Description

Well this is one hell of an upgrade. For the past five years I've been using my 4770k/1060 computer and it was time for something more powerful.

This new computer is primarily for workstation use (hence the 32-core CPU). My workloads include all kinds of engineering work, primarily CAD and CFD. The 64 processing threads come in handy for workloads such as finite element analysis in structures, meshing CFD models more quickly, and of course solving CFD models more quickly. What's even more useful is that I can be using several high-performance applications simultaneously without compromising the performance of either of them. Based on Cinebench scores I've achieved about a 7.3x increase in multi-core performance compared to my 4770k (and this is without an overclock). And, to match the CPU horsepower, the immense amount of RAM is extraordinarily helpful for the sort of thing I do. Especially CFD can eat up this amount and even more RAM quite easily. Overall this CPU really makes my productivity experience a lot better.

The GPU horsepower of the 2080Ti comes in handy for some productivity tasks as well - it facilitates working with large CAD models and visualizing CFD results. I thought this might be a little bit overkill - but today I observed one of my CFD packages maxing out my GPU. So it's safe to say paying the RTX tax was worth it.

In terms of non-productivity work... I am an avid flight simmer. I was looking for a bit of a performance boost in X-Plane 11 and looking to upgrade to 1440p. This is where I am very pleased with Threadripper. It is an excellent workstation CPU and still a very capable gaming CPU (though it's worth mentioning X-Plane is more CPU-bound than other games). I'm seeing about an 8% single core performance improvement compared to the 4770k (and the i7 has a 300MHz clock speed advantage). This definitely shows up in X-Plane; I am able to crank the settings up pretty far and get acceptable frame rates.

In terms of issues that I had:

  • RAM does not run at the full 3200MHz. I was able to get it to 2933, but out-of-the-box it was at 2133. This doesn't seem to be too unusual with Threadripper (especially with how much memory I have) but it would be nice to get the full 3200MHz especially since Ryzen likes fast RAM.

  • The pre-installed IO shield did more harm than good. It had these tabs at the top which interfered with the case. There were no instructions in the MB manual as to what to do with them, and I had to remove them entirely to get the MB to seat properly. I found myself quite frustrated during the build process and I certainly expect more from a $450 MB.

  • My GPU arrived with a bent bracket to mount to the case. This meant the PCIe slot was misaligned and could not be installed. Bending it back was easy enough but I expect a lot more for a GPU of this caliber.

Other minor comments about the components I chose:

  • The case is overall very nice but I have a few gripes. It barely fits an EATX board. The bottom of the board is right on top of the PSU shroud which means things like USB3 headers are kind of jammed in as opposed to neatly installed. Additionally, due to the backplate on my MB, I had to cut the rubber cable management grommets in half in order for my motherboard to seat. With both of these issues - Corsair has more than enough room to move the MB mounting up a half inch and the rubber grommets to the right a half inch and solve both these issues. It would be nice to see this so it can fit EATX boards more easily. Additionally, a little more room in the back for cable management would be good, especially considering that with the cable needs of my system, I had no hope of getting the cable management bracket on. Final complaint on the case - when picking it up it seems that everything I grab is a movable door, removable filter, or some other non-rigid thing. I'd definitely rather have filters and such, but it means that I'm always worried about breaking something on the case when I'm moving it.

  • The motherboard, considering the market they're trying to hit, seems like it's a little bit off in the style department. Between the 10Gbps Ethernet and explicit statement that it was designed for the 32-core 2990WX, it's clearly a workstation board. Yet, they advertise it as a 'gaming' motherboard on the box, despite nobody in their right mind buying a 2990WX for gaming. And the gaming styling - between the eagle thing and the Aorus logo - looks kinda out of place on a board that, based on its features, seems to be aimed for workstation users. Overall though I think the motherboard looks decent, and I wouldn't buy a case with an enormous side panel window if I didn't want to show off the MB.

In terms of the value proposition, I think it is excellent. Obviously nobody should spend this much on a PC without a good reason to, but if you need serious horsepower for creative or engineering application, a Threadripper-based build delivers excellent value in my mind. As one point of contact - this cost about 2.5x as much as my old i7 build, and delivers well over 7x the multi-thread performance, and manages a noticeable improvement in gaming performance.

Overall though I can't overstate how good of an opinion I have of Threadripper and the fantastic changes to the PC business that's being brought about by AMD.

Part Reviews

CPU

What can I say... this CPU is a beast. It absolutely destroys my productivity workloads and is still a capable CPU for when it is time to blow off some steam and play games. Considering how well this CPU performs and its price, I think it delivers excellent value for those that need such a powerful CPU.

CPU Cooler

This cooler has quite good performance and has kept my 2990WX nice and cool. My one complaint about it is that Corsair Link consistently just doesn't work well. I can control pump speed, but controlling fan speed seems to never really work. But in the end, my CPU is cool and noise is low so I'm happy.

Motherboard

Overall quite a good motherboard. It is handling my 32-core CPU like a champ and has had minimal issues. That said, the issues I did have were quite frustrating. The preinstalled I/O shield did more harm than good; the tabs on top of it interfered with the case and there was no clear guidance as to what to do, and I ended up having to remove them entirely in order for the motherboard to seat properly. I expect more from a board this expensive.

Comments

  • 3 months ago
  • 3 points

nice build! always wanted to know if the threadripper could be kept cool with a good aio. +1 also, dont use corsair link, its hot garbage. use the ICUE software from corsair. you should be able to make a nice fan curve with it.

edit: as for overclocking the ram to the wanted speeds, look up Buildzoid on youtube. his channel is "Actually hardcore overclocking". some great tips on ram frequencies there.

  • 3 months ago
  • 2 points

Thanks! A good AIO definitely can handle Threadripper. I'm running 65C right now with all cores at 100% load doing CFD stuff. I'll have to get ICUE, link is definitely pretty crappy.

Thanks for the tip on the RAM OC, I'll check it out

  • 3 months ago
  • 2 points

Fantastic build - use it well :)

I had the same issue with the same Corsair memory - a 2950X build. In the end I changed to the GSkill 8gb sticks (F4-3200C14Q-32GTZRX) and achieved 3200 simply by enabling XMP.

  • 3 months ago
  • 1 point

Hmm. I might try to RMA my Corsair RAM then. I'm also seeing poor RAM stability. I received at least 1 error on MemTest and I keep seeing apps crash on Windows. Is this consistent with your experience with Corsair RAM on TR?

  • 3 months ago
  • 1 point

exactly . . . I had only installed 4 sticks of the 16gb Corsair's but good never get them close to 3200. Just not stable - 2933 was the max and I still had an occasional crash. I may take a shot at manual tweaking the memory settings to see if I can get to 3600 but the reality is I use this for mostly photo editing. Even converting 100's of images in Lightroom takes so little time now I don't think a memory speed increase is going to help much. I am manually overclocking the cpu to 3.95 GHz.

  • 3 months ago
  • 1 point

Interesting, you have a very similar setup (I have a profile for a 3.925GHz OC). Out of curiosity...what are your timings for the RAM? Also what are you running for voltage? I'm in the same boat that 2933 seems to be plenty but stability is really key for what I'm doing

  • 3 months ago
  • 1 point

Here are the timing values from Newegg's site (where I found them)

DDR4 3200 (PC4 25600) Timing 14-14-14-34 CAS Latency 14 Voltage 1.35V

As I said all I did was enable XMP - GSkill designates AMD compatible memory with an X on the end of the model #. I honestly don't know what makes memory processor sensitive.

  • 3 months ago
  • 1 point

Hmm, I'll give that a shot. Thanks for your help!

  • 3 months ago
  • 1 point

Nice build! But I am always curious about the performance of TR in CFD applications, since they are more memory-bound. What type of CFD are you running? What software? Would you elaborate a little bit on the performance or benchmarks?

I ran my own CFD codes (LBM/Boundary element etc.) on a 1950x and the speed is pretty impressive, but I heard rumor that TR lines, with their interesting memory layout, is not very ideal for common commercial CFD software with FVM solver.

  • 3 months ago
  • 2 points

Thanks! I perform CFD mostly to assess external aerodynamics of aircraft. I'm using Autodesk CFD Ultimate 2019. I use multi-threaded meshing with 8 threads and let it use as much CPU resources as it needs for the actual solve. Autodesk's CFD can use 2n cores (not threads), so it's using all 32 cores but only 32/64 threads. In turn, during the largest CPU load Autodesk can throw at it, my CPU usage is only at 40% which is kinda neat. I've also noticed that despite enabling multi-threaded meshing, it can only really take advantage of a single thread during meshing; I have one thread at 100% load and a few others at maybe ~40% load. As for performance... I've never made a point of gathering data and comparing to my previous 4770k. That said, during a typical use case for me (800k fluid elements or so), it takes maybe 30-40 minutes to converge. I'll run 2000 iterations but the software usually detects convergence and stops it around 400 iterations. During some of the more high-fidelity and more complex models (1.5-2M fluid mesh elements), it uses as much as 56GB of RAM and takes significantly longer to converge, on the order of several hours. I also suspect that running it on 32 cores is kind of overkill in that doubling the cores likely does not even come close to halving the solve time, but I've never tried it out.

The other package I use is Autodesk Flow Design, which is a tool used early in the design process to ascertain high-level aerodynamics and give rough numbers for things like drag. Running at the default resolution (which is plenty for a lot of what I do with this particular tool), it converges in a few minutes... which is absolutely fantastic.

I've never compared this to a CPU designed with a more conventional memory layout with a similar core count, but I get the feeling that TR's odd memory setup probably isn't fantastic for everything, but in my view, the large core count makes up for that.

Not sure which benchmarks you're looking for... but Cinebench multi-thread is 13798 and single-thread is 401 at an all-core OC of 3.925GHz.

Hope all this helps!

[comment deleted by staff]