Like

I’ll remind you that single core performance is really close on all these CPUs. If most of your work is in Python, and it’s CPU bound, I don’t know what to say. Python is annoying in the sense that it’s not moved into the year 2006 and provided a method to bypass the GIL.

That’s fine, I just thought I’d point it out. Obviously the faster flash is going to cost a lot more :stuck_out_tongue:

I think that’s absolutely sound logic.

I think you’ll see that with significantly increased CPU power, you’re going to notice it more. With a higher thread count, your queue depth will likely rise, which is where the NVMe will shine.

but I don’t want to push you into a more expensive purchase on an assumption.

The big thing in my mind would be to make sure that when you’re compressing things, you use multi-threaded compression utilities. I’ve shaved hours off my work day by doing that, and I have a somewhat similar day-to-day, depending on the project.

haha yeah, that’s a good rule of thumb IMO.

It fits, check the case manuals on fractal.

This specific CLC is just happens to be really good for the money (in the ballpark of nh-d15).

Don’t get me wrong, I love my D15 in my old case, but in a define-C which is 40L including the PSU shroud space and with a 3950/980ti - how do you get the hot air out quietly?

positive air pressure.

Just fill up your front with intakes, throw your exhaust fan on the back and have at it.

I personally recommend the Meshify C over the Define C, but that’s just me. Much better airflow, and if you need smaller, there’s the Meshify Mini.

You do make sound arguments. Maybe save the $200 and just go with a 512GB 970 Pro for the boot drive with the savings.

And if i do that, i think i can just make the swap to drop the Pro SATA SSDs down to the kiddo, since I’ll have a bit peace of mind with a Samsung MLC drive handling the important stuff (i.e. the host OS).

I have, but there’s a difference between what the manual says, and what reality is. Top clearance is a bit sketch given the proximity of VRM heat sinks, and RAM. Front clearance is better as long as the GPU isnt long.

TBF, a similar thing happened with the D15 on this build. I was using a Gigabyte UD3 at first, but the x16 slot was a smidge too close to the CPU, and they ended up interfering.

I hear that, i just dont like that aesthetic. Admittedly my tower sits behing my monitors right now, so you dont really see it…but ill know its back there…not satisfying my aesthetic preferences.

Yeah, that’s fair. I know it’s definitely polarizing. I love the look of it, and it seems like the perfect mid-tower case. I wish they made it a bit less fractal-ish

Well, you can always spec it out and sleep on it.

dont you think its spooky how linus t’s TR build is spookily similar? :smiley: :smiley: :smiley:

1 Like

winkwink?

1 Like

I think it’s spooky for how much YT suppresses y’all in my feed, that it’s the one video they’ll recommend repeatedly.

It’s basically the Linus T build for someone who works remotely and has to finance their own work stations. Half the memory, half the cores (maybe fewer) and then maintain the focus on quiet, understated and reliable.

youtube be like

4 Likes

Well now I’m annoyed. I guess I shoulda ordered the Gigabyte Master the moment I saw it was in stock. Now, both it and the Taichi are out of stock everywhere.

I had considered possibly doing VFIO and moving back to linux for my host OS, but I guess I either have to wait, or write it off with the MSI Ace or ASUS Hero 8 :frowning:

Rona makes me sad. I guess I shoulda looked at the lack of PSUs and put 2 and 2 together, that there’s still a shortage.

Yeah, the shortage is going to last until all the international travel situations are sorted. And possibly even past that, due to ahem geopolitical tensions.

Luckily though, the hardware doesn’t stay out of stock too long. What I’ve had good luck with in the past is calling the newegg business line and asking them when they expect the part to be in stock. They might even let you, well not preorder, but you know.

I cant tell you how much i hate x570 cuz chipset fan. Do all the work to do a custom loop and you have that one annoying spinnny boy. No on really makes a block for it thats not some repurposed garbage and even on my Taichi ti would be a bit hard to mount nicely due to its location close to the top 16 slot. (not always an issue but when it kicks on its like MOTHER FUCKER)

For the sort of lab/dev environment you’re talking about IMHO cores all day every day. Anything faster than 3 ghz or so (IMHO) is plenty for doing things like real-time video/audio transcode even, and more cores means more granularity with committing resources to individual things, smoother overall experience, etc.

Sure, at the cost absolute single thread performance, but that isn’t what you’re doing.

2c.

Games? Don’t believe the hype regarding frequency. Sure, 5Ghz is absolute best twitch shooter pro performance in 1080p - but are you actually doing that and is that more important than work?

More likely you’re running games with signifiant detail settings (i.e., so the games don’t look like shit) and shooting for anything between 60-90 FPS (90+ for VR). And that just doesn’t require 5Ghz clocks.

Hence, again: I’d go cores over absolute frequency.

AS above though with the 3950X(T?) you can pretty much have both! Doesn’t QUITE clock as high as intel but its IPC is on par or sometimes better anyway, depending on the workload.

Yeah, I’ve slept on it, and I’m gonna wait for either the Master or Taichi to end up back in stock at regular price, and get a 3950x, and a 512GB 970 Pro.

Final: https://pcpartpicker.com/list/7XrThg

Thanks for the advice everyone.

P.S. don’t underestimate on how long it takes to start a webkit dev server for a giant angular project on a CPU with “low” frequencies. I can usually go scramble some eggs and grab a drink by the time my most frequent project is ready. Still faster than everyone else’s macbook though.

3 Likes