[alexp88]

Greetings,

I just finished building my first workstation. Here is the configuration I have:

Motherboard: Supermico X10DAX (just 1 CPU installed)
CPU: Intel XEON E5-2699 v4
GPU: nVidia QUADRO K1200
RAM: Crucial 288 Pin 32 GB (16 GB x2) CL15 DDR4 DIMM Memory Module (ecc registered)
Storage: Crucial MX300 525 GB SATA 2.5 Inch Internal Solid State Drive with 9.5 mm Adapter AND WD Red 4 TB NAS Desktop Hard Disk Drive - Intellipower SATA 6 Gb/s 64MB Cache 3.5 Inch

Though the system is quite high-end, I am totally unhappy with the overall performance. Quite honestly, my other workstation, a Dell Precision M3800, is performing way better. And here's the configuration as well, so you can compare them:

CPU: Intel i7 -4702HQ @ 2.20GHz
RAM: 16GB
GPU: nVidia Quadro K1100M

The issues I'm experiencing with the homebuilt workstation include lag, sluggish performance, and occasional freezing. This happens especially with Adobe Photoshop CC 2017 (most recent version). The overall response is very slow after about 15 minutes. It becomes unusable.

I have come across this thread (https://www.cfd-online.com/Forums/hardware/144936-singl...). The OP seems to be experiencing something quite similar. Now, one of the users has raised an interesting point about having created a memory bottleneck. I have just one processor and 2 RAM modules (16 GB each, so 32GB total). After reading the thread I realized I hadn't installed the RAM modules properly. My motherboard has 8 modules/CPU and dual CPU capability. The 8 modules are distributed evenly on each side of the processor (4 on the left side and 4 on the right side). When I installed the RAM, I place one module on one side and the other module on the other side. I eventually moved one module to enable the dual-channel memory architecture. I placed them in slots 2 & 4, which I think is the correct way.

The performance is still quite poor for a machine this expensive. I also came across maximum memory bandwidth, and I'm confused if I should necessarily try to compute/achieve the CPU's maximum memory bandwidth of 76.8GB/s. Will this make a difference in how the system performs? As I have only 32GB of RAM installed, this essentially lowers the memory bandwidth to about half, so maybe the performance is also about a half. Should I add another 2 x 16GB RAM modules? I know each core needs to be allocated a minimum of 2000MB.

Reading the same thread, one of the users says it's important to populate "bank 1 of all 4 channels for each CPU". What exactly does this mean? What is bank 1? I know a channel is, for example, A1, A2, and the other is B1, B2 (so I'd need to place a RAM module in the first slot of each channel, i.e. A1, B1, which I did). How many more RAM modules would I need to achieve optimum performance?

I did a few tests to stress the CPU and GPU. I used Furmark for about 30 minutes and all was fine. No artifacts whatsoever. Furmark hasn't crashed at all. I used Prime95 to stress the CPU for about 3 hours. No problem here either. For RAM, I ran Memtest86 for 10 hours. The results were 2 passed and no error. So no problem here either.

As I said, this is the first workstation I've ever built, and I'm not experienced with hardware. So any input/advice would be greatly appreciated!

Thank you!

[flotus1]

The link you posted was shortened, so I don't know which thread you are referring to.

Anyway, here are a few things that come to mind:
Did you consult Supermicro support in advance to find out about possible issues that arise from using only one CPU on this motherboard? You should at least study page 1-10 of your manual to find out which expansion slots belong to which CPU. Are you planning to upgrade with a second CPU any time soon? Otherwise you would be better off with a single-CPU motherboard.
Did you flash the most recent Bios version 2.0b? What is the revision number printed on your motherboard?
Did you connect the additional 8-pin power supply correctly for the CPU?
Refer to your manual page 2-14 on how to populate the 2 DIMMs. It would be best to put them in A1 and B1, provided you installed the CPU in socket 1. The names of the slots are printed on your motherboard.
Memory bandwidth is an issue, but not so much for with Photoshop than with CFD, but for feeding 22 cores providing maximum memory bandwidth is a good idea even for "non-CFD" workloads. At least 4 DIMMs are needed to get the maximum memory bandwidth for one of these CPUs.
During your tests with Furmark and Prime95, what did you monitor to make sure everything is fine? Temperatures? Frequencies? Which CPU cooler do you use?
And last not least: What kind of workload do you have? Are you sure that 22 rather slow cores are your best option? Or would a lower amount of fast cores be more beneficial? Puget systems came to this conclusion for Photoshop CC:: https://www.pugetsystems.com/labs/ar...rformance-625/
If Photoshop is the program you use the most, the CPU and motherboard choice was far from ideal. It seems like you really should have asked before buying.

[alexp88]

Sorry, here's the original link: single i7 MUCH faster than dual xeon E5-2650 v3 !!!

[alexp88]

To answer your questions:

1. I haven't contacted Supermicro in regards to possible issues that might arise from using only one CPU on this motherboard. However, I've been reading about using just one CPU on dual-CPU motherboards and the most likely issue that might occur with LGA2011 platforms is PCIe lanes not terminating so that Ethernet and SAS controllers may not work. I will contact Supermicro, but I expect this is not the culprit of the poor performance.

2. The bios version is 2.0b, 10/5/16


3. I have connected just one 8-pin power supply for CPU 1.

4. I have installed the 2 DIMM modules, as per my original post, in A1 and B1. As for maximum memory bandwidth, I figured that it won't be an issue with Photoshop.

5. The CPU fan/cooler I use is Noctua NH-D15.
I monitored the temperatures and frequencies while running Prime95 and Furmark, and all was fine. Anyway, the CPU and GPU have both passed the tests.

6. My workload typically consists of heavy Photoshop, Premiere Pro, and Dreamweaver use (and sometimes more, i.e. After Effects). I really wanted more stability rather than faster cores. I looked specifically for an option with a greater cache memory. Even with slower cores, Photoshop and also the entire workstation should still have literally flied, if you know what I mean. Returning the CPU is not a valid option in my case, and I wouldn't want that anyway.

I have selected the hardware components of the workstation to future-proof the it, as I might decide to install a second CPU in the future, and upgrade the other components as well. In fact, I've spent quite a lot of time doing research about the best configuration for my workload, and I noticed that most users tend to also choose Xeon CPUs for workstations. Maybe a CPU with fewer faster cores is another good solution for gaming, but not for a workload where stability is wanted.

I am starting to suspect that the actual issue may be the OS, Windows 10 because, as I mentioned above, the performance keeps fluctuating from very good to very poor. For example, it worked well yesterday, but after booting it this morning, I started to experience the issues I have described in my original post. This means something is happening when the workstation boots (not every time, of course).

Anyway, I will try to install a different Windows version and see how things go, but I'm still waiting for suggestions/advice. Thanks!

[flotus1]

I can not comment on issues concerning Windows OS or Adobe products.
So here are my last few general comments:

I really wanted more stability rather than faster cores.

These are not conflicting requirements. CPUs with lower core counts are just as reliable as their counterparts with higher amounts of cores.
Try to find out what causes the poor performance after working with your program for a few minutes. If it is insufficient memory, this can easily be seen in windows task manager. If the PC has a more general performance problem, this can perhaps be found out by comparing benchmark scores (i.e. Cinebench, Aida 64...) with known values for similar systems.
The conclusion from Puget Systems for Adobe products were:

• most of the workload for is single-threaded
• if it is multi-threaded, there is virtually no scaling beyond 6 cores
• using a dual-CPU setup even has a performance penalty

Since they sell workstations, it is safe to assume they would recommend expensive dual-CPU setups with pricey CPUs if these were any good for the workload.