My dad had a 256bit memory module on his desk. Made out of sandwich of two pieces of glass and 16 vertical and 12 horizontal wires. Each intersection had a small ferrous ring.
Around that time he was programming and early computer and you could actually hear the memory bits changing. He was excited listening to it, figuring each click was a loop. Then her heard a different sound and realized he as hearing instructions and not loops.
Some really early computers stored bits in waves in mercury. A series of waves would be started at one end of a tube full of mercury and they would be read back when they arrived at the other end. The memory of course then had to be refreshed. Core memory was however much more successful.
Is Optane really 'all that' compared to other PCIe storage?
I know that Intel are positioning it as some sort of system accelerator, and I know it does tend to score well in comparisons... but not "This is in a completely different category" well, AFAICT.
What's special about it compared to, for instance, Samsung's NVMe offerings?
I use LVM and have enough storage that all SSD would be a little expensive so for the last few years I've been using the block caching LVM feature with mirrored NVME drives as a writeback cache to my large HDD array. It greatly increases the performance of the storage with little hassle.
If haven't played with gluster for a few years but it has a tiered storage feature so the most recently used files actually get moved onto the high speed storage medium, which depending on your workloads might be even better, assuming gluster is now performent and stable enough.
I recently upgraded my laptop to a Thinkpad T470s, which supports M.2 NVMe, and M.2 Sata.
I upgraded from a 500gb NVMe drive to a 1TB M.2 sata drive.
Why do I say upgraded if it is slower?
Well, in a laptop, power usage is hugely important, as is heat.
To prevent throttling on the CPU, less heat is better.
I did a lot of research, and found that the sata based M.2 drives consume less power, and generate less heat. For my purposes the sata speed is sufficient. :)
From the brochure, it had a few million of 64-bit words worth of RAM that was used as regular RAM ("central memory") and a lot more of ECC DRAM that was used as fast storage ("SSD"). What was the API for the SSD? Was it mapped into virtual memory as RAM, or did the ISA have I/O in,out opcodes for it with word address, or did it use the API for disk storage, or what?
Register latency is usually around 4ns (2.5-3.5 Ghz with a pipeline depth of 12-16). The article claims main memory latency is 7x more, which is around 24ns to 45ns. I tried to come up with plausible numbers that fix the 7x ratio they mention.
These days main memory latencies are more around 70-80ns (single sockets) and 80-100 ns (dual sockets).
Being off by so much does make me wonder about the accuracy of the rest of the article.
language is symbolic so there doesn't have to be a literal connection between the reference and referent. for example you still "boot" a computer, which comes from the concept of "bootstrapping", ie, pulling oneself up by the straps of their boots. no boots are involved.
Contrary to what others said, it will likely fall out of favor the way the floppy icon faded for "Save" functionality in GUIs. It's everywhere until one day you realize you can't find it in any of the apps you use every day.
sliken|6 years ago
Around that time he was programming and early computer and you could actually hear the memory bits changing. He was excited listening to it, figuring each click was a loop. Then her heard a different sound and realized he as hearing instructions and not loops.
jcoffland|6 years ago
https://en.m.wikipedia.org/wiki/Delay_line_memory
agumonkey|6 years ago
Avery3R|6 years ago
spectramax|6 years ago
Nursie|6 years ago
I know that Intel are positioning it as some sort of system accelerator, and I know it does tend to score well in comparisons... but not "This is in a completely different category" well, AFAICT.
What's special about it compared to, for instance, Samsung's NVMe offerings?
chapium|6 years ago
edoo|6 years ago
If haven't played with gluster for a few years but it has a tiered storage feature so the most recently used files actually get moved onto the high speed storage medium, which depending on your workloads might be even better, assuming gluster is now performent and stable enough.
ntw1103|6 years ago
brianpaul|6 years ago
wolf550e|6 years ago
theideaofcoffee|6 years ago
I don’t have a hunk of the SSD, but I have a (30lb) board of the central memory perched against my desk right now.
sliken|6 years ago
These days main memory latencies are more around 70-80ns (single sockets) and 80-100 ns (dual sockets).
Being off by so much does make me wonder about the accuracy of the rest of the article.
3xblah|6 years ago
tyingq|6 years ago
bluedino|6 years ago
modzu|6 years ago
NikolaeVarius|6 years ago
I think colloquially, it's pretty much understood to mean "time between turning on and being useful"
niccl|6 years ago
jyrkesh|6 years ago