It's only an advantage if the two threads are requesting memory from the two banks (one thread per bank). I don't believe Windows is optimised for this and both threads will likely end up contending for the same memory. In practice, dual channel memory doesn't give anything like the theoretical 2x bandwidth improvement.
The biggest improvement will be seen where an integrated graphics chipset can be setup so that the processor uses one memory bank (channel) most of the time and the graphics uses the other.
This is the way I would design dual channel memory for high performance systems, but it's actually completely wrong in the case of PCs.
In PCs, all that "dual channel" does is double the width of the data bus from 64 to 128 bits. The two memory channels are run in lock step so there is still only ever *one* access to memory at any time.
IMO it can't be emphasised enough that software design has a much greater impact on system performance than hardware speed.
Anyhow, the human operator is so slow that a gigahertz chip can load, run, and unload a dozen threads between keystrokes or mouse gestures. A lot of the purported gain in speed in practice makes no difference in ordinary computer use. The CPU is idle or nearly idle about 98% of the time. Just open the Performance tab in task manager, and watch the graph as you type. ;-)
Until I got 8GiB of RAM, my dual 3.2GHz dual core (which Windows optimistically reports 8 cores due to Hyperthreading, LOL!) would max out such that a single keystroke or mouse gesture could resultin a delay of 30 seconds or more before the response was completed.
I'm sure you are correct and my increasing/fixing the page space had absolutely no effect at all. Could you tell me why the PC decided to run much faster though ?
MBQ
Nope meant it was set for 500mB of ram, had 512-756Mb of page space. If remember correctly MS recommend 2* installed ram. The settings were as per delivered, base config with 500mB of ram but an additional 500 mB was installed at the shop at the time of purchase. Shows the danger of your assuming. Well as an ex-mainframe systems programmer that moved on to UNIX admin esp performance and finished up a C and C++ programmer having written a few realtime programs, maybe, just maybe I have an inkling as to the use of page space.
Think your a bit behind the times on threading, some of us have been doing it for a while. Even the latest development software package from microsoft will use multiple threads for complie and linking - soon messed everyone around as now they need to put some proper dependencies in their projects.
Philip wrote in news: snipped-for-privacy@gmail.com:
I would have also thought anyone wanting a gap to exploit might consider EMUs and DMUs. Modern image modellers have some headway here, but even some of the most common types of first generation DMU seem to be a missed opportunity. Whilst the ex-Lima model might do, I have yet to hear anyone say how good it is, even "for it's age". The 108 is obviously a good model, the 110 wasn't bad "for it's age" but had a relatively limited area of operation, then there's iconic DMUs like the bubble cars and the Class 120. Even the single car parcels units would be a good starter to test the water - only one mould and I'm pretty sure they had a broadish area they operated in, at least the LMR ones. And I daresay you could include the 104 as a good all rounder too.
EMUs are probably a harder gap to fill due to mainly limited areas of operation and lack of catenary but now Peco have announced they plan to design and manufacture british outline catenary, it may possibly be the start of a rush to see who can release some of the more common EMU models eg 314 and family. Or given the apparent popularity of obscure prototype diesels, some of the classic oddities like the Oerlikon sets. I suspect quite a few people would take a good model of one of these for the novelty factor alone!
Cannot see much of a market to be honest, those who care seem to manage using the components you mention or those from the P4 track company. Even a reasonable representation of 3rd rail needs a fair bit of customising to the track layout beyond what could be expected from an affordable ready made system. Despite the efforts of EFE and its tube stock four rail must be an even smaller market even if there is the odd person modeling the pre 1950's Mersey Railway or the LNWR /LMS/LMR lines to Watford or Richmond.
Rivet counters would be able to say you had the wrong style for the area. Though through running was compatible different operators used different components .
If you are interested in the subject this chap has done an excellent site with a lot of information I have never seen elsewhere.
You know the answer. Like you said it was probably crippled by lack of page file space. I was challenging the assertion that dynamically expanding page files have any great effect, assuming that they are set to system managed and not crippled by a faulty custom setting limiting their maximum size.
By fixed size I was referring to it not being set to "system managed". Being set for 512-756MiB of page space is still a fixed upper limit. Every Windows PC I've used has been set, from new, to use a system managed page file. I assumed correctly in this case that someone has fiddled, without understanding the implications, and set a custom size with a fixed upper limit and didn't change it when more RAM was installed.
It's not exactly rocket science, and you don't need those qualifications to understand the issues.
You know the answer. Like you said it was probably crippled by lack of page file space. I was challenging the assertion that dynamically expanding page files have any great effect, assuming that they are set to system managed and not crippled by a faulty custom setting limiting their maximum size.
By fixed size I was referring to it not being set to "system managed". Being set for 512-756MiB of page space is still a fixed upper limit. Every Windows PC I've used has been set, from new, to use a system managed page file. I assumed correctly in this case that someone has fiddled, without understanding the implications, and set a custom size with a fixed upper limit and didn't change it when more RAM was installed.
It's not exactly rocket science, and you don't need those qualifications to understand the issues.
MBQ
Sorry I cant be bothered to explain it unless someone actually wants to know. Its to do with disk contention if the page space isnt fixed and pre-allocated.
PolyTech Forum website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.