Building engineering number cruncher.

[dmorgan182]

I want to build a high speed number crunching machine. I have engineering programs that typically run on 2-3 cores but could use up to 5 cores simultaneously on a rare occasion. I'm interested in building a system with an i7-6700K, 16 gig of ram and 2 or 3 Samsung 950 Pro 512 GB PCIe NVMe operating under RAID 0 for maximum speed as the boot drive. Video is not critical, just speed. Any recommendations on motherboard and best method to incorporate m.2 SSD for optimum performance?

 

[kanewolf]

You are wasting resources with multiple PCIe storage cards on a number cruncher. You would do MUCH better to rework the code to use GPGPU processing and put something like a GTX1080 in one of those slots.

 

[bambiboom]

I want to build a high speed number crunching machine. I have engineering programs that typically run on 2-3 cores but could use up to 5 cores simultaneously on a rare occasion. I'm interested in building a system with an i7-6700K, 16 gig of ram and 2 or 3 Samsung 950 Pro 512 GB PCIe NVMe operating under RAID 0 for maximum speed as the boot drive. Video is not critical, just speed. Any recommendations on motherboard and best method to incorporate m.2 SSD for optimum performance?

dmorgan182,

I agree with Kanewolf that the priorities might be redirected. The disk in calculation intensive applications will be used for opening the programs, files, and savings. More important will be the swapping between CPU and RAM. It also important to maintain high double precision and I would strongly recommend a Xeon and ECC RAM.

As the applications have a limit of fivecores to effective multi-threading, I'd suggest using a 6-core with a high single-threaded performance. One of the best in that parameter is the Xeon E5-1660 v2 6-core @ 3.4 /4.0Ghz. On Passmark, the average CPU score is 13790 and the single-threaded rating is 2105. For comparison, the i7-6700K the average CPU score is 10999 and the single-threaded rating is 2329. The CPU score is weighted in a way that includes single-threaded as well as calculation cycles. The single-threaded the i7-6700K is superb, but the E5-1660 v2 is getting through a noticeably greater number of calculation cycles within the same period. The Xeon E5-1660 v2 uses DDR3-1866 ECC RAM which can have a latency of 10 or 11 whereas DDR4-2400 as used by the i7-6700K is 14.

The other aspect to consider is whether your applications can take advantage of GPU computing and it worth investigating that aspect carefully as the total computing power can be exponentially increased. I was visiting a research facility recently that runs particle experiment simulations using eleven parallel dual Xeon systems which each have four Tesla K20 GPU coprocessors. The "Titan" supercomputer at Oak Ridge for a while the fastest computer in the World uses 18,600 Opterons and 18,600 Tesla K20X's.

One tactic to your system which could fast and cost-effective would be to buy a used HP z620 with a low specification- about $700-800. Add the E5-1660 v2- about $750, 64GB of DDR3 1866 RAM, an Intel 750 PCIe SSD for the OS and Program, an LSI 9260 or 9361 RAID controller running a RAID 0 on a pair of Samsung 850 Pro SSD's that contain the active projects and libraries, and a RAID 1 on appropriate sized enterprise storage drives.

If your can benefit from GPU coprocessing, configure an NVIDIA Maximus using a Quadro and Tesla co-processor. The GPU therein have to match so depending on the budget, you might have a Fermi or Kepler set. A Quadro K4200 and Tesla K10 (Kepler, about $1,000) would be a power combination. If the application does not benefit from GPU coprocessing buy a used Tesla C2075 (Fermi, about $700) which is a GPU coprocessor with a single DVI monitor output.

If you would set a budget and determine the status of GPU co-processing in your use, a more specific recommendation is possible.

Cheers,

BambiBoom

Modeling:

1. HP z420 (2015) > Xeon E5-1660 v2 (6-core @ 3.7 / 4.0GHz) > 32GB DDR3 1866 ECC RAM > Quadro K4200 (4GB) > Intel 730 480GB (9SSDSC2BP480G4R5) > Western Digital Black WD1003FZEX 1TB> M-Audio 192 sound card > 600W PSU> > Windows 7 Professional 64-bit > Logitech z2300 speakers > 2X Dell Ultrasharp U2715H (2560 X 1440)>
[ Passmark Rating = 5064 > CPU= 13989 / 2D= 819 / 3D= 4596 / Mem= 2772 / Disk= 4555]
[Passmark V9.0 Beta Rating = 5019.1 > CPU= 14206 / 2D= 779 / 3D= 5032 / Mem= 2707 / Disk= 4760] 3.31.16
[Cinebench R15 > CPU = 1014 OpenGL= 126.59 FPS] 7.8.15

Simulation:

2. Dell Precision T5500 (2011) (Revised) > 2X Xeon X5680 (6-core @ 3.33 / 3.6GHz), 48GB DDR3 1333 ECC Reg. > Quadro K2200 (4GB ) > PERC H310 / Samsung 840 250GB / WD RE4 Enterprise 1TB > M-Audio 192 sound card > Logitech z313 > 875W PSU > Windows 7 Professional 64> HP 2711x (27", 1920 X 1080)
[ Passmark system rating = 3844 / CPU = 15047 / 2D= 662 / 3D= 3550 / Mem= 1785 / Disk= 2649] (12.30.15)

 

[dmorgan182]

I located an HP z420 online with an E5-1650 3.2/3.8 processor, 32 GB RAM, 1TB 7200 HD and a Quadro 2000 as a base system. I can add the SSD Raid 0 for the application and operating system and see if this helps. I can also change the processor and increase the RAM to test the impacts, but the base system is only $700 (refurbished) so the cost is minimal to see if the change in processor will significantly reduce run times. Thanks

 

[bambiboom]

I located an HP z420 online with an E5-1650 3.2/3.8 processor, 32 GB RAM, 1TB 7200 HD and a Quadro 2000 as a base system. I can add the SSD Raid 0 for the application and operating system and see if this helps. I can also change the processor and increase the RAM to test the impacts, but the base system is only $700 (refurbished) so the cost is minimal to see if the change in processor will significantly reduce run times. Thanks

dmorgan182,

The z420 / E5-1650 is a very good choice as a basis for this use, given the healthy clock speed and 6 cores that can accommodate the sometimes 5-cores utilization. I have two z420's and can say these are ultra-reliable and very quiet-running.

The CPU swaps to RAM may be more important than disk speed but a RAID 0 for the OS / programs SSD's can't hurt. It may be more advantageous though to use these in series -one drive with OS and programmes, and the second for active projects, although the read / write is so immediate on SSD's that may be moot as well. If you do have a RAID 0, keep a system image on a separate drive as rebuilding a RAID 0 using the hexidecimal file analysis looks tedious.

As you will have a Quadro 2000, which uses a NVIDIA Fermi GPU, you might consider whether a Fermi-based Quadro /Tesla "Maximus" coprocessor configuration can accelerate the processing. I've seen Tesla M2090's with 6GB memory and 512 CUDA cores-$2,600 new- selling for $100-120.

What kind of problems will the system be running and using which programs?

Cheers,

BambiBoom