This question has an answer in English! Translate Steve's answer.
Sort by
Profile photo of Shunichi Arai
Shunichi Arai
-Steve
Follow me
Software entrepreneur with a background as a programmer. Last update: 2 years ago
I would love to put RAM on CPUs, but so far, CPUs only have megabytes of cache memory due to technology limitations.
As Manabu Yoshida says, CPU and DRAM are incompatible and it is difficult to put them on the same die. eDRAM technology has been adopted by IBM, but it seems to have some disadvantages and it is not yet possible to put a large amount of RAM on it.
What is attracting attention now is a stacking technology that creates the memory on a separate silicon die from the CPU and mounts it in the same semiconductor package.
If this technology is realized, CPUs will be able to access memory by leaps and bounds, and computer performance will be greatly improved.
PS: Fujitsu's latest CPU A64FX seems to have 32GB of HBM2 memory inside the CPU package. It is probably the first CPU with HBM2 memory in the package, and it is expected to be highly usable as it is the first time in a long time that a calculator without accelerator such as GPU is on the Top500. [1] [2]
Footnotes.
[1] http://www.jpubb.com/press/2236430/
[2] Prototype of supercomputer "Fugaku" ranked No.1 in the world by Green500 : Fujitsu
morioka kazuo
morioka kazuo
-morioka kazuo
Follow us on Twitter
Reconcile with the cat 1 year ago
CPUs are too fast, and RAM that can keep up with the CPUs is too expensive. So, RAM that is fast enough to keep up with the CPU is built into the CPU and used as cache memory. Of course, if the cache memory is increased to the size of the current main memory, a very fast CPU can be created, but it will be very expensive. 10 million yen per unit might still be a good deal ( Nikkori
So, I have to choose between memory that is so fast that it can keep up with the CPU but is expensive as hell, and memory that is so slow that it makes the CPU wait for me! (psst!) However, the current situation is that the memory is so large that it can keep up with the CPU, and these are combined to get the best performance and the cheapest price.
Profile photo of Masahiro Ishizuka
Oba Hideyuki's profile picture
Oba Hideyuki
-Oba Hideyuki's profile picture
Follow us on
Systems Engineer, Legal (Scrivener) (2010-Present)2 years ago
You can do it if you want to, right?
You can integrate the computing performance of a current 10,000 yen CPU and a few dozen MB of memory into a CPU you can buy for 100,000 yen, or you can pay the price of a luxury car for a CPU with memory large enough to run Windows 10 at a minimum, or the price of a large house for memory large enough to run comfortably.
Long ago computers used SRAM based on the same mechanism as CPUs that ran at the same speed as CPUs, but it was very expensive and had low integration, so they just loaded it with DRAM because it was slower but cheaper and had enough capacity compared to not having enough memory or the machine being very expensive.
The fact that it is external is one reason why it is slow, but even inside, memory that runs at the CPU's computing speed is very expensive. The memory (cache) currently inside the CPU is also divided into levels due to cost and speed issues, and the memory that is installed in a few MB is L2 or L3, which is much slower than the full speed of the CPU.
The balance between price and performance dictates that a slower method of RAM is selected and divided.
0 コメント:
コメントを投稿