News Silicon Motion is developing a next-gen PCIe 6.0 SSD controller

[OldAnalogWorld]

You are mistaken - without a DRAM buffer, the translator quickly stops fitting into the small dram buffer built into the controller (usually 32-64 MB) and each data search operation goes to the flash, with an order of magnitude drop in performance. The HMB buffer is useless here - the translator is not stored there and it is also too small - 64 MB. Therefore, for a system disk - a loss of a dram buffer of at least 1GB, especially when the disk is busy with internal work with the flash, and data comes from the OS/software for writing, without it everything is very bad and this is easily proven even with such a load as simultaneous distribution / receipt of files in torrents. Such disks behave very badly there. As well as at the beginning of intensive simultaneous through the OS and in other software in parallel. Buying a bufferless disk is justified only as an operational backup and cache of the most frequently used data.

P31 will be faster especially in 3.0 slots. And there is no alternative to it. I just looked at the prices for the 990 evo plus in local large retail chains and it costs more than the 990Pro ($210 vs $200) - this is crazy, but that's the reality.

I just need to replace one old drive that came with the laptop, because the wear and tear there is approaching the dangerous limit of 35% and I don't want to risk it any further, but there is simply nothing else to choose from - the P31 in local retail comes with a only 1-year warranty from sellers and is ~$50 more expensive than the 990Pro in the 2TB version.

 

[bit_user]

You are mistaken - without a DRAM buffer, the translator quickly stops fitting into the small dram buffer built into the controller (usually 32-64 MB) and each data search operation goes to the flash, with an order of magnitude drop in performance.

I understand your claims. Your logic sounds reasonable, but that's not what matters. What actually matters is whether it's true, and the only way we can know that is by testing it.

So, when I look at the data, it seems to me like Samsung has adequately mitigated such concerns through clever design of their FTL (Flash Translation Layer) and effective utilization of however much on-chip memory it has. If what you're saying is right, then their low block size random performance (esp. QD1) should be horrible, but it's not at all!

​

It's worth pointing out that the Crucial T500 does have DRAM, yet only performs a little better in these tests!

without it everything is very bad and this is easily proven even with such a load as simultaneous distribution / receipt of files in torrents. Such disks behave very badly there.

Okay, so show me a benchmark like that. You've defined a test it should fail, so now let's see if it does. It's a popular drive, so it shouldn't be hard for you to find some benchmark like that.

P31 will be faster especially in 3.0 slots.

I already showed you a benchmark where it wasn't faster - the 50 GB folder copy, although that was almost certainly in a PCIe 4.0 slot. The Evo Plus had a higher efficiency, while using a little more average power. That means its performance must've also been higher, or else its efficiency would've been worse.

 

[OldAnalogWorld]

I don't believe in these tests, because they are artificial tests, and empirical experience has clearly shown me how much worse unbuffered drives are than models with a full DRAM buffer. And in any case, in local retail 990 Evo plus is simply pointless to buy, because it costs more than 990Pro. By the way, local networks recently sold out KC3000 for 2TB for $125-130. Because it is even hotter and even worse compatible with many laptop models, because it is double-sided (and they sell many times more of them than desktops). And MSI 480 Pro 2TB for $155-160 is also freely available, very hot, but with much less reliability in batches than KC3000 according to statistics from large retail chains - where, by the way, Toshiba 112L or 176L Micron chips can accidentally arrive, which is an unpleasant fraud in batches from Kingston.

It's very sad, but not a single manufacturer, except Hynix, even thought of making a low-power version with a DRAM buffer using modern technical processes tailored for laptops. Although the same Samsung already has the ability to make chips using "3nm", which will sharply reduce the consumption of controllers on old 3.0/4.0 buses and this is the mainstream in the world now, and no one needs 5.0 at all. They are all in collusion and for the sake of profit, because they need to sell new models with a higher margin than the old ones with new controllers on "3nm". They all promote only new hardware by force and M$ is in collusion with them too.

Capitalism (or rather oligopolistic collusions) in the endless pursuit of profit regardless of anything leads to such sad consequences when instead of progress for the masses we get an increasingly worse situation in each individual sector...

 

[thestryker]

I don't believe in these tests, because they are artificial tests

You can certainly feel free to believe this, but that doesn't make it even remotely true. They're an accurate reflection of how the drives perform.

empirical experience has clearly shown me how much worse unbuffered drives are than models with a full DRAM buffer.

There are absolutely bad DRAMless drives, but there are also good ones. This tends to be mostly dependent on the NAND being used. Just because you've used bad drives doesn't make them all bad. The SN770 is the oldest good DRAMless drive I can think of and even then its shortcomings were mostly due to its NAND. Cheap drives also tend to be the ones which have component swaps to maintain margins more often.

 

[bit_user]

I don't believe in these tests, because they are artificial tests,

The nice thing about synthetic tests is that they are actually more extreme than real world usage, because the kind of I/O pattern you tend to have in real life is where the the CPU does some other things in between submitting I/O requests and those requests are often larger (especially when accounting for read-ahead due to buffered I/O and the kernel). So, if the synthetic tests don't show a huge difference, then it's unlikely real world tests will, either.

They have other benchmarks, too. PCMark 10 is based on I/O patterns they captured from of real software applications. So, it's designed to closely replicate real world performance. Here, as well, the drive performs competitively:

​

NAS Compares liked it, noting its power efficiency and even recommending it for laptops:

https://nascompares.com/2024/10/11/...msung_990_EVO_Plus_Gen_5_SSD_Review_-_Verdict
​

Tweak Town liked it so much, they gave it an overall rating of 99% and their "Must Have" Editors Choice Award!

https://www.tweaktown.com/reviews/1...us-2tb-ssd-the-dramless-powerhouse/index.html
​

and empirical experience has clearly shown me how much worse unbuffered drives are than models with a full DRAM buffer.

Ah, but that's the issue here. You're just assuming it's bad, on the basis of being DRAM-less. You might be right that DRAM-less SSDs tend to be worse, but they're always the budget models. The thing about budget SSDs is that they don't just cut one corner. They cut lots of corners. So, just because some DRAM-less SSDs are bad doesn't mean they all are.

That's why people actually test these drives. There are always exceptions to the rule - DRAM-based drives that perform poorly and DRAM-less ones that perform well. You can't know which is a good or bad drive without testing. So, that's why I asked where are the test results that show it's a bad drive? I don't know how you can expect anyone to believe you, instead of these professional reviews, absent any evidence.

And in any case, in local retail 990 Evo plus is simply pointless to buy, because it costs more than 990Pro.

Maybe that's specifically because it's more suitable for things like laptops? Also, the 990 Pro is stuck in a weird spot, because it's meant to be a premium drive - but a lot of people wanting premium drives are going for PCIe 5.0 models now. So, they're probably having trouble charging as much as they normally would for it. Another factor could be that the Evo Plus launched much more recently and had less time to build up inventory ahead of the holidays.

Anyway, it must be a short-term market distortion. It certainly doesn't reflect the long term pricing trends for those products.

It's very sad, but not a single manufacturer, except Hynix, even thought of making a low-power version with a DRAM buffer using modern technical processes tailored for laptops.

I actually wonder if some of the OEM SSD models aren't geared towards lower power. Perhaps even some retail SSDs have configuration settings that laptop OEMs use, which run in lower-power modes. As you mentioned, Samsung's 990 Pro has such a mode, but it's really not good for much. I'll bet other SSDs have similar modes that might work better.

Another factor is that the budget laptops are still largely PCIe 3.0. If you take any of these SSDs and run it at that speed, they're going to use less power. Maybe Crucial's T500 is a good 4-channel, DRAM-based drive to look at, in such situations?

this is the mainstream in the world now, and no one needs 5.0 at all. They are all in collusion and for the sake of profit,

I don't see why it has to be "collusion". There's clearly some demand for these products, whether it's people like you who just want the highest-end specs or not, the fact is that enough of them are willing to pay the extra money for PCIe 5.0 drives, at a time when very decent PCIe 4.0 options still exist.

Capitalism (or rather oligopolistic collusions) in the endless pursuit of profit regardless of anything leads to such sad consequences when instead of progress for the masses we get an increasingly worse situation in each individual sector...

Speed vs. efficiency tradeoffs are nothing new, in tech. For a long time, lithography processes made it possible to get Moore's Law type gains at the same power. This scaling broke down not because of market forces, but because of physics.

https://en.wikipedia.org/wiki/Dennard_scaling#Breakdown_of_Dennard_scaling_around_2006
​

Efficiency gains were once "free", but that hasn't been true for a long time.

 

[OldAnalogWorld]

I don't care how many tests you show, my personal experience clearly shows that drives without DRAM buffer behave very badly in difficult situations. And they are worth buying only if they cost 1.5 times less than normal versions with DRAM buffer. There is no point in showing me these graphs - I will simply ignore them.

I actually wonder if some of the OEM SSD models aren't geared towards lower power. Perhaps even some retail SSDs have configuration settings that laptop OEMs use, which run in lower-power modes. As you mentioned, Samsung's 990 Pro has such a mode, but it's really not good for much. I'll bet other SSDs have similar modes that might work better.

What prevents manufacturers from adding such settings for regular customers in their proprietary utilities, like managing the TDP of processors? Just a maximum power (current) regulator. That's all. And so that this setting is written to NVRAM or somewhere else so that even when moving the disk to a new laptop, it immediately consumes no more than the specified threshold at the start-up stage. Approximately how charging controllers are programmed - they do not depend on the settings in the OS. If I set the necessary thresholds, they will always be applicable, even after turning off and on the laptop under another OS that knows nothing about it. This is the kind of non-volatile setting I expect from ALL manufacturers of SSDs with a DRAM buffer.

Maybe Crucial's T500 is a good 4-channel, DRAM-based drive to look at, in such situations?

It is not sold in our local retail. I do not consider ordering remotely on principle due to the complete absence of a normal warranty. For this reason, I do not consider Hynix either, although it is the best - with its high price and minimum warranty of only 1 year, this is an exorbitant level of risk - 50/50 that my copy will fail outside the short warranty period. Only models of working local 3+ year warranty.

Speed vs. efficiency tradeoffs are nothing new, in tech. For a long time, lithography processes made it possible to get Moore's Law type gains at the same power. This scaling broke down not because of market forces, but because of physics.

This is a priori a false argument. Many old models with old controllers (as well as with new flash) can be easily converted to new technical processes, reducing consumption several times. But greedy capitalists do not do this precisely because they need to sell new hardware frequently, although there are hundreds of millions of old laptops and PCs in the world, the owners of which will gladly buy 2.5" SATA/PCIE 3.0/4.0 versions with maximum efficiency using the most modern technological processes. But capitalists are not interested in this - because there is no way to justify a new increased margin level, because there will be no "newness" of the product, it will be the same model with the same performance, but with a sharply reduced energy consumption. This also proves that they do not care about the "green economy". Profit is above all, even the ecology of the planet.

What prevents them from making old optimized controllers for the 3.0/4.0 bus with 2-3 times lower consumption at "2/3 nm", instead of "7-12 nm"? There are at least 100+ million people willing to buy these old models with new energy-efficient versions of controllers and flash - 2ТБ+ versions. Which is already economically justified for R&D and launching new batches.

Efficiency gains were once "free", but that hasn't been true for a long time.

I am the one who writes most often about the silicon impasse - there is no need for me to point out a fact that is well known to me (and you know it).

 

[bit_user]

I don't care how many tests you show, my personal experience clearly shows that drives without DRAM buffer behave very badly in difficult situations.

But, because those are generally cheap drives, you don't know that their poor performance was entirely (or even primarily) due to the lack of DRAM. And even if a generalization is well-founded, there are often exceptions. By refusing to look at facts and others' experiences, you're not allowing for this possibility.

Like @thestryker said, DRAM-less drives are usually the cheap drives that also use cheap NAND. The cheap NAND will only compound the performance impact of not having a DRAM buffer. However, the 990 Evo Plus is a TLC drive, not QLC. So, it's not using Samsung's cheapest NAND (as further underscored by its 1,200 TBW endurance rating, for the 2 TB model).

And they are worth buying only if they cost 1.5 times less than normal versions with DRAM buffer.

At "1.5 times less", they would be paying you to take it!

There is no point in showing me these graphs - I will simply ignore them.

LOL, it's like the old saying:

"Don't confuse me with facts, I already know what I believe!"
​

This is a priori a false argument.

I think you missed my point, which was that efficiency is no longer free. The designer has to make a conscious decision whether to opt more for performance or efficiency. Because there's generally more demand for performance, efficiency often gets de-prioritized.

Many old models with old controllers (as well as with new flash) can be easily converted to new technical processes, reducing consumption several times.

That costs money. So, the return on investment would need to make it worthwhile, which it rarely is if it's just exactly the same controller.

there are hundreds of millions of old laptops and PCs in the world, the owners of which will gladly buy 2.5" SATA/PCIE 3.0/4.0 versions with maximum efficiency using the most modern technological processes.

Have you seen any data on PCIe 4.0 SSDs running at PCIe 3.0? Or PCIe 5.0 SSDs running at PCIe 4.0? I'm sure it's better - I just don't know how much. It might be pretty close to what you're asking for.

This also proves that they do not care about the "green economy". Profit is above all, even the ecology of the planet.

They sell what consumers demand. If consumers demanded greater efficiency, then we'd get it (i.e. at the expense of some performance).

 

[OldAnalogWorld]

not QLC

I have never had and never will have QLC drives.

(as further underscored by its 1,200 TBW endurance rating, for the 2 TB model

Which is a shameful 600 cycles, although the firmware calculates wear at 1500 cycles. But in reality, 3D TLC has 2500-3000 cycles. For YMTC 3D TLC generally set fake 5050 cycles in most cheap Chinese disks.

which it rarely is if it's just exactly the same controller.

Which is not at all obvious with such a massive demand for energy-efficient drives with a DRAM buffer for laptops with PCIE 3.0/4.0. At least for very old hardware with SATA3 slots, the 870 Evo is an obvious choice and has no problems with power and heat dissipation. There used to be an MX500, but it is EOL, and its resource is 3 times lower (relative to 1500 cycles according to the firmware - I have them all). Well, there is also a KC600, but with its price on SMI2258/2259 and an unknown NAND grade, there is no point in it compared to the 870 Evo. Everyone prefers to take Samsung with its own NAND memory in 100% of cases and its own controllers.

True, Samsungs, as I already wrote, have a key drawback - if the disk fails (the translator or firmware crashes due to firmware errors or power failure), it is impossible for an ordinary buyer to restore it independently for free - there is no proprietary software on the Internet for reflashing in firmware mode. While for other controllers, such proprietary utilities are available. And I have restored disks in this way, which have an obvious problem in the firmware when the disk is heavily filled for bufferless versions. Which requires either a paid operation in an authorized service center outside the warranty period (which is sometimes dangerous in the event of data leaks) or throwing the disk in the trash.

They sell what consumers demand. If consumers demanded greater efficiency, then we'd get it (i.e. at the e

I constantly read in reviews and on technical forums about problems in laptops of fast and hot series and the lack of any choice in terms of energy-efficient ones. Those buyers who install hot series often find out that the M.2 slots do not have enough power or they are low-profile and double-sided drives do not fit there, and without a radiator and with the usual poor airflow in laptops, the M.2 slots are a constant thermal bath. And then people, after 1-2 years, suddenly discover a failure of the "reliable series" in such conditions.

The demand for energy-efficient drives with a DRAM buffer for laptops, including old ones (for upgrades, since old drives from the factory have long been worn out and are small in size - a key factor encouraging owners to upgrade old, but quite functional laptops for several more years, at least) is huge. And the mass opinion of incompetent buyers is manipulated by professional marketers. Anyone who has at least a little understanding of the topic understands the deplorable situation on the SSD market for laptops, especially the capacious 2-4-8TB series.

I read a detailed review on the website of a local retail chain some time ago about a customer who had inserted an 8TB Adata drive (S70) into an Asus TUF 17" model 2024 and discovered that under heavy loads (he has heavy tasks), it spontaneously disappears/turns off, and then reappears after turning the laptop off/on (he was lucky that he did not lose the working data on it), i.e. Asus' failure to comply with the power supply standard (maximum current) for M.2 slots of 15W is obvious in this series of laptops (or the SSD manufacturer AData, if this is also likely - the peak current is higher than the maximum level according to the standard, there is no way to check this without a technical examination). Since he replaced the drive with the same one and the same symptoms in both slots. But how could he know and how can he prove under warranty that the laptop is of poor quality (does not meet the power standard for M.2 slots, if this is exactly the problem in the slots), If Asus did not specify the maximum power of the slots, but only publishes "compatibility" lists? And this is the case with all manufacturers - no one specifies the maximum power of M.2 slots, USB ports, real DP/HDMI protocols. Everything needs to be found experimentally. But what to do if such information is not even in the reviews, and a new laptop is urgently needed? What to do when this becomes clear after the purchase - what to appeal to, for example, after six months? This is a completely abnormal situation - the lack of explicit standardization and requirements for the mandatory transfer of information to consumers on the part of the developers of the standard, which obliges compliance with the standard or does not allow the use of slots/ports in end products under the trademark of this standard.
And where, who, in reviews, measures the peak power of M.2 slots/usb ports, the real bandwidth of DP/HDMI ports? I have never seen this, in any review of a laptop or PC motherboard.

 

[bit_user]

Which is a shameful 600 cycles,

It's the same endurance rating as the DRAM-based 990 Pro (i.e. for drives of the same capacity). So, either that's saying something good about the 990 Evo Plus or something bad about the 990 Pro. In the reviews of the 990 Pro, I didn't see anyone criticize its endurance.

And where, who, in reviews, measures the peak power of M.2 slots/usb ports, the real bandwidth of DP/HDMI ports? I have never seen this, in any review of a laptop or PC motherboard.

In PC motherboards, I think there's not a problem providing adequate power to the M.2 slots, so long as the user has a sufficient power supply. I almost never read laptop reviews, so I can't comment on them.

Regrading USB, it can negotiate its power. So, the stakes are considerably lower and it should be non-fatal if a port can offer less than the maximum amount allowed by the standard.

 

[OldAnalogWorld]

In PC motherboards, I think there's not a problem providing adequate power to the M.2 slots, so long as the user has a sufficient power supply. I almost never read laptop reviews, so I can't comment on them.

I have seen reviews where there were obvious power supply problems in a number of desktop boards. If you do not clearly indicate the power of slots and ports in the product specifications - this is 100% a reason for fraud with characteristics for a mass (read technically illiterate) buyer. And these are the manufacturers who constantly engage in fraud with strict compliance with standards in order to save money and unfairly increase the profits of the company owners.

Regrading USB, it can negotiate its power. So, the stakes are considerably lower and it should be non-fatal if a port can offer less than the maximum amount allowed by the standard.

If the port does not have sufficient power supply (and this is a common practice of saving), then no agreement will help - it simply cannot provide so much, and if it gives (by mistake) more than it can - the power circuits will burn out. And this has happened more than once in desktop motherboards and in laptops, I have read about such reviews.

Therefore, it is necessary to force all manufacturers at the legislative level to comply with previously introduced industrial standards, quickly ratifying them at the level of mandatory state standards. If the "invisible hand of the market" does not want to behave decently - society must force them to do so by force through official requirements for the product and the mandatory publication of all this data. But now there is wild chaos on the market, artificially and intentionally created by unscrupulous manufacturers of components and finished gadgets. There are many examples of such behavior.

 

[bit_user]

I have seen reviews where there were obvious power supply problems in a number of desktop boards.

If there's a desktop board with sub-standard power delivery to these slots, I'd like to see it. Can you find one of them for us?

If the port does not have sufficient power supply (and this is a common practice of saving), then no agreement will help

It can, though. Something like a graphics card can limit its clock speeds to stay within the limits of the available power, for instance. Same thing with a SSD, actually.

if it gives (by mistake) more than it can - the power circuits will burn out. And this has happened more than once in desktop motherboards and in laptops, I have read about such reviews.

Again, I'd love to see one of those reviews. If it's not in English, I can use Google Translate to read it.

Therefore, it is necessary to force all manufacturers at the legislative level to comply with previously introduced industrial standards, quickly ratifying them at the level of mandatory state standards.

That's not realistic for all countries. Perhaps a few major countries or trading blocs, though.

But now there is wild chaos on the market, artificially and intentionally created by unscrupulous manufacturers of components and finished gadgets. There are many examples of such behavior.

Well, it's a good thing reviewers discovered these practices, so that at least their readers can avoid the boards.