RAM 或随机存取存储器(Random Access Memory )是任何现代计算机中极其重要的部分。计算机的CPU（中央处理单元）需要数据和指令才能执行工作。该信息必须存储在某个地方。“某处”被称为计算机内存。
有各种类型的RAM内存，每种都有自己的优缺点。CPU(CPUs)内置了非常少量的内存，称为CPU“缓存”。这种内存速度非常快，本质上是CPU本身的一部分。但是，它非常昂贵，因此不能用作计算机的主存储器。
这就是RAM发挥作用的地方。RAM以硅计算机芯片的形式出现，连接到内存总线。CPU本身的缓存实际上也是RAM的一种形式，但是当这个术语被普遍使用时，它是指这些位于CPU之外的内存芯片。
内存总线只是一组专用电路，用于在CPU和RAM本身之间传输信息。操作系统从系统的慢得多的机械或固态硬盘驱动器(solid-state hard drive)中移动信息，为 CPU 的需求做准备。例如，当视频游戏“加载”时，数据正在从硬盘驱动器移动到RAM。
打个比方，将RAM视为桌面，将抽屉视为硬盘，而您自己则充当CPU。处理桌面上的物品既快速又容易，但空间有限。这意味着您需要根据需要在桌面和抽屉之间移动物品。
当今使用的计算机、智能手机、游戏机和所有其他类型的计算设备都具有某种类型的 RAM(some type of RAM)。我们将逐一介绍，解释它的工作原理和用途。具体来说(Specifically)，我们将介绍以下类型的 RAM：
- SRAM
- 动态随机存取存储器
- 动态随机存取存储器
- 特别提款权内存
- DDR SDRAM
- GDDR
- HMB
如果这听起来像是令人生畏的胡言乱语，请不要担心。这一切很快就会变得非常清楚。
SRAM – 静态随机存取存储器(SRAM – Static Random Access Memory)
RAM的两种主要类型之一，SRAM是特殊的，因为它不需要“刷新”来保留它当前存储的信息。只要有电流通过电路，信息就会保持在原处。
SRAM 由许多晶体管 (4-6) 构成，并且由于其性质而速度极快。然而，它相对复杂且昂贵，这就是为什么您会在作为超高速缓存存储器投入使用 的CPU中找到它的原因。(CPUs)
在数据必须快速移动但可能会遇到瓶颈的地方，也有少量的SRAM缓存。(SRAM)硬盘(Hard)缓冲区就是这个用例的一个很好的例子。无论设备在哪里需要更多数据，都有可能会有一些SRAM帮助平滑传输。
DRAM – 动态随机存取存储器(DRAM – Dynamic Random Access Memory)
DRAM 是另(other )一种常见的RAM设计类型。DRAM内存是使用晶体管和电容器构建的。除非您刷新每个存储单元，否则它将丢失其内容。这就是为什么它被称为“动态”而不是“静态”的原因。
DRAM比SRAM慢得多，但仍比硬盘驱动器等辅助存储设备快得多。它也比SRAM便宜得多，并且计算机通常具有数 GB 的板载DRAM作为主要(DRAM)RAM解决方案。
SDRAM – 同步动态随机存取存储器(SDRAM – Synchronous Dynamic Random Access Memory)
有些人似乎认为SDRAM是SRAM和DRAM的混合体，但事实并非如此！这是已与CPU时钟 同步的DRAM 。
DRAM模块将等待CPU响应数据输入请求。由于其同步特性以及SDRAM内存如何配置到 bank 中，CPU可以同时完成多条指令，从而显着提高其整体性能。
SDRAM是当今大多数计算机中使用的主要(SDRAM)RAM类型的基本形式。它也被称为SDR SDRAM或单数据速率同步动态随机存取存储器(Single Data Rate Synchronous Dynamic Random Access Memory)。尽管它与当今计算机中使用的内存类型基本相同，但它的普通SDR形式几乎已经过时，取而代之的是我们列表中的下一种类型的RAM。
双倍数据速率同步动态随机存取存储器(Double Data Rate Synchronous Dynamic Random Access Memory)
您应该知道的第一件事是有多代DDR内存。第一代，我们回想起来称为DDR 1 ，通过让读写操作在时钟周期的高峰和低谷同时发生，使SDRAM的速度提高了一倍。
DDR2、DDR3和今天的DDR4在第一代DDR的基础上得到了指数级的改进。这些内存模块的性能以每秒兆传输数(Mega Transfers per Second )或“MT/S”来衡量。一次兆传输基本上相当于一百万个时钟周期。最快的第一代DDR芯片可以执行 400 MT/s。DDR4可以达到 3200 MT/s！
GDDR SDRAM – 图形双倍数据速率随机存取存储器 (GDDR SDRAM – Graphics Double Data Rate Random Access Memory )
GDDR目前处于第六代，几乎完全被发现连接到视频卡或游戏机上的GPU （图形处理单元） 。GDDR与常规DDR相关，但专为图形用例而设计。强调大量带宽，同时不太关心低延迟。
换句话说，这种内存的响应速度不如普通SDRAM快，但它可以在响应时立即移动更多信息。这对于需要流入许多 GB 纹理数据来渲染场景的图形应用程序来说是完美的，并且少量的延迟并没有真正的后果。
尽管名称如此，但GDDR可以用作普通系统RAM。例如，PlayStation 4有一个GDDR内存池，开发人员可以按照他们喜欢的方式拆分，根据需要将部分分配给CPU和GPU。
HBM – 高带宽内存(HBM – High Bandwidth Memory)
GDDR有一个HBM 内存形式的竞争对手，它在(HBM memory)AMD制造的数量有限的显卡上得到了体现。目前(Currently)最新版本是HBM 2，但不确定它是否会取代GDDR或失效。
内存性能最重要的部分是在给定时间内可以移动的数据总量。做到这一点的一种方法是使内存非常快。提高总带宽的另一种方法是使“管道”数据被推得更宽。
HBM内存以低于GDDR的原始时钟频率运行，但使用独特的 3D 堆叠芯片设计，为数据提供了非常宽的物理路径以及更短的信号传输距离。最终结果是内存解决方案与GDDR(GDDR)相比具有相似的总带宽，但延迟更少。
HBM的问题在于制造起来很复杂，而且由于它的物理设计，它还无法实现GDDR微不足道的各种能力。如果最终解决了这些问题，它可能会取代GDDR，但不能保证会发生这种情况。
Thanks For The Memories!
很明显，RAM是任何计算机的重要组成部分，当它出现问题时，很难弄清楚问题到底出在哪里。
毕竟，这里或那里的流氓位可能会使您的系统微妙地不稳定或看似随机崩溃。这就是为什么当您遇到莫名其妙的稳定性问题时， 您应该始终测试坏的 RAM 内存。(test for bad RAM memory)
有一天，我们可能会超越RAM，但在可预见的未来，它将成为计算性能难题的重要组成部分，因此我们不妨了解一下。
Understanding Types Of RAM Memory & How It’s Used
RAM or Random Access Memory is an incredibly important part of any modern computer. The CPU (central processing unit) of a computer needs data and instructions in order to perform work. That information has to be stored somewhere. The “somewhere” is referred to as computer memory.
There are various types of RAM memory, each with their own pros and cons. CPUs have a very small amount of memory built into them, known as the CPU “cache”. This memory is incredibly fast and essentially part of the CPU itself. However, it is very expensive and so can’t be used as the primary memory of the computer.
That’s where RAM comes into play. RAM comes in the form of silicon computer chips, attached to a memory bus. The cache memory on the CPU itself is actually also a form of RAM, but when the term is generally used, it refers to these memory chips that sit outside of the CPU.
A memory bus is simply a dedicated set of circuits that move information between the CPU and the RAM itself. The operating system moves information from the much slower mechanical or solid-state hard drive of the system, in preparation for the CPU’s needs. For example, when a video game is “loading”, data is being moved from the hard drive to RAM.
As an analogy, think of RAM as the top of a desk and the drawers as the hard drive, with you yourself acting as the CPU. It’s fast and easy to work with items that are on the desk, but there’s only so much room. Which means you need to move things between the desk surface and the drawers as you need them.
Computers, smartphones, game consoles and every other type of computing device in use today has some type of RAM. We’ll be going over each one, explaining how it works and what it’s used for. Specifically we’ll be covering the following types of RAM:
- SRAM
- DRAM
- SDRAM
- SDR RAM
- DDR SDRAM
- GDDR
- HMB
Don’t worry if that sounds like intimidating gibberish. It will all become very clear shortly.
SRAM – Static Random Access Memory
One of the two primary types of RAM, SRAM is special because it doesn’t need to be “refreshed” to retain the information it’s currently storing. As long as there’s power flowing through the circuits, the information stays right where it is.
SRAM is built from a number of transistors (4-6) and is incredibly fast thanks to its nature. It is however relatively complex and expensive, which is why you’ll find it in CPUs put into service as hyper-fast cache memory.
There are also small amounts of SRAM cache wherever data has to move quickly, but might be bottlenecked. Hard drive buffers are a good example of this use case. Wherever a device has to more data around, chances are there will be some SRAM helping smooth that transfer out.
DRAM – Dynamic Random Access Memory
DRAM is the other common type of RAM design. DRAM memory is built using transistors and capacitors. Unless you refresh each memory cell, it will lose its contents. This is why it’s called “dynamic” rather than ”static”.
DRAM is much slower than SRAM, but still much faster than secondary storage devices like hard drives. It’s also far cheaper than SRAM and it’s typical for computers to have multiple gigabytes of DRAM onboard as the main RAM solution.
SDRAM – Synchronous Dynamic Random Access Memory
Some people seem to think that SDRAM is a mix of SRAM and DRAM, but it’s not! This is DRAM that has been synced to the CPU clock.
The DRAM module will wait for the CPU before responding to data input requests. Thanks to its synchronous nature and how SDRAM memory is configured into banks, the CPU can complete multiple instructions at the same time, significantly increasing its overall performance.
SDRAM is the basic form of the main RAM type used in most computers today. It’s also known as SDR SDRAM or Single Data Rate Synchronous Dynamic Random Access Memory. Although it’s fundamentally the same type of memory used in computers today, the vanilla SDR form of it is pretty much obsolete, replaced by the next type of RAM on our list.
Double Data Rate Synchronous Dynamic Random Access Memory
The first thing you should know is that there are multiple generations of DDR memory. The first generation, which we refer to as DDR 1 in retrospect, doubled the speed of SDRAM by letting read and write operations happen at both the peak and trough of the clock cycle.
DDR2, DDR3 and today DDR4 have exponentially improved on that first generation of DDR. The performance of these memory modules are measured in Mega Transfers per Second or “MT/S”. One mega transfer is essentially the equivalent of a million clock cycles. The fastest first generation DDR chips could perform 400 MT/s. DDR4 can be as fast as 3200MT/s!
GDDR SDRAM – Graphics Double Data Rate Random Access Memory
GDDR is currently sitting at the sixth generation and is almost exclusively found connected to a GPU (graphics processing unit) on a video card or games console. GDDR is related to regular DDR, but is designed for graphics use cases. Emphasizing massive amounts of bandwidth, while being less concerned with low-latency.
In other words, this memory does not respond as quickly as regular SDRAM, but it can move more information at once when it does respond. That’s perfect for graphics applications where many gigabytes of texture data needs to be streamed in to render a scene, and the small amount of latency is of no real consequence.
Despite the name, GDDR can be used as normal system RAM. For example, the PlayStation 4 has a single pool of GDDR memory that developers can split any way they like, allocating portions to the CPU and GPU as needed.
HBM – High Bandwidth Memory
GDDR has a competitor in the form of HBM memory, which has featured on a limited number of graphics cards made by AMD. Currently the latest version is HBM 2, but it is uncertain whether it will supplant GDDR or become defunct.
The most important part of memory performance is the total amount of data that can be shifted within a given amount of time. One way to do this is to make memory that is very fast. The other way to improve the total bandwidth is to make the “pipe” data is being pushed through wider.
HBM memory runs at lower raw clock frequencies than GDDR, but uses a unique 3D-stacked chip design that provides a very wide physical pathway for data as well as much shorter distances for signals to travel. The end result is a memory solution that has similar total bandwidth compared to GDDR, but with less latency.
The problem with HBM is that it’s complicated to make and thanks to its physical design it’s not yet possible to achieve the sorts of capacities that are trivial with GDDR. If those problems are eventually overcome, it could replace GDDR, but there’s no guarantee that this will happen.
Thanks For The Memories!
It should be obvious that RAM is an essential component of any computer and, when it goes wrong, it can be hard to figure out what the problem actually is.
After all, a rogue bit here or there might make your system subtly unstable or be behind seemingly random crashes. This is why you should always test for bad RAM memory whenever you have an inexplicable stability problem.
One day we might move beyond RAM, but for the foreseeable future it will be an essential part of the computing performance puzzle, so we might as well get to know it.
