Originally Posted by
mdocod
Memory can be configured on a DIMM with different numbers of memory chips of different densities and widths to achieve a particular capacity and rank configuration..
A unbufferd (64 bit) 4GB DDR3 DIMM could be constructed in any of the following ways:
4 X 1024MB (16bit) chips = 4GB single rank
8 X 512MB (8bit) chips = 4GB single rank
8 X 512MB (16bit) chips = 4GB dual rank
16 X 256MB (4bit) chips = 4GB single rank
16 X 256MB (8bit) chips = 4GB dual rank
16 X 256MB (16bit) chips = 4GB quad rank
32 X 128MB (4 bit) chips = 4GB dual rank
32 X 128MB (8bit) chips = 4GB quad rank
Quad rank (or any 32 chip configuration) is pretty well unheard of in desktop/consumer systems. Most DIMMs made for non-server systems are either single or dual rank with 4-16 chips.
To be honest its something that I hadn't payed much attention to until recently, when I was running a bunch of benchmarks and tests to compare the effects of CPU-NB and RAM overclocking on the AM3+ platform. I switched memory from our APU rig into my AM3+ machine because it was *supposedly* nicer memory in order to run the tests. I noticed an immediate drop in memory benchmarking results across the board on the ripjawX kit vs the cheap Crucial Sport kit. Upon investigation:
Code:
Code:
Handle 0x0030, DMI type 17, 34 bytes
Memory Device
Array Handle: 0x002C
Error Information Handle: Not Provided
Total Width: 64 bits
Data Width: 64 bits
Size: 4096 MB
Form Factor: DIMM
Set: None
Locator: Node0_Dimm1
Bank Locator: Node0_Bank0
Type: DDR3
Type Detail: Synchronous Unbuffered (Unregistered)
Speed: 800 MHz
Manufacturer: Undefined
Serial Number: 00000000
Asset Tag: Dimm1_AssetTag
Part Number: F3-2133C9-4GX
Rank: 1
Configured Clock Speed: 800 MHz
The only difference that made any sense is that the ripjaw kit is single rank, while the crucial sport kit is dual rank. Sure enough, the dual rank kit performs better on the APU and tests better on the AM3+ platform. Further investigation suggests that this is not abnormal, many 3rd parties recommend ensuring at least 2 ranks are installed per memory channel, especially in server environments where rank interleaving can improve real-world performance by up to ~10%.
I guess my problem with the whole issue is that these flashy brands are selling "high speed" single rank kits as "performance memory" when really, it's just the cheapest way to make memory that is error free at a high cycle speed. There will always be a crowd chasing irrelevant numbers:
ghz and
mhz are a never ending story of betrayal in computing. I feel a bit betrayed because I purposely picked out a "performance" kit for an A10-6800K in order to get the best possible performance from the iGPU, only to find out after the fact that my efforts were a waste.