Chess workstation - home edition - noise

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

Doppelbauer

New Member
Aug 8, 2016
13
4
3
64
Build’s Name: Chess Workstation / Home Edition
Operating System: Windows 10

CPU: Dual E5 2683 v3, CPU-coolers: 2x Noctua U9DX-i4, standard front to back fan orientation, low noise adapters

Motherboard: Supermicro X10DAX

Chassis: Supermicro SC732D4F-903B, Mid-Tower. Case fans: 1x exhaust case fan SM 12 cm (came with case), 1x intake Noctua 12 cm.
Drives: SSD, Crucial MX200 - 1TB
RAM: 128 GB, 8 x Crucial 16GB DDR4 2133 ECC
Add-in Cards: Asus NVIDIA 760 mini

Power Supply: PWS-903-PQ, 900 Watt, 80 plus Gold (came with the case)

Usage Profile:
Extended chess analysis, using standard chess GUI and chess engines, e.g. Stockfish 7 and Komodo 10. Optimal analysis with chess engines typical uses all CPU resources. System should be stable under full load 24/7 but still be relatively quiet, as the box is sitting in a living room/apartment.

Build and Performance:
Build was easy, system is performing to specs, 3300 CPU-score in Cinebench R15, Chess benchmark: Stockfish chess engines >30'000 kN/sec in base chess position (multithreading off, as suggested for chess analysis). Typical max core temps under extended full load are stable at 54 C and 64 C for the 2 CPUs, CPU fans spin at 1300 RPM, case fans at 950 RPM

Noise Levels:
System can be reasonably quiet even when under full load from chess analysis, exhibiting moderate levels of typical low frequency fan noise, but only if side panel is off/removed. PROBLEM: when the side panel is mounted and system is under full load (as it will be most of the time) noise level increases slowly by about 10 dB (guess using a phone) as opposed to full load and no side panel mounted. The additional noise is annoying and distinctively different from the CPU fans and case fan noise. Opening the side of the box halfways, giving the PSU breathing space, eliminates the noise problem.
Other observations: Side panel mounted/off does not influence core temps or fan RPMs. Disabling the CPU hyperspeed options in the BIOS the X10DAX to reduce power consumption did not help. My guess: the additional noise seems to come from the PSU/PSU fan, due to a combination of relatively high load and and high ambient temperature.
Any ideas how to reduce or eliminate this additional noise? Some ideas: (1) Replace PSU (costly, outcome uncertain) (2) mod the case and flip the PSU on its back to let it take in fresh air from the top (add dust filter), (3) mod the side panel and add an intake fan to the side panel (4) simply use the system without side panel.
Thank you.
 
  • Like
Reactions: Patrick

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Doppelbauer - what OS does this use? I was looking for a good chess benchmark.
 

Doppelbauer

New Member
Aug 8, 2016
13
4
3
64
My system runs Win 10, and standard Chessbase GUI software, but that's just the GUI. The engines matter. Stockfish 7 is available for many platforms.

Chess benchmarks, nontrivial issue. :)

Mostly kN/sec (kilo nodes per second) are reported, the number of chess positions evaluated by the system per second. My system seems to evaluate 30'000'000 positions per second, in the starting position of chess.
Sounds nice, but kN/sec are no absolute measure of analysis / playing strength, as they depend on the specific position used (e.g. basic chess starting position, vs a position 15 moves later) and the engine used. Eg. Stockfish 7, best free engine will report higher numbers than Komodo 10 (best commercial engine) using the same hardware. It seems likely that in terms of real playing strength, even some penalty should be given to results from higher core counts, as the analysis task does not scale perfectly to higher number of cores. Therefore, some people have called for depth of search, expressed in term of half-moves, plys, per time as a better benchmark. E.g. how long does it take to calculate up to 30 plys (half moves)? Again, such a measure would be highly program and position specific, as the selection and dismissing of moves (pruning) are very different among chess programs. As Stockfish is available for may platforms, in practical terms, kN/s in the basic starting position is a reasonable (not perfect) benchmark. The Stockfish engine allows you to run it in a benchmark mode from a line prompt in Windows, so there is not even a need for a OS specific GUI. I guess this would be the same in Linux. Hope this helps.

Noise
As you own the same SC732 case, what do you think, how can I quiet this thing under full load? :)
 
  • Like
Reactions: Patrick

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
My system runs Win 10, and standard Chessbase GUI software, but that's just the GUI. The engines matter. Stockfish 7 is available for many platforms.

Chess benchmarks, nontrivial issue. :)

Mostly kN/sec (kilo nodes per second) are reported, the number of chess positions evaluated by the system per second. My system seems to evaluate 30'000'000 positions per second, in the starting position of chess.
Sounds nice, but kN/sec are no absolute measure of analysis / playing strength, as they depend on the specific position used (e.g. basic chess starting position, vs a position 15 moves later) and the engine used. Eg. Stockfish 7, best free engine will report higher numbers than Komodo 10 (best commercial engine) using the same hardware. It seems likely that in terms of real playing strength, even some penalty should be given to results from higher core counts, as the analysis task does not scale perfectly to higher number of cores. Therefore, some people have called for depth of search, expressed in term of half-moves, plys, per time as a better benchmark. E.g. how long does it take to calculate up to 30 plys (half moves)? Again, such a measure would be highly program and position specific, as the selection and dismissing of moves (pruning) are very different among chess programs. As Stockfish is available for may platforms, in practical terms, kN/s in the basic starting position is a reasonable (not perfect) benchmark. The Stockfish engine allows you to run it in a benchmark mode from a line prompt in Windows, so there is not even a need for a OS specific GUI. I guess this would be the same in Linux. Hope this helps.

Noise
As you own the same SC732 case, what do you think, how can I quiet this thing under full load? :)
Awesome! I will look at that when I get back from Machu Picchu. I may send you a PM then.

Right now I am using E5-2650 V3's. Going with the larger tower coolers (Hyper 212's fit) and then adding an auxiliary 120mm fan works really well. Most NAS systems are significantly louder than my 2P workstation. You have much higher-end CPUs though!

If I had your CPUs I may be inclined to see if any other cooler would fit. More heatsink surface area may help.