Would You Rather QLC w/DRAM or TLC without?

[Bombastinator]

22 hours ago, Blue4130 said:

Don't look at some random peak number. Look at sustained transfer. Here's an example... 

 

It may peak at over 400, but it doesn't last lomg... 

So faster only for short spurts.  There’s also the latency advantage. Implication is the more cache the better then.

[Blue4130]

2 hours ago, Bombastinator said:

So faster only for short spurts.  There’s also the latency advantage. Implication is the more cache the better then.

No, it's 'look at the overall performance of a "thing", don't look at a single aspect.' 

 

You need to look at every metric and base it off of your needs. Avoid most of the marketing buzz because it's usually heavily cherrypicked. Research how something is made. Why did they make it this way, how will it be a benefit or how will it hinder. 

[Bombastinator]

11 minutes ago, Blue4130 said:

No, it's 'look at the overall performance of a "thing", don't look at a single aspect.' 

 

You need to look at every metric and base it off of your needs. Avoid most of the marketing buzz because it's usually heavily cherrypicked. Research how something is made. Why did they make it this way, how will it be a benefit or how will it hinder. 

Sure but what does this have to do with your statement?  There seems to be a sort of repetitive “No because [insert unrelated truism]”. You seem more interested in creating some sort of “no” than actually arguing the point. 

[Wooden Law - Black]

On 2/24/2022 at 1:52 AM, SquintyG33Rs said:

TLC vs QLC don't seem to make much of a difference in real world performance. I'd just get whatever is the cheaper option tbh.

It always depends on what SSD is talking about. For example, between a MX500 (TLC + DRAM) and a 870 QVO (QLC + DRAM) there is a massive difference as performance under sequential write (SLC cache), the MX500 does 500-550 MB/s while the 870 QVO around 80 MB/s. Even a BX500 (DRAM-less TLC up to 960GB SKUs) is better than the 870 QVO post-SLC. 

[Blue4130]

3 hours ago, Bombastinator said:

Sure but what does this have to do with your statement?  There seems to be a sort of repetitive “No because [insert unrelated truism]”. You seem more interested in creating some sort of “no” than actually arguing the point. 

To me it seems like you are overly focused on the initial peak speed as some high bar that should be focused on. I am just trying to show that number is highly irrelevant. If that's not the case, apologize.

[Bombastinator]

1 hour ago, Blue4130 said:

To me it seems like you are overly focused on the initial peak speed as some high bar that should be focused on. I am just trying to show that number is highly irrelevant. If that's not the case, apologize.

It’s true that I don’t think the number is completely worthless, because the majority of general uses will be in that area.  How often does it take more than a few seconds to do various things? There are things where it matters though such as game level loads and whatnot.  Those still seem to be a good bit faster than with an HDD. It may depend on how much of that is done.  I remember watching various videos that tested real world actions like game and level loading times and video editing stuff, where the difference between a cheapass SSD and a high end SSD while still large when views as fractions of each other were big, but the actual number was simply not that large.  The other problem that got run into with games is sometimes such things are timed for HDDs and there is pointless waiting with either system because both were faster than even good HDDs

[NewMaxx]

Best QLC drive is the 670p since it has Intel's 144L QLC, but we have a lot of good QLC coming in that was shown off at ISSCC this year. However, as far as the current market goes, that's the only drive I can recommend with QLC in the NVMe space. E12(S) + 96L QLC is acceptable but generally you don't get enough of a discount to make it worthwhile. Most people fail to realize that reads come from native flash and 96L QLC (even Intel's) is far slower than TLC. In realistic terms you might not notice this in 99% of cases but why save a bit of money for something that's objectively worse?

 

Plus, NVMe DRAM-less drives with TLC can be quite good, for example the SN770. You shouldn't compromise with QLC (aside from maybe the 670p at this time).

 

When it comes to SATA drives, simple: DRAM or go home.

 

I should note that most DRAM-less drives have poor steady state performance for multiple reasons. Directly, yes, the lack of DRAM can hurt maintenance (garbage collection), but also sustained speeds tend to be low since firmware forces writes through SLC first in order to reduce write amplification among other things (on DRAM-less drives). So it's not a binary "DRAM or no DRAM" because DRAM-less drives are designed for a (cheap) market segment such that other design decisions focus on hiding performance pitfalls. Which is to say, you can wind up with a full drive and terrible consistency; DRAM-less drives should be considered as having less effective capacity if you are price-comparing.

 

Re: the "good enough argument" - this is true. Just about any SSD is better than nothing (a HDD). You will likely not notice the difference between an entry-level and a high-end SSD. However, there are caveats. SATA (AHCI protocol) absolutely benefits from DRAM. If we're talking about NVMe instead, TLC objectively is more consistent and reliable than QLC with or without DRAM. However, when comparing edge cases, which can include a fuller drive doing a bunch of small updates (Windows + Steam for example), the user experience is clearly better with TLC and a more conservative SLC caching scheme. To turn the argument here around: you're not buying one SSD over another for the 99% of the time performance, but the 1% of the time where the difference is huge. How much is that worth to you? That's the real question.

[Generalkidd]

So I did ultimately get the Samsung 870 QVO QLC w/ DRAM and I decided to keep it too rather than exchange it for a DRAM-less TLC SSD. From my personal usage so far, it *feels* a lot better than the previous DRAM-less TeamGroup SSD I had. The Samsung Magic software's performance mode for the SSD seems to help even further too. One thing that's really eyebrow-raising for me is that in CrystalDiskInfo, after only writing 586 GB to the SSD so far, the health status has already been reduced to 99% whereas other SSDs I've written over 100 TB and still remained at 100%. I'm sure it's nothing to be concerned about but it was a bit surprising. 

[mariushm]

The Samsung magician "performance" mode basically implements a cache in your computer's RAM so it locks away some computer ram that otherwise could be used by games or software.

You also risk losing data if you have sudden power failure, because the software may cache writes for a short time so anything changed in that period may not be actually sent to the drive.

 

The health thing is typical for QLC ... like I said, their erase count is quite low, at under 1000, closer to 500 ... so some cells were already erased a few times based on that amount.