33 ms·
Here's a thought: What if, instead, Intel began keeping a single socket revision alive for longer than 1 release cycle of CPU? This seems like the fundamental
by jaytaylor 7y ago
Here's a thought:
What if, instead, Intel began keeping a single socket revision alive for longer than 1 release cycle of CPU?
This seems like the fundamental issue. I remember ~20 years ago when there were only 2 sockets for x86 to worry about: Intel P4 & AMD Athlon. Motherboards lasted at least a generation or two, it was nice.
- TylerE 7y agoI wonder if they consider that a feature, and not a bug. Don't want people slagging off your cool new CPU when it's actually their old mobo bottlenecking things.
- dbcurtis 7y agoMemory bandwidth. You can’t make the code run faster without more ability to move data to and from mamory. All questions about modern CPU architecture get down to two questions of first principle: How much memory bandwidth is available? How is the available die area being used?
- BubRoss 7y agoThis is not true at all. I think you might mean memory latency. Modern CPUs are primarily designed around memory latency. Very few programs are limited by memory bandwidth. At that point you have done a good job of optimizing already.
- wmf 7y agoMost people don't upgrade frequently enough for it to matter; letting people put a new CPU into a 4-5 year old motherboard may be more of a disservice. If you do upgrade frequently then you just need to consider the cost of the motherboard as part of your upgrade cost.
- pjmlp 7y agoIt never mattered to me, when it came to upgrade I always had to upgrade the motherboard as well, thus ending up buying a full PC, as in Portugal there was hardly any monetary benefit building yourself the PC vs buying a new one.