- Vista Workshop: More RAM, More Speed
- High-End DDR3 Memory on the Hook
- DDR3-1333 Speed and Latency Shootout
- PC Memory: Just the Facts
- The New Arms Race: DDR3-1800 RAM
- Hardcore DDR2 RAM by Corsair, G.Skill, OCZ and Patriot
- Overclocking 9 Value-Priced DDR2-800 Kits
- Corsair's PC10000 Super-Ober-Uber-RAM Sets Sails
- Corsair's XMS2 Dominator: The World's Fastest DDR2?
- Live Memory Test: Overclock 'Em Till They Crash
Some Memory Basics
Source: Tom's Hardware – Keywords: ram, scaling, notebook
Syndication:
Some Memory Basics
Advertisements usually only talk about the capacity of memory installed on systems: 1 GB, 2 GB or sometimes even 4 GB, although anything at 4 GB and above requires either a 32-bit operating system capable of handling the memory addressing to take advantage the memory capacity, or a 64-bit operating system, which natively supports much more RAM. Any system running Windows XP or Windows Vista should at least have 1 GB of memory. On the one hand, this will prevent Windows from swapping too much work data onto the slow hard drive. On the other hand, RAM has become really inexpensive, making memory one of the most affordable components.
Memory is always added to a system on so-called modules. Desktop PCs utilize Dual Inline Memory Module (DIMMs), the “dual” means that there are pins on both sides of the connector. Memory chips can also be located on both sides of a module. The first DIMMs had 168 pins and were used for SDRAM memory. The second generation for double-data rate RAM (DDR) had 184 pins, while the current DDR2 and DDR3 DIMMs have 240 pins. Differently positioned notches prevent users from confusing DDR2 with DDR3.
Due to space constraints, notebooks and laptops utilize different memory modules, which are called Small Outline DIMMs (SO-DIMMs). These are roughly 1/3 of the size of regular DIMMs, and have only 200 pins. Most notebooks offer two sockets for SO-DIMMs, although ultra-portables oftentimes come with permanently installed memory and only one slot for upgrades.
It’s important to know that the best performance is achieved by pairing two physical memory banks, which are referred to as channels. All memory controllers, whether they’re integrated into an AMD Athlon 64 X2 or Phenom X3/X4 processor or part of a chipset northbridge, are capable of running dual-channel mode, which means that they utilize two banks in order to double the bandwidth by widening the memory data path from 64 bits to 128 bits.
Good RAM, Bad RAM
While memory is categorized by memory type (DDR2, DDR3) and memory speed (clock speed), there are also various parameters that need to be set, called timings, and they’re also relevant to performance. However, the performance difference between mainstream memory and high-end RAM has been rather small. It is highly important to use a reasonable memory technology, which currently is DDR2-667/800 for notebooks and DDR2-800/1066 for desktop PCs. Running fast memory timings is favorable, but we recommend against spending a lot of money on enthusiast memory unless you’ve already optimized other key components such as the processor or the graphics card. In the case of notebook memory it’s usually not possible to alter memory timings anyway.
Mix or Match?
Memory vendors try to sell so-called matched memory pairs, which are often tested together in a standardized test environment. While two different DIMMs with identical specification should technically work together, issues may occur. Huge compatibility problems, resulting in functional incompatibility, are very rare these days, but it’s fairly possible that RAM speed or timings may have to be adjusted to slower settings for compatibility reasons.
You can always operate any memory type at slower clock speeds and more conservative timings than they were specified for. Running a DIMM above the specification, though, puts you into the overclocking corner, and must be done at your own risk. If you mix a 1 GB DDR2-667 and a 1-GB DDR2-800 DIMM, both will run at DDR2-667 speed.
- Previous page Buy RAM to Increase Your Battery Runtime
- Next page Synchronous vs. asynchronous Dual...
I'll stick with XP and my 2GB of Corsair until a version of Linux that doesn't require you to wander in to the console every five minutes comes along.
...one nice thing though is that the reviews consider reasonably priced equipment. I don't care about 8800 Ultras, 9800GX2s, or thousand dollar CPUs. I've got more important priorities then trying to impress people who aren't getting any work done.
The fastest SSD I've seen has read/write speeds of approximately 110MBps/80MBps and an access latency of about 0.1ms.
If significantly faster SSDs are currently for sale, please forgive me for not keeping up with storage tech the way I used to. (BTW, FusionIO is neither an SSD nor affordable outside of datacenters)
To simplify things we'll just use the faster of the two.
110MBps = 880 Mbps
This is roughly the effective throughput of single channel DDR-400 on an AthlonXP chipset. (Memtest86(+))
It is a little more than half the effective throughput of the single channel DDR-400 controller on the socket 754 Athlon 64 systems.
It is less than a quarter of the effective throughput of the dual channel DDR2-800 controller on an Athlon64 X2.
I don't have readily available numbers on an Intel DDR3 system, but you can see where this is going.
(I could have put in Intel memory controller results and/or AthlonXP dual channel results, but these numbers fit better to get the general feel of things)
These results aren't bad actually.
SSDs may not have reached even the last generation of memory technology, but at least in throughput, they match RAM technology that was widely used to run Windows XP.
Side note: If you use theoretical bandwidth, then SSDs can't reach DDR-400 (3200Mbps) until the interface gets upgraded to 4GHz (~400MBps).
If the numbers seem funny to you, it's because I'm taking into account the 10/8 encoding scheme that SATA uses.
The real kicker is the access time.
0.1ms seems very fast in HDD terms, but on a 2GHz processor, this is about 200000 cycles of latency.
Compare that to:
150-165 cycles for the AthlonXPs
125-140 cycles for P4s
~60 cycles for Athlon64s
RAM is made to be accessed very often in relatively short bursts.
Until SSDs can get the latency down, they have no hope of replacing RAM.
If mass storage were as quick as RAM, you wouldn't dedicate part of it to be used as RAM.
Rather, you would manipulate the files directly from the mass storage device.
This would eliminate RAM all together, reducing overhead and potentially power consumption.
Unfortunately, no current mass storage device has the potential to do that.
I'm still waiting for MRAM to make its appearance.
It used to hold promise, but after waiting so long I lost interest and stopped keeping up with it.
Perhaps it can do what flash can not.
Feeding the OS endless amounts of ram is not a long term solution to an underlying performance issue. I can run Ubuntu linux with all the fancy compiz-fusion eye candy with just 512mb and it's smooth. Vista doesn't run smooth even with 1gb; it's a real dog. I understand why everyone is avoiding it like the plague, except of course those poor unfortunates who bought into the Direct X 10 marketing fallacy, and others who bought a new machine with it preinstalled (microsoft tax).