Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> Common Characteristics of Memory > - Zero access latency

It's incorrect, and simply wrong.

 help



One of many errors. Another is saying that all memories support writing data. Apparently this author has never heard of ROM. Or CD and DVD drives. Or neglected to carefully read the ai assisted parts.

Yes I agree, ROM, CD, and DVD are read-only drives. But the manufacturer writes the data. My perspective comes from hardware design, where even for read-only drives, the data-writing decision is made during manufacturing.

Example: When using a precomputed lookup table for the AES S-Box, ROMs are designed to create memory banks for faster arithmetic operations. During the design phase, you must make careful trade-offs regarding how much space to allocate for storing precomputed values. From the start, you need to floorplan how to write these precomputed arrays to achieve realistic space allocation. For this reason, algorithms like Canright's Boolean network are more suitable for compact design and are more efficient at reducing physical memory area.

It was my mistake to write it as general-purpose, and my wording was confusing. Thank you for highlighting this. I will read the article again and improve the writing.


You are right, I wrote in an oversight that "Zero access latency" is practically possible within some margin. I was reading the book "The Cache Memory by Jim Handy" and on section 2.1.3 which says:

"Occasionally, a designer will allow the CPU address to propagate through the address buffers to the bus at the beginning of every cycle, whether the cycle turns out to be a read hit or a read miss. The cache is designed in such a way as to initiate a main memory access at the same time as the cache access, rather than to wait for a cache miss before commencing the main memory access. This approach can shorten the main memory access time in the event of a cache miss and will serve to improve the operation of singleprocessor, single-tasking systems, but can work to the detriment of multitasking systems and multiprocessing systems, since the cache will allow the CPU to waste considerable amounts of main memory bus bandwidth. Such caches can be referred to as look aside designs. Since the processor has the option of going to either the cache or main memory at all times to request data, the look aside cache can be an add-on to the computer system."

Because the book mainly focused on the older i386 processor memory design, I could have been clearer when describing common characteristics. I updated the common characteristics to include "Higher data-rate processing," which is relevant to the Prefetch architecture and its intent to support more data bits.


That's where I stopped reading. It's just so egregiously wrong I didn't have faith in the rest of the article being useful.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: