DDR4 offers no advantages over DDR3.
DDR4 offers no advantages over DDR3.
Hi, So I was doing some research on memory layouts for a graduate project and was reading some papers on it. I came across this interesting research article from Carnegie Mellon University published and accepted in a reputable and prestigious computer architecture journal. If somebody wants to have a look at it, this is the link to the article : https://dl.acm.org/doi/pdf/10.1145/3366708 I think this is a more digestible presentation (ppt) version of the paper : https://people.inf.ethz.ch/omutlu/pub/Wo...9-talk.pdf The gist : DDR4 is not really much better than DDR3 in some of the workloads. In fact in certain benchmarks, it is quite regressive when compared to DDR3. The authors compared it with 9 popular memory technologies such as GDDRx, LPDDRx etc. and concluded that the LPDDR scores slightly better due to its low power usage and embedded nature. When Linus keeps telling that DDR4 improves the performance, I think we really need to put this to test or atleast have a discussion or a disclaimer as to if this is credible? Why? Because memory doesnt scale really well in terms of manufacturing like CPU. CPU is classified as logic and manufacturing a CPU or logic block is vastly different from manufacturing a memory block. Because transistors leak charge all the time, and when you try to cram as much capacity inside a single DIMM as possible, the latency actually increases since now the charge needs to travel longer distance to get sensed as 1 or 0. But hey Akhil, isnt it that when transistor size decreases, the performance increases? It is true for a logic block or CPU, but for memory, it doesnt hold true because as you decrease the transistor size, the charge literally leaks much easier now. Think of transistor like a tap and capacitor holding your 1 or 0 bit as a water tank. Now when the tap is made smaller, the tank puts so much pressure on the tap that the water keeps dripping drop by drop. This is what happens when you try to decrease transistor size in DDR memories, the charge keeps leaking. Memory companies need to come up with various ways to arrest this leakage and cut corners else where. (In electronics, the famous saying goes: you can't have free lunch). Hence, thats why you dont have 7nm memories even thought CPUs are on 7nm. So essentially having DDR4 doesnt really improve your speeds by that much, or atleast it is really debatable.
Someone seems to be confusing DDR4 speed with the actual DIMM specifications. It’s interesting how people often mix up these terms. Did you hear about this discussion? Was it a fresh insight or just another observation?
The analysis made here is incorrect. It’s essential to revisit the material thoroughly, paying close attention to how speed is defined—whether it refers to bandwidth or latency. Mixing these concepts into a single vague interpretation undermines the actual discussion about performance.
I recall hearing someone mention the differences between DDR3 and DDR4 latency a few years back. It seems like the topic wasn’t new to anyone familiar with the subject.
He mentioned in a recent interview on the channel discussing his video that the concepts are quite similar when moving from DDR3 to DDR2. It involves a transition with higher latency but offers faster performance and more storage capacity.