Page not found – ShopingServer Wiki

What is NVMe memory and what are its features?

msrxp — Sat, 18 Jan 2020 17:40:12 +0000

The high speed of NVMe SSDs has made them very popular, but what is the difference between them and SATA SSDs?

NVMe stands for non-volatile memory express is a communication interface for SSDs developed by a consortium of companies such as Intel, Samsung, SanDisk, Dell and Sigit; this protocol actually uses bus Enables PCIe for SSDs. NVM Express enables host software and hardware to utilize maximum parallelism in modern SSDs.

NVMe has long been considered an unnecessary but widely used option in the information storage industry. But now that capability has become an important item, and buyers of new computers are better off paying attention to it. If you have just purchased a computer and want to speed it up, you can achieve your goal by using this technology.

Using the NVMe protocol improves the read / write speed of I / O; it also reduces the latency dramatically. The improved process of checking multiple rows of commands (especially long commands) also greatly improves memory speed and performance in this type of technology. NVMe is an alternative to SCSI and ATA, with the goal of maximizing memory capacity.

NAND, their controls, and their structure allowed SSDs to reach high speeds, but they remained unused until there was a process and port for high speeds. NVMe, developed based on PCIe, has actually removed the barriers to data transfer. In fact, due to the use of old interfaces, many SSDs cannot reach their maximum capacity for data exchange speeds, and the new standard can further increase the gap between these types of memory and old mechanical memory.

Brief History

The first details and news about the new standard of memory became Media Year 2. The technical development of the NVMe interface began in year 2, with more than five companies led by Intel’s Huber Huffman advancing the development process. March 8, version 2.3 was released. Version 2.3 added SSD support with multiple PCIe ports to the previous version.

The Version 2.3 on November 6 added improvements such as live support for more streamlined updates and better usage. Version 2.3, which we are currently seeing in memory, was released on June 6, which brings clean-up of the lower level blocks on NAND and new formats for NVMe memory.

Simulation improvements also defined how to use shared storage when there was a physical and virtual controller. Initial information from version 2.3 was also released on June 6, but no product has yet been released.

How does NVMe work?

NVMe maps the I / O commands and their responses to shared memory on the PCIe interface on the host computer. This interface supports parallel I / O and multi-core processors to achieve high throughput and reduce bottlenecks.

NVMe acts as the host computer writes an I / O command line and executes the NVMe controller by removing the I / O rows and sends the executed commands back to the host.

By allocating more paths to process an I / O request than SCSI and ATA, the NVMe needs to be reduced to the CPU. NVMe supports 7000 commands in a single message row and 2 I / O rows. For comparison, it should be noted that the SAS-based device supports a maximum of 2 commands and one SATA memory at most 1 command per row.

The main problem is memory!

The advances of processor and graphics card manufacturers in the last decade cannot be denied, but the main reason for the rise of laptops in current generations is SSDs . For many years, storage slowdowns have been one of the main causes of PC slowdowns, and increasing the number of processor cores or frequency of the processor has not helped make systems faster. Mechanical memories were wasting the potential of processors and graphics cards, and their speed of rotation was not effective either. Eventually SSDs were able to fill the gap created between CPU power and memory speed. Now, with all the powerful systems including computers and laptops for gaming , video editing and graphic content production, we’re seeing a new kind of memory.

For example,

if you’ve bought a MacBook Pro in the past two years, you’ll find that the device has grown significantly faster than its predecessors. Applications are opened in a blink of an eye and files are quickly stored on the device. The device also turns on and off in just a few seconds. All of the above is due to the use of NVMe SSDs in recent generations of MacBook Pro, which can read and write information up to 3 times faster than the SATA SSDs used in previous generations.

Data path detection has been up to 5x faster than before. However, their SATA SSDs are several times faster than HDD variants, and all of these figures can be indicative of the high speed of NVMe-based memory. The table below shows a comparison between the performance of the three main technologies in the information storage industry.

The average speed obtainable in mechanical memory is about 1MB / s, while this for SATA SSDs is 2MB / s and for NVMe SSDs is more than 1GB / s. It is clear that many users do not need this speed, and even professional users do not achieve the maximum speed potential in NVMe standard in everyday use, but increasing this amount can dramatically improve the performance of computers.

The results you see in the table below clearly indicate the difference between the two NVMe types in different classes, the SATA and NVMe differences, and the difference between the hard drives and the SSDs:

* All benchmarks listed in this table

As you can see, the NVMe SSDs were much faster in the zoom tests than the SATA ones. The Samsung 970 Pro is an upgraded NVMe memory and the ADATA XPG SX6000 Lite is an economical NVMe memory and although they both use NVMe 1.3, there is a huge difference in speed. Of course there may be products of all three types of memory that perform better or lower than the stated level, but the values listed in the tables above are considered average.

Storage Memory / Read Speed (Write Information)	Kind of	Order in MB / s	4KB Random Single IOPS
Samsung 970 Pro	NVMe SSD	۲۷۹۵ (۲۰۸۷)	۱۴۰۱۴ (۴۱۰۷۸)
Lexar NS200	SATA SSD	۵۲۳ (۴۷۸)	۸۵۷۷ (۲۴۸۲۰)
Samsung 970 Evo Plus	NVMe SSD	۲۷۰۶ (۲۴۴۷)	۱۵۰۸۲ (۴۱۷۳۶)
ٰ Western Digital WD10EZEX	HDD 7200 RPM	1/3 (0.8)	۵۹ (۳۱۴)

* All benchmarks listed in this table are executed by Zoomit.

Undoubtedly, if a graph of the amount of memory progressed in the last five years can undermine advances in the processing of personal computers. But lower speeds have not eliminated older generations, including HDD . These types of memories are still a better option for storing large amounts of data because they are much less expensive due to SSDs, and in larger volumes, this distance is noticeable. But it is best to place operating systems, applications, and data on NVMe SSDs or if not on SATA SSDs to increase system speed.

Although the SATA port has been able to reach 8 GB / s in its version 2.3,

most of the commercial ports on PCs do not exceed the nominal capacity of 5 GB / s and in actual performance they can barely reach 1 MB / s.

Even version 4.0 of the technology is much slower than the potential of today’s SSDs, especially if used in RAID configuration.

The next step was the use of PCI Express technology by many manufacturers,

which by default were available on computers for transferring graphics card data. The technology from the third generation onwards provides the ability to use multiple paths (lanes) up to a maximum of 2 numbers, each of which can carry about 1 GB (exactly 1 MB) of data per second.

PCIe is also know to be a major contributor to the emergence of the Thunderbolt interface. This interface is now the primary option for connecting external gaming cards to computers.

NVMe’s external caches also use the said interface, which makes them almost identical to internal caches. Many users now realize how effective Intel has been in preventing Thunderbolt from being forgotten.

Although the advent of PCIe dates back to several years before the emergence of the NVMe standard, its use for storing information has not been common.

Previous protocols used in the field, such as SCSI and AHCI, were all developed during the heyday of mechanical memory and were therefore not compatible with the advanced features of SSDs.

Ultimately,

NVMe was able to eliminate the limitations of previous standards by offering many capabilities, including reducing the delay in executing commands and increasing their execution capacity by up to 4,000.

The capability mentioned in SSDs is very important because

the data in this type of products is widely stored in storage units,

while the process is done in a continuous and circular mechanical memory. The NVMe standard is still evolving and newer generations are coming to market. Version 4.0 of this protocol provides the usability of the computer’s memory as a cache .

NVMe memory formats

M.2: It is the most common form factor for NVMe memory,

with up to four lines of PCIe 3.0 or PCIe 4.0 connectors.

U.2: This form factor is also know as SFF-8639 and uses up to 1 PCIe max. Such form factor can be found mostly on computer servers.

U.3: Developed based on U.2 specifications and uses SFF-8639 connector. A controller can be used to combine SAS, SATA and NVMe. U.3 caches are compatible with U.2 but U.2 cache cannot be used in U.3 context.

AIC: Almost all of the original NVMe caches are HHHL AIC

or FHHL AIC type and uses PCIe 2.0 and PCIe 3.0 interface. A HHHL AIC memory connects to the PCIe server port.

How to use an NVMe memory?

Adding such memory to a computer without NVMe boot support would be pointless unless you are a gamer

It is possible to add an NVMe-type memory to PCIe ports by purchasing an adapter. All popular and important operating systems provide dedicated drivers to do so, and regardless of your computer’s age, speed will undoubtedly increase. But it’s not going to be that simple. To maximize the power of an NVMe SSD, you need to have your operating system installed.

Memory booting requires the BIOS to support the standard used by memory. Most bios on older computers do not support such a feature, and it does not appear that the makers intend to provide an update to add that capability.

So it would be pointless to add such a memory to a computer without NVMe boot support unless you are a gamer and install your games on that memory or from heavy processing software with high processing needs such as high quality video editing (such as 2160p). ).

Most NVMe SSDs on the market use the M.2 form factor. But having an M.2 port doesn’t mean your computer supports the standard.

This port is designed to support USB 3.0 alongside SATA and PCIe, but its early generations only support SATA, so it is advisable to read your motherboard information or buy compatibility online before purchasing new memory.

Make sure it meets the NVMe standard. Also note that the MSATA port, which is a previous generation M.2, is very similar but does not support said functionality.

The appearance of a port cannot support PCIe and NVMe, but it does seem to differentiate between PCIe x2 and PCIe x4 ports, and only the latter can support said standards. The first prototype uses a B-shaped key that separates one pin or junction. The key term is referred to as port bumps on memory troughs.

The PCIe x4 port uses an M-shaped key that, in addition to the previous bump, has another bump on the opposite side that separates the five connecting pins from the others. There is no general rule for these ports, but most B-key ports only use the SATA standard.

Nowadays, both B and M keys are common ports that can be used by all standards. Sometimes referred to as ports, two sockets and three sockets are also referred to.

If your port is older then you should use a PCIe M.2 $ 2 converter. There are products on the market like Plextor’s M9Pe that will be usable only after being ported and no additional action is required to launch them.

Also, as a simple user, you should avoid buying the 2.0-inch version of NVMe memory because they require the SFF-8639 interface designed for small memory.

This interface features 2 PCIe 3rd-generation connectors, along with 2 SATA ports and several side channels that utilize the 1.5V and 8V ports to power.

Of course,

this interface is only reachable in high-end enterprise memory and systems. If you are a holder of rare PCs with a Thunderbolt port (like many Asus motherboards), you can use this port to connect NVMe memory to your device, which gives the user a lot of power.

Not all NVMs are the same

In general, any kind of NVMe memory can make your computer faster, but not all of them are alike. For example, Samsung’s 970 Pro memory can process data at speeds of 1 GB / s and write data at 1.2 GB / s, while these values for Toshiba’s RC100, respectively, equals to 1.2 GB / s. And 2 MB / s. The difference between different models can be even more noticeable when the amount of written data is greater than the cache volume. Various factors such as the type of controller, the number and type of NAND units and the number of PCIe paths affect memory performance. Here are some of the differences and explanations for some of the following:

NVMe SSDs have PCIe x4 connectivity faster than PCIe x2 variants.
Increasing the number of NAND chips increases the number of paths and units that the controller uses to distribute and store information. For this reason, lower-capacity memories are usually slower than larger ones, even if they are of the same model.
The type of NAND chips used in memory also affects its speed. The SLC chips are faster then the MLC and TLC respectively and eventually the QLC is the slowest.

Finally, NVMe SSDs can be used for a long time on your computer or laptop,

and you will no longer need to replace parts or devices unless you are a gamer or processor. The high speed of this type of memory allows you to make the most of your computer’s other components, including the processor and graphics card.

Frequently Asked Questions

۱. How fast do NVMe 1.3-based caches reach?

It all depends on the number of PCIe lines and the type of memory. NVMe-type SSDs currently carry a maximum of 4 lines, so in PCIe 3.0 technology the maximum rated speed is 1 GB / s, but in practical terms, this speed is lower.

۲. Are M.2 and NVMe Memory the Same?

No, these two categories are quite different. NVMe is a data transfer interface that can be compared to SATA, but M.2 is a form factor that accommodates both SATA and NVMe memory.

۳. Do all laptops support NVMe-based memory?

No, a large percentage of laptops do not have a port for connecting NVMe memory,

and some of them are not modifiable if memory is available (such as new MacBooks),

so be sure to support M.2 before and after purchasing this type of memory for your laptop. Note the NVMe version.

۴. Are PCIe 4.0-based memory available in the market?

Some companies have launched PCIe 4.0-based products,

but due to the limitations of the motherboards and their newness, we are not yet seeing much of this memory.

۵. Couldn’t your computer’s SSD memory port be of type x4?

Yes. Some motherboards have x2 (dual-line) M.2 memory ports due to their limitations,

which means that your data transfer rate will probably be half the amount announced by companies or benchmarks.

Everything you need to know about the performance of the RAM modules

msrxp — Fri, 13 Sep 2019 17:41:41 +0000

The most important features of desktop RAM memory modules are their frequency, delay and timing, which play a key role in memory selection and purchase. In this article we will explore these concepts.

Buying memory for a PC is not a complicated process for some users. Those who don’t have much mental conflict about the details of memory can easily pick a product for their computer. The selection process for them only involves choosing the memory capacity of the RAM. In some cases, they only rely on the advice of the seller or other people. There are plenty of resources available in terms of detecting the amount of RAM needed, but it’s best to know that most gamers and gamers recommend an average of 1 GB of memory.

Highly professional users always seek the highest level of performance and speed. Of course, they also demand the same at a reasonable price. As a result, they will have more in-depth reviews of the products. Meanwhile, brands such as Patriot, Corsair, GSkill, Idita and others that offer high-performance high-end customers have more customers and fans.

There are a number of users in the computer world who are looking for more complete and detailed information about hardware. They need to identify the factors and terms of the hardware world to make better choices. Such information can also identify the difference in performance, price and cost-effectiveness of the products. In this post, Zoomit explores concepts in the PC world that are useful for better product selection.

The basics

To explore and explain the concepts of memory in the computer world, we focus on a variety of DDR4 memories and related terms. Today, DDR4 has become an industry standard in the memory sector, and in the last four or five years we have generally seen memories of this kind. Of course, most of the terms used in the present article also apply to past generations. However, today’s user generally works with DDR4 memory unless he has a very old computer. The following is an introduction to the basic terms of computer memory:

DIMM :

This is the abbreviation for Dual Inline Memory Module or “Dual Inline Memory Module”. DIMM variants today come with two 2-bit data interfaces on either side of the module. In terms of sales, there are two categories of UDIMM modules for desktops and SODIMM for laptops. However, some compact desktop motherboards also use SODIMM. This solution is used to accommodate four modules on such a motherboard, otherwise the motherboard will probably only support two modules.

SDRAM :

stands for Synchronous Dynamic Random-Access Memory or “Dynamic Random Access Memory”. The structure of these memories has rows and columns of data storage cells, like spreadsheets. Random access memory can access the data of each cell according to the command issued by the memory controller. The term “random” in the above definition means that the controller does not need to read the data of the entire row to access the data of a particular cell and go directly to the address in question. The term “dynamic” means that each cell must be regularly refreshed so that its data is not lost. In contrast, there are static memories that generally perform slower. All the available memory in the system is also synchronized by the external clock generator with the desired frequency.

Data Rate :

The data rate is the number of times the module sends and receives data in seconds. The clock signals generate a square wave. Here is a term called Double Data Rate, which means both data transfer and signal loss. Duplicating the data transfer rate means that a 1MHz wave will be able to move data at a rate of 2 times per second. So we now know that the DDR data frequency is double the clock frequency. The term MT / s is used for the above concept.

In the DDR memory of each signal, two data are transferred

DDR4 : The fourth generation of Double Data Rate memory that is nowadays a comprehensive standard among PC memory products. In each frequency generation, capacity and some other characteristics of memory have improved steadily.

IC : Perhaps the most familiar term in this series can be called integrated circuit. Integrated circuits are generally known by end-users as chips. IC in DRAM generally has an 8-bit interface, but there are 8-bit examples.

Data Rate: The more, the better

By definition, we know that a higher data transfer rate means more data transfer per unit of time. Of course, there are some limitations in memory controller support. Most of today’s professional processors are capable of working with 2MHz (DDR4-3600) DDR4 memory. Of course, some speed limits have been artificially changed and increased, with the sole purpose of developing the market. In this situation, however, a chip maker like Intel wants to charge more for a non-overclocked K-series processor on the pretext of supporting more memory speeds, which means it needs a more professional motherboard. Here are some examples of this artificial increase.

AMD’s family of Series 4 processors are capable of working with memory faster than the DDR4-3600, but the manufacturer in the main firmware has some limitations that cause the DDR4-3600 to exceed half the speed of the controller if it exceeds the defined frequency range of the DDR4-3600. And the deceleration is proportional to the other I / O units on the mainframe.

Practical tests on a sample CPU with G.Skill Trident Z RGB DDR4-3600 memory module confirm the same reduction in performance if CPU limitations are enabled.

In the first experiment,

the defined constraints caused the mentioned parameters to drop at a frequency of 1 MHz, but when the constraints were lifted, the performance level improved at a speed of 2 MHz in the memory module.

The earliest members of the Series 6 family of adopters accept the DDR4-3467 module frequency without functional drawbacks, but higher frequencies will definitely interfere with performance and induce noise (especially in cross-talk signal). In addition, the connection paths between the CPU socket and the DIMM on some motherboards are not ready for such an increase in frequency. However, if you are running on such processors with a lower motherboard than the X470, a data rate increase of more than DDR4-2933 is not recommended.

More data transfer rate means more data transfer per clock signal per unit time

The Intel LGA 1151 family of processors have memory controllers that perform well above the DDR4-3600. Intel, of course, designed non-Z chips to limit any processor (even the K series) to higher rates. Experimental experiments with the Z370 Core i3-8350K Core i3-8350K processor also showed higher limitations.

Finally, the easiest way to break the DDR4-2666 limit is to use Z Series chipsets with Core i5 processors or higher.

The Intel Memory Controller operates at a frequency of 1 or 2 MHz and occasionally reaches higher coefficients of 1 or 2.3 MHz. The lower the memory coefficients, the more stable they are and the more stable they are on older platforms.

For example, the Z270’s designs are more stable. This makes the DDR4-3467 13x 266.6 MHz more stable than the 17x 200 MHz DDR4-3400 and it performs better.

In summary, given the above information, we can conclude that the family of Series 6 CPUs support the DDR4-3600 when the motherboard is ready. On the one hand, Core i5 or Core i7 K Series processors can handle such data rates when installed on a suitable Z390 or Z370 board. For coordination and other limitations of CPU and motherboard combinations, you should refer to hardware forums and especially overclocking groups.

The delay, the less, the better

Latency is said to be the amount of time spent in memory to begin executing an operation (such as reading and writing). It is interesting to note that this factor has not changed much since decades. The old PC-100 system and the new, professional DDR4-3200 blend have both CAS delays of 1 nanosecond. Why didn’t there be any progress? To better understand the basic timing of operations, we need to look at the concept of memory cells being row and column. The following statements should therefore be described:

CAS or Column Address Strobe : The number of clock cycles that will be needed to access data in the new column when the correct row is opened.

tRCD or RAS delay to CAS: Minimum cycles the memory controller modules has to wait for a new row to open.

tRP or Row Precharge : Minimum cycles that the memory controller must wait for the current row to open.

tRAS or Row Active Time : The minimum number of cycles the memory controller must wait between opening one row and closing another.

CMD or Command Rate : The number of cycles that instruction is given to memory to ensure that it is read by memory. Generally the values of 1T or 2T mean a clock cycle and 1 cycle clock.

If we assume that the correct row is opened from memory, CAS will be the time we need to access the next bit in memory. And If all rows are closed, a row must first be opened to access a cell and then the correct column activated (tRCD + tCAS). If the wrong row is open (the row that does not contain the desired data), the current row must be closed, the correct row opened, and the correct column found in the new row (tRAS + tRCD + tCAS) to access the data cell. Finally, when the CMD is increased from 1T to 2T, a clock cycle will be required for each memory instruction.

Delay definitions use the time and clock cycle expressions

In the above definitions the term “time” was first used,

but in the definition of delay types we use the concept of clock cycles. In fact, the delay time is defined by the clock cycle. Again, the question is how the old PC-100 and the modern DDR4-3200 are parity delayed? The duration of a 1MHz clock cycle is 2 nanoseconds. As a result, PC-100 CAS 1 (with a CAS delay of one clock cycle) takes about 2 nanoseconds to access data. In contrast, the DDR4-3200 operates at a 1MHz clock, and a 1MHz clock cycle takes only 2.5 nanoseconds. Finally, the DDR4-3200 CAS 16 requires 6x of data access time, again equal to 1 nanosecond delay.

Since clock cycle time is inversely proportional to the clock,

the faster the memory, the more clock cycles will be needed to achieve a standard 5 nanosecond delay. The DDR4-3600 performs this activity in 1 cycle. The DDR4-4000 has a clock delay of 4 cycles. Finally, reducing the time to less than the standard 2 nanosecond requires fewer latency cycles per frequency. As a result, the DDR4-3200 C14 breaks down to a high standard record in 0.5 nanoseconds and the DDR4-3600 C16 delivers 3.5 nanoseconds.

Increase in rank, decrease in delay

In mainframe modules processes, waiting for each write or reading process to finish before running the next process slows down the processes considerably. Interleaving is a term that allows one to start a command while another is running. Users can increase the CPU’s ability to perform this process by increasing the number of ranks per channel from one to two. To do this, two single-channel DIMMs or one double-DIMM per channel must be installed.

The following information will be used for this purpose:

Most of the memory produced since year 1 uses eight gigabytes of memory ICs.
Most memory ICs have an 8-bit interface.
Eight 8-bit ICs will be needed to fill a 4-bit rank.
The total capacity for eight eight-gigabit ICs is eight gigabytes.
As a result of the above information, most 1GB memory kits have four ranks.

Structures consisting of four ranks perform the process of adding two ranks per channel to two-channel motherboards. If you look at the information above, the term “mostly” has been repeated many times. As a result, there are some exceptions to these topics that we will consider below:

Demand for 4GB ICs is high right now, and companies can’t afford to waste these chips on building single-gigabyte modules. As a result, 2GB modules are built using two 8GB IC ranks.
Consumer-specific 2GB modules use two 1GB IC ranks. As a result, two 1GB DIMMs provide a 2GB dual-channel kit with four ranks.
Current 4GB modules use four eight-gigabit ICs each with a 2-bit interface. Four of these modules will be needed to build four ranks.
Older 4GB ICs are not in high demand. However, shoppers who need 4GB of memory to access the four ranks use them. However, such ICs will be difficult to identify.

At the bottom edge of the memory module’s heatsink to detect the presence of eight ICs on each side. Of course, there are exceptions to this that can only be found online resources and forums.

Conclusion

Higher data rates will mean increased modules performance. Of course, the limitations of the CPU and motherboard should always be considered. Low latency also increases performance without increasing data rates. Four ranks perform better than two ranks, resulting in a 2 GB DDR4-3200 performance better than a 1 GB DDR4-3600.

Given the above information, we understand more about them when purchasing PC memory. Of course, there are suggestions from online retailers and experts as well, but in the end you should be careful that the memory you buy is no more than CPU processing and management. In the meantime, you need to be very careful when buying single-color kits, as they will ultimately reduce overall speed and efficiency.

Hynix Demos First 16Gb DDR5-5200 DIMMs

msrxp — Sun, 25 Nov 2018 18:45:01 +0000

SK Hynix announced on Thursday that it has successfully tested both conventional DDR5 RAM and

the first registered DDR5 DIMMs as well.

The new chips have a capacity of 16Gb (that’s per-IC, with typically 8-16 ICs per physical DIMM,

for 8-16GB of total memory capacity) and are clocked at DDR5-5200,

which provides up to 41.2GB/s of memory bandwidth per channel.

A conventional desktop outfitted in a standard dual-channel configuration would offer 82.4GB/s of overall memory bandwidth.

That’s a significant improvement over current desktop DDR4. While you can buy DDR4 at a variety of clocks,

the commonly available DDR4-3200 provides 51.2GB/s of bandwidth in a dual-channel configuration.

DDR5 should be more power-efficient than DDR4,

thanks to at least some voltage improvements.

Hynix is advertising a 9 percent decrease in core voltage on their 1Ynm process and claims that their chip meets all JEDEC specifications.

Since the DDR5 spec hasn’t actually been published yet we don’t know if this voltage decrease is typical for the class, but either way,

it’ll yield at least some improvements in power efficiency.

The chips aren’t set to be commercialized until 2020,

and we probably won’t see instant adoption from Intel or AMD — AMD has promised to support current AM4 sockets through Zen 3,

which means either they make a clean break with the old platform with Zen 4 in 2021 or support two different memory standards on the same silicon.

AMD has gone both of these routes in the past, so we’ll have to wait and see what the company opts to do this time.

Intel has not commented on when it will introduce DDR5,

but it’s not unusual for it to take several years between early memory standard demonstration and full commercialization.

As is typical,

we’d expect DDR5 to start off at a cost premium relative to DDR4 before crossing over with the older standard and eventually becoming less expensive.

One thing to always keep in mind is that latency doesn’t improve nearly as quickly as bandwidth. In fact,

DRAM latencies don’t necessarily come down at all unless you overclock or pay for top-quality RAM.

Even as clock speed increases, latency timing keeps pace,

which is why you don’t see DDR4-3200 being sold at the 2-2-2-5 timings that typified really good DDR-400. RAM timings tend to impact the performance of AMD systems a bit more than Intel rigs,

but in both cases,

the adoption of integrated memory controllers and large on-die caches has ameliorated the impact of using slower RAM.

Relative RAM latency

The big winner in all this will likely be integrated graphics solutions.

Dual-channel DDR5 would represent a 1.63x improvement in memory channel bandwidth,

and solutions like AMD’s integrated GPUs continue to be highly sensitive to increases in bandwidth.

APUs like the Ryzen 2400G scale quite well when you open the proverbial throttle and with 1080p’s enduring popularity,

we suspect we’d see these gains translate into higher frame rates as opposed to an immediate push to bump resolution (AMD, of course, would likely advertise both,

but driving higher frame rates in 1080p would be a higher priority we think).

While there’s always been a push and pull between the level of quality offered by integrated graphics and the increased demands of newer titles,

if you look at the situation over the long term,

integrated graphics solutions have improved faster than GPU minimum requirements.

We’ve gotten much better at driving games from iGPUs than we were in 2010 and memory bandwidth increases + on-die GPUs tightly integrated with their CPU counterparts are the two principal reasons why.

There was a time when it looked like HBM might cut into the desktop and laptop market,

even if only a little, but that moment seems to have passed courtesy of HBM’s enduring high prices. DDR5 looks set to have the desktop market to itself,

barring some unusual moves from AMD or Intel, neither of which has shown much inclination to pick up the banner of alternative memory technologies.