not impressed - nvme vs ssd

[Waldorf]

@dirtyferret
start to wish i hadnt "retired" to here.
bottle of coolant is 40-50E online, what i had gotten in the states for 10-15$,
same for hw parts, and sub 200$ Lian li is just over 300E (plus 50 shipping).

 

[bonehead123]

@Waldorf

Just curious, but is there some reason why you don't use the full width of the forum's webpages when posting your comments ?

Are you afraid someone will get lost or not be able to follow along if you do, or is it just some kind of parsing thing on your end ?

Can you please enlighten me, TIA

 

[Waldorf]

lol.
no, i just hate left/right scrolling (mobile), and assume others do too,
but only seems to be an issue on other forums im using, just forgetting to use full width here

 

[SchumannFrequency]

One such workload where its clearly better than SATA is simply downloading a steam game on a gigabit or faster connection. I did this on my 9900k test bed, which has 2 SATA SSD attached as storage, and the SSD was bottlenecking the download once OS write cache was exhausted, 100% load on the drive.

I suspect you will have to download at more than 200 MB/s?

More than 95% of people have download speeds of <200 MB/s.
And most servers don't even give users >200 MB/s..

To be completely honest, there are also large performance differences between file systems in many workloads.

XFS has things it is fast for. ZFS is lightning fast for some tasks. There are also obscure file systems such as HAMMER2 and FFS.

For example, OpenBSD's FFS is, in my experience, slower than other file systems for opening apps.
But when I do a search for files in FFS it seems to be much faster than Windows 10/11 and most Linux file systems.

I still use an EVO 500 GB (SATA) because I'm afraid that if I buy the fastest NVMe drive, I'll regret that there will hardly be a difference.

On my Samsung EVO 500 GB (SATA) with XFS (on Calculate Linux) I see this disk usage in Steam:

At times I see 'disk usage' going to 4.1 Gbps and higher.
What values would I see with an NVMe on Windows 10/11?

HAMMER2 outperforms any file system by very large margins for database workloads.
It also has the best reliability under heavy load of any file system.

I don't see many persons using it professionally except a few Cambridge graduates.
But NVMe (more expensive) is something that's worth it when HAMMER2 (100% free) is not worth it?

Seems like a logical issue for me.

 

[Dragam1337]

I suspect you will have to download at more than 200 MB/s?

More than 95% of people have download speeds of <200 MB/s.
And most servers don't even give users >200 MB/s..

To be completely honest, there are also large performance differences between file systems in many workloads.

XFS has things it is fast for. ZFS is lightning fast for some tasks. There are also obscure file systems such as HAMMER2 and FFS.

For example, OpenBSD's FFS is, in my experience, slower than other file systems for opening apps.
But when I do a search for files in FFS it seems to be much faster than Windows 10/11 and most Linux file systems.

I still use an EVO 500 GB (SATA) because I'm afraid that if I buy the fastest NVMe drive, I'll regret that there will hardly be a difference.

On my Samsung EVO 500 GB (SATA) with XFS (on Calculate Linux) I see this disk usage in Steam:

At times I see 'disk usage' going to 4.1 Gbps and higher.
What values would I see with an NVMe on Windows 10/11?

Really depends alot... i have a 500 gb evo sata aswell, and it doesn't have the best write speed - the bigger capacity sata models are alot better in that regard. But between my 1 and 4 tb sata ssds and 980 pro 2tb (nvme), the difference is miniscule in most cases. Like a straight file transfer, the 980 pro is obviously alot faster, if what you are transfering to is equally fast. But verifying files as in your example, you will more often than not be singlethread limited on your cpu. But for the few instances where that aint the case, the 980 pro can do it at like 1500 MB/s, where as the others do it at 500 MB/s. Loading games, and playing games, i really don't see a difference, as the latency difference between is so small that you can't tell (and that's what matters for games).

 

[A Computer Guy]

I am not impressed with NVME too, and because of that I added to it a RAMDrive

View attachment 346582

Please tell me you have a side of ECC with that.

( AMD Radeon RAMDisk 8GB )

 

[SchumannFrequency]

Really depends alot... i have a 500 gb evo sata aswell, and it doesn't have the best write speed - the bigger capacity sata models are alot better in that regard.

RND4k Q1T1write speed is more than 2x faster than the RND4k Q1T1 read speed for this drive. It's not really 'slow'

Generally F2FS and XFS are the fastest on the consumer solid-state storage devices generally followed quite quickly (and commonly trading blows) with EXT4.
But for launching apps, F2FS is frequently twice as fast as the popular EXT4.
Btrfs with its copy-on-write behavior leads to it having a lot of features but at least in its out-of-the-box behavior generally being a fair amount slower than EXT4/F2FS/XFS.

I understand what you are saying about the file transfer, but how many times a year do you do large file transfers? I do it very rarely anyway.
And it happens quickly. I mean even if you copy the full 500 GB drive, if it's large files (sequential read) it would take 1000 seconds. Or just 16 minutes

I also installed Manjaro with F2FS on this EVO 850 500GB. Then I did a large system update, more than 1.3 GB. You then have several factors.
The package manager Pacman is fast. And the F2FS file system is fast. But what I saw was that this large full system+app update happened almost at the speed of light on the EVO 850 500GB.

 

[Dragam1337]

RND4k Q1T1write speed is more than 2x faster than the RND4k Q1T1 read speed for this drive. It's not really 'slow'

Generally F2FS and XFS are the fastest on the consumer solid-state storage devices generally followed quite quickly (and commonly trading blows) with EXT4.
But for launching apps, F2FS is frequently twice as fast as the popular EXT4.
Btrfs with its copy-on-write behavior leads to it having a lot of features but at least in its out-of-the-box behavior generally being a fair amount slower than EXT4/F2FS/XFS.

I understand what you are saying about the file transfer, but how many times a year do you do large file transfers? I do it very rarely anyway.
And it happens quickly. I mean even if you copy the full 500 GB drive, if it's large files (sequential read) it would take 1000 seconds. Or just 16 minutes

I also installed Manjaro with F2FS on this EVO 850 500GB. Then I did a large system update, more than 1.3 GB. You then have several factors.
The package manager Pacman is fast. And the F2FS file system is fast. But what I saw was that this large full system+app update happened almost at the speed of light on the EVO 850 500GB.

Well, that was kinda my point - it doesn't really matter. Even a mechanical drive can read and write faster than the 1 gpbs connection transfers, which is the fastest internet speed most people can get.

Mechanical drives have terrible access times though, and suck for anything other than mass storage, but sata ssd's are absolutely adequate for 99% of consumer pc's.

 

[SchumannFrequency]

Well, that was kinda my point - it doesn't really matter. Even a mechanical drive can read and write faster than the 1 gpbs connection transfers, which is the fastest internet speed most people can get.

Mechanical drives have terrible access times though, and suck for anything other than mass storage, but sata ssd's are absolutely adequate for 99% of consumer pc's.

I agree with your point that SATA vs NVMe doesn't matter for standard users:

But I've been trying to make the nuance that the file system can have a significant impact for the everyday user:

https://openbenchmarking.org/embed.php?i=1901053-SP-LINUX421F77&sha=87c01c6&p=2

https://openbenchmarking.org/embed.php?i=1901053-SP-LINUX421F77&sha=48f8de1&p=2

 

[Waldorf]

maybe not so much of a difference playing most games, but enough out there that are map based FPS,
so not having to wait on the only member thats using a fast hdd/slow ssd is not fun (up to 20-30s),
and loading games that have +80gb installed usually tends to be faster as well, and becomes more noticeable if
ppl are doing things like shadowplay recording at the same time, especially if its a single drive for everything,
a nvme doing ~1500 R/W would still be a better idea vs ~500 MB on sata.

and short of FS limitations, i do think stuff like AV scans etc, should do better with drives 3-4x faster..

 

[AVATARAT]

Please tell me you have a side of ECC with that.

( AMD Radeon RAMDisk 8GB )

View attachment 347203

What do you mean by "a side"?
DDR5 chips have partial support on-die error correction code.
If you mean that this can "break" my file system, I see no difference from before to started using it.

And I use not only RAMDisks but PrimoCache, so it's a bit complex but work very well

 

[Wirko]

except not every user knows that, and when looking how many found at online retailers
are just listed as M.2 in the title, best case they bought a drive up to 10x slower than it should be (gen 4).

Alright, I'll say this and then stop. The OP is long absent anyway. The motherboard in your signature has M.2 slots in this configuration:

M.2 SATA SSDs will work in slots M2_1 and M2_6 Those are the "best cases", or rather, special cases. They will not work in the other four slots - at all.

 

[A Computer Guy]

What do you mean by "a side"?

It was a bit of humor. Ramdisk without ECC is like a deathwish for data integrity but I am exaggerating a bit. Old tech like Acard ANS-9010 with DDR2 you could observe corrupted data within minutes if you didn't use ECC RAM or use the chip function to enable ECC correction for non-ECC RAM.

 

[Waldorf]

@Wirko
not sure why keep trying to tell me what my hw can do/how it works, as i knew about any of it even before buying,
and has nothing to do with my response about the avg user possibly buying the wrong type of M.2 (sata),
either because it wont work in the slot they are trying to use, or at best, will be slower than what it could be doing on a pcie (R/W wise).

@A Computer Guy
if using any form of ram cache, i would rather be concerned about data corruption/loss (possible bricking of a drive), when there is any brown/blackout,
so my priority would be on a UPS, not so much as ECC ram.

 

[chrcoluk]

I suspect you will have to download at more than 200 MB/s?

More than 95% of people have download speeds of <200 MB/s.
And most servers don't even give users >200 MB/s..

To be completely honest, there are also large performance differences between file systems in many workloads.

XFS has things it is fast for. ZFS is lightning fast for some tasks. There are also obscure file systems such as HAMMER2 and FFS.

For example, OpenBSD's FFS is, in my experience, slower than other file systems for opening apps.
But when I do a search for files in FFS it seems to be much faster than Windows 10/11 and most Linux file systems.

I still use an EVO 500 GB (SATA) because I'm afraid that if I buy the fastest NVMe drive, I'll regret that there will hardly be a difference.

On my Samsung EVO 500 GB (SATA) with XFS (on Calculate Linux) I see this disk usage in Steam:

At times I see 'disk usage' going to 4.1 Gbps and higher.
What values would I see with an NVMe on Windows 10/11?

HAMMER2 outperforms any file system by very large margins for database workloads.
It also has the best reliability under heavy load of any file system.

I don't see many persons using it professionally except a few Cambridge graduates.
But NVMe (more expensive) is something that's worth it when HAMMER2 (100% free) is not worth it?

Seems like a logical issue for me.

I cannot remember the throughput but did post a task manager screenshot on another forum (which showed the i/o stats etc.), so I can hunt it down and post it here, the issue with steam is its not a simple sequential write, it downloads loads of many clusters, so is kind of a hybrid random/sequential. Small games would probably be ok as the OS write cache initially absorbs it.

 

[AVATARAT]

It was a bit of humor. Ramdisk without ECC is like a deathwish for data integrity but I am exaggerating a bit. Old tech like Acard ANS-9010 with DDR2 you could observe corrupted data within minutes if you didn't use ECC RAM or use the chip function to enable ECC correction for non-ECC RAM.

I have not investigated that, but they probably have ECC or something similar integrated into the programs for that.

if using any form of ram cache, i would rather be concerned about data corruption/loss (possible bricking of a drive), when there is any brown/blackout,
so my priority would be on a UPS, not so much as ECC ram.

I have a UPS, but it didn't help when the PC reset by too high/unstable CPU/GPU overclock

 

[Waldorf]

@AVATARAT
lol. how about finding stable oc BEFORE using ram cache?

 

[AVATARAT]

@AVATARAT
lol. how about finding stable oc BEFORE using ram cache?

I have stable OC, but often I experiment again when sometimes someone does a few points more than me

 

[quickchangethechannel]

Never heard of those brands. But here's my Sandisk SSD vs Samsung Evo 980 Pro. Doing a lot of small jobs compared to 1 big job takes much longer. But should still be 50-100% faster. That would be like the bottom test. Lots of small files.

I also use a Samsung Evo 970 for Windows / boot and my laptop takes about 3 seconds to get to login after switching on.

 

[Dragam1337]

Never heard of those brands. But here's my Sandisk SSD vs Samsung Evo 980 Pro. Doing a lot of small jobs compared to 1 big job takes much longer. But should still be 50-100% faster. That would be like the bottom test. Lots of small files.

I also use a Samsung Evo 970 for Windows / boot and my laptop takes about 3 seconds to get to login after switching on.View attachment 347368

What matters in daily usage is access times and random speeds, and there they are fairly close. Sequential speeds only matter in few instances.

 

[quickchangethechannel]

I am not impressed with NVME too, and because of that I added to it a RAMDrive

View attachment 346582

Dam that is FAST bro. Here is my C drive with the same app and settings. When I first got the drive, I also removed the intel optane drive and put in another nvme drive without first disabling it in windows, so my drives are stuck in RAID mode and could be affecting performance. Yours is 10x faster than mine. Mind you, laptop is a few years old now, but the prices are ridiculous for a new one at the moment.

Edit: Just checked PCIE version, only 3 unfortunately.

 

[Dragam1337]

Never heard of those brands. But here's my Sandisk SSD vs Samsung Evo 980 Pro. Doing a lot of small jobs compared to 1 big job takes much longer. But should still be 50-100% faster. That would be like the bottom test. Lots of small files.

I also use a Samsung Evo 970 for Windows / boot and my laptop takes about 3 seconds to get to login after switching on.View attachment 347368

I just did my 980 pro 2tb vs my 860 evo 4tb (sata)

And while the sequentials on my 980 pro are literally 10 times higher than the 860 evo, the random speeds are much closer - there is still a difference mind you, but not to near the degree that one would perhaps think, if one was to believe the hype. My 860 evo actually gets better random speeds than your 980 evo in 3 out of the 4 tests.

 

[quickchangethechannel]

Trying to find the intel optane drive so I can disable raid and see if it makes a difference. I love Acer, pop the bottom off and it's all there

 

[Dragam1337]

Trying to find the intel optane drive so I can disable raid and see if it makes a difference. I love Acer, pop the bottom off and it's all there View attachment 347375

Might wanna give those fans some cleaning while you are in there

 

[quickchangethechannel]

I did just that unfortunately it says my pcei link is x4 and capped at 4GB/s. Bummer!