CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > General Forums > Hardware

M.2 SSD as substitude for RAM...optimum core number

Register Blogs Community New Posts Updated Threads Search

Like Tree1Likes
  • 1 Post By flotus1

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   January 13, 2018, 20:20
Default M.2 SSD as substitude for RAM...optimum core number
  #1
New Member
 
Join Date: Nov 2017
Posts: 22
Rep Power: 8
fonograph is on a distinguished road
If I use fast SSD like 960 PRO as substitude for RAM,is it waste to buy cpu with many cores?

Due to the huge RAM price increase,there is no way for me to afford even close enough amount of RAM,two years back and I could afford 64gb for the money that I have,but now I can get only 16gb,and that amount is just not enough.

On a positive side SSD technology advances rapidly and this new breed of super fast M.2 and PCIe NVMe SSDs is just amazing,especialy the 960 series from Samsung,its still far cry from RAM when it comes to bandwidth,but its much better than old SATA SSDs,the price per gigabyte is about order of magnitude less.

Before purchasing PC components its common thing to study benchmarks and tests done by other people to choose such combination within constrains of budget that will offer best performance.Problem is that CFD is not a mainstream thing,most people don't even know what CFD is.As a result,there is far less people discussing and testing PC hardware from perspective of CFD user,my point is I cant find tests and benchmarks that I need.

I read a lot about CFD in last two months and there is couple of generalizations and tips that I remember.

1. CFD needs powerfull CPU,it can make use of very large number of cores.

2. CFD needs large amount of RAM,more RAM= bigger/more accurate simulations

3. CFD needs big RAM bandwidth,the more memory controllers and higher frequency RAM,the better.

4. 2 or 3 cores per memory controller is optimum.More cores will not decrease simulation time becose the bottleneck is in memory bandwidth.

Due to my limited budget,I am forced into mainstream platform ( AM4 ) with only two memory controllers,my RAM options are either new fast Samsung B die 2x8gb 3466 MHz or second hand bulk 2133 MHz RAM 2x16gb.

My CPU options are any of Ryzen line of processors from 100 euro 4 core,180 euro 6 core to 270e 8 core.I will buy 512gb 960 PRO SSD.


The question is this : If I run simulation that requires 100 gb of space,and I have only 16gb of RAM,does it even make sense to buy 8 or 6 core cpu over 4 core? If I compare the SSD bandwidth to RAM,then it would seem I would see no improvement beyond ONE core! Single 2133 MHz RAM running on single memory controller is 17gb/s bandwidth,maximum theoretical bandwidth of SSD that uses 4x PCIe 3.0 lanes is like 4gb a second,going by the tip of 2-3 cores per memory controller,there is no point in using more than one cpu core even if the bandwidth of my SSD quadrupled.

And what about RAM,if wast majority of simulation doesn't fit onto RAM and PC starts using SSD,does it even matter if I have 32 or 16gb? Does it even matter if the RAM is 3466MHz or 2133 MHz when the SSD is far slower than RAM?
fonograph is offline   Reply With Quote

Old   January 14, 2018, 05:26
Default
  #2
Super Moderator
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,399
Rep Power: 46
flotus1 has a spectacular aura aboutflotus1 has a spectacular aura about
SSDs, no matter how fast they are, can not replace system memory.
Even a full-fledged PCIe-SSD delivers a bandwidth that is an order of magnitude lower than RAM. Same with latencies.
And then there is the problem with durability. A flash SSD that is abused as a swap drive for HPC workloads will reach the end of its lifespan rather quickly.

If you have simulations that require 100GB of RAM and budget is that limited, your only option is a dual-socket 2011 platform. The boards are rather expensive, but processors and DDR3 reg ECC are cheap.
Spanner likes this.
flotus1 is offline   Reply With Quote

Reply


Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
SigFpe when running ANY application in parallel Pj. OpenFOAM Running, Solving & CFD 3 April 23, 2015 14:53
Compressor Simulation using rhoPimpleDyMFoam Jetfire OpenFOAM Running, Solving & CFD 107 December 9, 2014 13:38
DecomposePar unequal number of shared faces maka OpenFOAM Pre-Processing 6 August 12, 2010 09:01
[blockMesh] BlockMeshmergePatchPairs hjasak OpenFOAM Meshing & Mesh Conversion 11 August 15, 2008 07:36
Unaligned accesses on IA64 andre OpenFOAM 5 June 23, 2008 10:37


All times are GMT -4. The time now is 05:16.