I have seen noticeable differences in reliability between some brands. For example I have had very high failure rates with Mushkin and AData. Very low rates on Kingston, almost zero on HyperX (also technically Kingston) and so far, zero failures on Samsung which has become my brand of choice for now.
Also avoid no name SSDs from ebay etc, many lack adequate DRAM caches, and have slow controllers. So they look quite quick on small transfers, but the performance drops off very quickly on large reads and writes.
It seems that a range of CPUs were fitted to the E21-521 and that some had fitted the A8-6410. This is an interesting article where it seem bus speeds went down after fitting a second module.
formatting link
Another article said that they really ought to be matching pairs, including manufacturer, in order to use any dual channel feature, assuming that is an option for Harry's laptop.
Having more memory allows mucking about with virtual machines, keeping gazillion tabs open in web browsers, and almost no use of your paging file - keeping things snappy.
On the other hand, 8GB is more than enough for most casual uses, and your machine might be bound by other factors like your 1.8GHz Quad Core CPU or hard drive speed. Your Radeon R4 Graphics memory (integrated in he A6-6310 CPU) is deducted out of your system memory?
I've seen it said that going dual channel (memory amount staying the same) is a gain of 10% over single. Whether that is worth it?
I am no expert but I don't think that's how Windows manages memory. If you put more RAM in then the same system will use more RAM, and work better. At least that's how Windows used to be before I lost interest in it.
Surely the kernel can only 'see' 4Gb of memory on a 32-bit machine so it can't hand more than a *total* of 4Gb to all the processes it's managing. That is unless there's some clever memory management hardware in addition to the 32-bit addressing.
For Windows, some drivers were found to be buggy when dealing with more than 4GB, so for desktop versions this was artificially limited to 4GB (there are hacks to switch off the 4GB limit) and only allowed on servers where better qualification of drivers was enforced.
An extra level of page tables.
Depending how they're tuned, server versions of windows allow either 2GB or 3GB of memory per process, or there's an API that allows memory hungry programs (e.g. SQL databases) to 'bank switch' memory to access more than 4GB in a single process.
formatting link
All in all, for many years, it's been easier to use a 64bit O/S.
But when you have more than one loaded all the time so you have an instant switch to it when you want to use it, its more complicated than that. And its much better to have virtual OSs than do full reboots to do something in another OS too and that takes lots of ram.
But those with a clue don?t boot much at all. I only do it very infrequently, I suspend instead of shutting down.
And it doesn?t really make any sense to do anything much with a clapped out old laptop like that.
Given the number of Raspberry Pi's sold, I'm sure one of the main distros is Raspberry Pi OS (previously known as Raspbian), even accounting for not all pis running it. And Raspberry Pi OS is 32 bit (yeah I know there's a 64 bit not quite alpha version out there).
But in the Intel/AMD world you are probably right. However 64bit is not always better, as I know to my cost after I upgraded to 64 bit on my modest Intel based desktop. It had worked pretty well, if modestly, with
32 bit linux for quite a time, but the memory use of 64 bit really hits performance - it swaps a lot. Can't be bothered with a memory upgrade, so am testing whether I can do most things on an 8Gb RPI4.
Previous sockets going back, had this problem too (reduced settings to ensure stability with all slots filled).
This is AMD.
Intel used two cycle Command, to hide this on some of their stuff. Virtually all of the hardware now runs two cycle Command, just because of the speeds involved. Two cycle command, leaves extra Tsu to clock edge.
If the motherboard had been wired with two independent channels, that's when adding the second RAM stick, makes no difference to settings. As the two sticks are completely independent of one another. In this case, the bus is shared. One bus, two sticks. And it stays this way, since the wiring plan is "sealed in copper". It's not magically dynamic.
CPU -------+----+ Not a dual channel motherboard. | | The OPs config. Load power only SODIMM1 SODIMM2 increases by autorefresh power level of second DIMM (~1W). Settings drop.
+------- CPU -------+ A dual channel product, the DIMMs | | have no effect on one another. SODIMM1 SODIMM2 Commissioning the second bus, increases power consumption by a watt or two just for bus termination power and pad power.
The commit charge is the total amount of virtual memory currently in use. In win 10 it tends to show as two numbers separated by a slash. So currently my system is showing 26.8G/36.7G. The second is basically the sum of the physical ram and any space available in page files(s) (I have
32GB RAM, and a bit over 4GB available in page files).
The first number is the amount of VM actually committed. So generally if that total currently committed is is less than the physical ram, then there is no need for the OS to keep anything likely to be needed in a hurry in page file space rather than actual ram. So low paging overhead.
Adding more ram may coax it to keep more of the allocated VM in ram, but it will be stuff that is not likely to be needed in a hurry - so you may not see much positive impact on performance.
Once you have reached the point where it can hold everything it needs in RAM, having more does not help much further.
(that's why adding some ram can have a big overall effect on machines that were a little bit short - since extra ram will prevent it needing to page as much in the first place, and the SSD will ensure that if it does, you don't get anything like the performance hit as previously)
HomeOwnersHub 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.