why no tri-channel ram?
why no tri-channel ram?
I discovered that DDR3 supports three RAM channels, which led me to question why tri-channel memory was removed from DDR4/5. It would have increased bandwidth without requiring stricter timing or higher clock speeds. The reason behind this decision isn’t just about cost or difficulty—it might involve other factors I’m not aware of.
More channels increase silicon intricacy and result in bigger sockets with additional pins, leading to pricier motherboards. A notable issue with memory channels is that higher counts raise memory latency. Consumer systems aim to balance affordability and speed, so most don’t push dual-channel DDR4 for bandwidth gains. Triple channel isn’t significantly more complex than triple, and beyond dual-channel bandwidth needs, you typically require even more channels. It’s an odd compromise that doesn’t offer clear advantages.
This involves product grouping, though diminishing returns also play a role. While three or four channels can be quicker for certain tasks, the improvement isn’t consistent across all workloads. Some applications do scale well with core count, similar to rendering and compression tasks, but most common uses—like gaming, browsing, office work—reach their peak efficiency around six to eight cores. Memory bandwidth behaves in a comparable way; it’s pricier to build, as triple or quad channels demand more space than dual channels, yet most users won’t notice much gain. Still, some buyers are willing to invest in extra channels, which is why high-end platforms like Threadripper and Core X exist. Both AMD and Intel have been less active in this area for about three years, but rumors suggest they may re-enter soon, possibly bringing more than two channels to consumer and prosumer devices. For now, achieving four or more channels typically requires EYPC or Xeon server-grade components.
the only system they relied on was the first-gen core HEDT. Now HEDT (thread ripper) can reach quad levels, but adding channels makes things tricky. It complicates the signals and needs more board layers, plus adjusting timing to keep signals stable. The price is better covered by faster RAM in consumer devices since regular chips can't really go beyond 6200mt/s for meaningful gains