Support Board
Date/Time: Tue, 04 Feb 2025 14:32:58 +0000
Post From: CPU usage for multi-core processor
| [2013-11-27 07:36:43] |
| namlem - Posts: 94 |
|
FFTrader, if I may share my point of view.. 2. If I load SC and its data directory in a RAMDisk, is it advisable for the purpose of gaining some speed?
since I have heavy calculations in Spreadsheets, putting whole SC into a RAMDisk didn't help me. Loading of charts was a bit faster and this is it (cpu i7-2600K, 16gb dd3).Overall, good performance gain I received by installing SC CLR version (non.NET version loads charts faster) on SDD drive and yes, buying faster CPU obviously would help me also. Actually, instead of buying new CPU every year :) I was thinking about use of distributed computing, but hit few obstacles on a way - first, an application must be programmed in a way to support cpu process distribution and 2nd - I didn't find an application for this task for WinOS (apple OS and linux have it). So, as you may know, Amazon suggests its cloud computing services, so I tried their windows cluster c3.8xlarge (with 32 virtual CPUs, 108 ECU, 60GB RAM) and you know - SC load time was even slower than on my PC... Having this in mind, as far as I understood, process can be shared efficiently if app developed in a way to support it and runs on linux. I think it's much easier to implement cuda support on linux for SC. Linux support is on a way, only one thing left :) In the mean time, I'm concentrated on spreadsheets' logic tuning. In my case, it was taking more than 3 hours to start SC each time(.net version). That's why, for example, I suggested SC team, to create a feature to load already calculated spreadsheets (afaik, it will be supported in non.CLR version). Overall, from my experience, best performance gain obtained by tuning application itself, in our case spreadsheet. |