alt Hacker NewsYou seem to think that "compute in memory" would mean the disappearance of commodity DRAM parts and their replacement with parts that had compute capabilities. This is impossible. Adding compute to memory will make it more expensive, so it will get zero adoption outside of the niches where it actually proves to be useful and valuable. Nobody is going to be putting compute in memory chips onto standard consumer DIMMs.
Replies
CPU's are silicon where some of the gates, some of the circuits that are etched into that silicon are turned into RAM.
That is, Modern CPU's (as opposed to truly ancient ones, around 50 or more years ago) implement caches -- regions of gates/circuits/transistors or mostly transistors -- which implement the equivalent of RAM.
When a silicon wafer specifically destined to become a stick of consumer RAM is etched, most of its space is etched in a gate/circuit/transistor (or mostly transistor!) pattern that implements RAM.
There may be other timing/coordination/refresh etc. circuits etched on the region of the silicon wafer which is to become the RAM, but those are usually a small portion of the space.
But, fundamentally CPU's and RAM -- are created from the same substrate, transistors and other very small electronic components etched into a silicon wafer, although these days due to manufacturing constraints and cost issues, RAM usually uses a larger process node, a larger (less expensive, older) size, i.e., the latest CPU might use 2nm process, and consumer RAM might use 10nm or 12nm process.
But, fundamentally, at the lowest level, both RAM and CPU are mostly transistors and other minature electronic components.
So, if they're both the same at that level (ignoring process node size), then my engineering question to you is as follows:
Why exactly can a CPU contain RAM electronic circuitry, but RAM can't contain CPU electronic circuitry, if both, at the lowest level, the level of the minaturized transistor, the minaturized electronic circuit, are created from the same substrate, the same basic building blocks?
>You seem to think that "compute in memory" would mean the disappearance of commodity DRAM parts and their replacement with parts that had compute capabilities.
You seem to imply that I think that, but I am merely speculating on possible futures, possible outcomes...
>This is impossible.
That might be speculation on your part...
>Nobody is going to be putting compute in memory chips onto standard consumer DIMMs.
That also might be speculation on your part...
Seymour Cray was heralded as a genius back in time in computer history, because he put memory physically closer to the CPU. Since his time, compute and memory have been getting closer and closer together, in various forms, including, but not limited to CPU caches, or say, 64GB HBM on Intel Xeon Max 9480... Cerebras, for example, mixes as much memory and compute as they can, on the same wafer...
But historically, memory and compute have been getting closer and closer together... that's because higher performance (which is the goal, the continual holy grail of the computing industry) typically requires less space, less latency, less delay, between the two...
Is putting a CPU or other processing unit/computing device/computing circuitry into a stick of DRAM a good idea?
Well, that I don't know, but the industry, always in search of better performance, might head that way, in whole or in part, in the future... we don't as-of-yet-know...
Maybe it would make sense for a single type of a desktop/laptop, for some use cases.
I would be more expensive, but maybe the performance would be worth it.