Samsung 980 PRO quietly launched, read speeds up to 7GB/s

[leadeater]

That fact that it's TLC and only up to 1TB makes the least sense of everything, if it were still MLC I could see some kind of logic to it, maybe.

[RejZoR]

58 minutes ago, Medicate said:

With the new consoles having high speed SSDs in them I completely expect games in the coming months and years to utilise those insane read and write speeds of PCIe 4.0 drives. I can imagine 500mb/s not being enough anymore in the near future for games that want to leverage those console SSD speeds.

Which just won't happen outside consoles. There are still so many people running spinning drives it won't happen for many years.

[AlTech]

So basically samsung have created a 980 EVO and are rebranding it as a 980 PRO because profits?

 

Okay. Time to not buy Samsung cos they need to stop being greedy.

 

PRO is supposed to be MLC.

EVO is TLC

QVO is QLC.

 

This is just big joke to be honest.

[Dabombinable]

My Vertex 460A has a (calculated) durability of 21.9TBW over its 3 year warranty with MLC (wear amount supports that - 81% after 4 years). Unless the 980 Pro is used as a scratch disk, or in high res video camera you won't notice the lower rated endurance.

[S w a t s o n]

The fuck is Samshit trying to pull here?? I've been pateintly waiting and they decide to throw some TLC into a drive and sell it as a PRO model? Big Pass, other drives will similarly be able to max pcie 4.0 I can keep waiting.

[Eaglerino]

6 hours ago, TempestCatto said:

I do not like this. It should be called the 980 Evo instead. I know TLC has come a long way, but I still prefer MLC (or if I can get my paws on it, SLC).

Gotta admit of all the SSD's I've bought over the years I never once looked at the type of memory they had and if you put them all together, even the older ones, guarantee I wouldn't be able to tell the difference on day to day use.

[GDRRiley]

50 minutes ago, Eaglerino said:

Gotta admit of all the SSD's I've bought over the years I never once looked at the type of memory they had and if you put them all together, even the older ones, guarantee I wouldn't be able to tell the difference on day to day use.

QLC is fairly new and easy to tell. hit it with a big write and then it's write speed will tank

[DeScruff]

4 hours ago, RejZoR said:

TBW ratings are rather meaningless unless you do absurd sequential writes every day. My Samsung 850 Pro 2TB is rated at 300 TBW. I know torture tests where people rather easily pushed them past 1PBW without any errors. Yeah, 1 Petabyte of writes. I used to have this write anxiety in the beginning, but now I just don't care and just use it however I use it. I'm currently at almost 32TB written and roughly 35.000 operational hours. Life percentage dropped to 97%. It just doesn't matter anymore at this point with exception of very specific stupid high intensity write scenarios. Which are just not gonna be happening on home systems even of most demanding users.

Agreed, SSDs have pretty much long since passed the point of the TBW being relevant metric on home/office/gaming PCs. These aren't the 64GB TLC drives of yesterdecade.
In those environments the PC themselves, are likely to become outdated before the drive starts wearing out. Heck If the SSD gets moved over to the next PC, it might even outlive that. Most people are probably more likely to upgrade their SSD due to their capacity being too low then due to the SSD wearing down.
- I mean... I took my 250GB 850 Pro outa my system, cause my parents needed a SSD, even a 120GB would do and I wanted more capacity. With <20TBW written, I feel confident it will last throughout the life of their machine.

I can imagine in a server enterprise environment the same can't be said. But I don't believe the Samsung Pro drives have ever been marketed for that so.

[StDragon]

3 hours ago, AluminiumTech said:

So basically samsung have created a 980 EVO and are rebranding it as a 980 PRO because profits?

 

Okay. Time to not buy Samsung cos they need to stop being greedy.

 

PRO is supposed to be MLC.

EVO is TLC

QVO is QLC.

 

This is just big joke to be honest.

It's all about density per chip. MLC is on the way out. TLC and QLC is the future. If you can obtain desirable speeds and acceptable write endurance, the type of NAND being used doesn't matter in the end.

 

If they weren't marketing performance, it would be QLC without question.

[Rune]

aaaand waiting for a phison e18 now..

[Bombastinator]

29 minutes ago, rrubberr said:

It really becomes impractical. Games don't write, and reading 500mb/s would read the entirety of a 100GB game in under four minutes. If anything latency and IOPS to read data, if developers get even sloppier about caching the data in RAM and VRAM.

 

Content creation requires these speeds, while companies desperately try to convince consumers who don't know any better that they "need the best."

There have been descriptions of how programs are coded to deal with the lower speeds of mechanical HDs and how much more effort and other resources it takes.  What I unknown is what is going to actually be done.  Buying something with the expectation that it could only possibly be useful becomes sort of questionable if the price is too high.  These drives might still be too slow for example and no more useful than the cheapest SSD currently available.  The storage differences between consoles aren’t as easy to determine as say AMD64 core counts and MHz.  This one is much more fuzzy.  3600x and 3700x at least looked like pretty sure things.  Hopefully soon there will be enough data about how the consoles work to be able to make a determination.   I don’t expect much to be buyable in an informed way until probably November at the earliest barring useful leaks. If info doesn’t appear what will likely have to happen is Someone will have to get ahold of a couple production consoles, put them through their paces and take them apart. 

[LAwLz]

11 hours ago, Kisai said:

The 1TB model has twice the TLC performance of the 500GB drive. I reasonably suspect the same is true of all manufacturer's using the same scheme. So the reason we likely don't see 2TB drives in M2 form factors is probably because they can't physically fit more chips on the 2280's used for laptops. I've also not seen any engineering laptop come with a 2TB NVMe drive yet either.

Ehm, there exist plenty of 2TB NVMe M.2 type 2280 drives.

Here is a list of 59 different drives that fit that criteria. Some examples:

Samsung 970 EVO (Plus)

Sabrent Rocket Q

Corsair MP600 FOrce

WD Blue

Sabrent Rocket 4.0

Intel 660p

 

Edit:

There are even TLC drives that fit 4TB of storage on them, like the Corsair MP510.

[Soppro]

Very curious to see how this stacks up against upcoming Phison E18 drives which should offer performance in the same range as well. Good to see more PCIe 4.0 drives 

[porina]

9 hours ago, comander said:

For what it's worth many current games are designed with harddrives (or worse OPTICAL DRIVES) in mind. Seeks are VERY VERY costly. Streaming data into memory and then processing it is relatively cheap. 

This means that games take their assets and convert them into a handful of VERY VERY large files (with lots of the same stuff repeated) that get streamed and processed into memory. 

This is a gripe of mine at the moment. I wish games would allow you to specify what format your game data files are optimised for. They can still use a highly compressed version for distribution, but why not have an option to repack it locally after download e.g. if you have a SSD use faster compression (reduce CPU bottleneck) and de-dupe content.

[porina]

13 minutes ago, comander said:

One thing to point out - deduplication is sometimes used to refer to data at a block level. Instead of writing out every single block on disk blocks that are identical are only written once and then any use of that data is then pulled by linking to the single copy. Useful for repeating patterns in a database. 

 

Duplicating assets is a bit of a different concept altogether. 

 

Forgive any pedantic rambling I dove into the deep end of data storage architecture lately. 

Agree with that. I was going to reply to more of your previous post but thought it would end up going deeper than I can be bothered with. To keep it simple, I'm just talking about it from a user space viewpoint. We have graphics settings already to cover for low to high end systems, why not file optimisation settings? In my game of choice, load times hit some limit, which I think is CPU limiting due to the compressed files. We get a big jump from HD to low end SSD, a smaller jump with high end SSD, smaller again to Optane, and that little more for ramdisk. I'd hope to see better scaling if they remove the CPU element somewhat. If the storage is slow the CPU is not a limit, but fast storage reverses that situation, where the disk isn't maxed out while it waits for CPU. Or allow more parallel reads to use more CPU threads. Extremely bad for a HD, but could help SSD. Removing intentionally duplicated data would help offset any footprint increase from less aggressive compression. Actually, even then, the requirement is low CPU decompression. A high compression cost is not a concern for the user, so it doesn't necessarily follow faster decompression means less compression.

[Doobeedoo]

I was looking forward to it finally. But wait what it's not MLC huh. That's so odd, the price better be lower at least that's for sure. I'm very interested to see a review for this, caching better be as good as MLC though. Speeds are looking great. 

Their Pro drives were always quite decently more expensive and hardly got cheaper, so if they manage to offer speeds and sustained with no worries and prices lower for Pro should be very expected then. 

It really sucks how 'SSDs are getting cheaper' is actually with an asterisk that they do but with lower quality nand flash that scuff endurance and sustained performance. So many drives have scuffed TBW and especially pathetic cache once exhausted performance falls appart in two digit MB/s for some. 

 

Another thing, where is their consumer version of theie Z-SSD that will be amazimg to see too. 

[Smackaroy]

21 hours ago, TempestCatto said:

I do not like this. It should be called the 980 Evo instead. I know TLC has come a long way, but I still prefer MLC (or if I can get my paws on it, SLC).

TLC is fast enough and has a longer longevity than anyone needs 

[Curufinwe_wins]

21 hours ago, RejZoR said:

TBW ratings are rather meaningless unless you do absurd sequential writes every day. My Samsung 850 Pro 2TB is rated at 300 TBW. I know torture tests where people rather easily pushed them past 1PBW without any errors. Yeah, 1 Petabyte of writes. I used to have this write anxiety in the beginning, but now I just don't care and just use it however I use it. I'm currently at almost 32TB written and roughly 35.000 operational hours. Life percentage dropped to 97%. It just doesn't matter anymore at this point with exception of very specific stupid high intensity write scenarios. Which are just not gonna be happening on home systems even of most demanding users.

The only manufacturer that is ever mattered on was Intel because they self-bricked after a certain amount of writes. But SSDs have almost always done much better than their ratings suggest.

 

Note these are from small 5 year old ssds including the dreaded sandforce tlc drive lol. 

https://techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead/

 

[DeScruff]

10 hours ago, LAwLz said:

Ehm, there exist plenty of 2TB NVMe M.2 type 2280 drives.

Here is a list of 59 different drives that fit that criteria. Some examples:

Samsung 970 EVO (Plus)

Sabrent Rocket Q

Corsair MP600 FOrce

WD Blue

Sabrent Rocket 4.0

Intel 660p

 

Edit:

There are even TLC drives that fit 4TB of storage on them, like the Corsair MP510.

I know a couple of those drives are double sided while Samsung drives are all single sided (better compatibility with Thin&Light laptops). I dont think there are any 4TB single sided drives but I could be wrong. So not all of em are apples to apples.

 

If I recall the 970 series SSDs also had a delayed launch of 2TB drives? I think its likely the same is gonna happen here. Maybe a release in a few months?

[Kisai]

10 hours ago, LAwLz said:

Ehm, there exist plenty of 2TB NVMe M.2 type 2280 drives.

Here is a list of 59 different drives that fit that criteria. Some examples:

Samsung 970 EVO (Plus)

Sabrent Rocket Q

Corsair MP600 FOrce

WD Blue

Sabrent Rocket 4.0

Intel 660p

 

Edit:

There are even TLC drives that fit 4TB of storage on them, like the Corsair MP510.

Look at the configurations that the machines come with.

 

 

The standard SSD on all high end laptops is 500/512GB, if you select "more than 1TB" on the HP page you're still sent to laptops with 512GB. The Dell configurations won't let you select another hard drive capacity.

 

I'm not saying you can't buy 2TB drives. I'm saying they aren't making them because the OEM's aren't using them.

 

 

[StDragon]

5 hours ago, comander said:

Generally speaking, at least with contemporary systems, compression/decompression with LZMA or ZSTD is often faster than reading uncompressed data. The CPU ends up being MANY times faster than a bundle of SSDs so shifting some of the burden to the CPU from the storage ends up as a performance win. 20% compression means 20% more bandwidth and 20% more cache space. 

 

In the case of the PS5, they have a specialized co-processor just for decompression, that supposedly offers 9 Zen2 cores worth of performance. 

https://www.tweaktown.com/news/71340/understanding-the-ps5s-ssd-deep-dive-into-next-gen-storage-tech/index.html#:~:text=The PS5 uses a derivative,to decompress Kraken-level data.

The problem with compression/decompression is that it incurs higher latencies (lower IOPS). So for streaming and sustained data pull access, it's not much of a big deal. But when you're dealing with lots of files (as in, millions of small files), random access, SQL queries etc, it's a substantial performance hit. I've experienced this first-hand with a ZFS array involving lz4 compression.

[leadeater]

55 minutes ago, StDragon said:

The problem with compression/decompression is that it incurs higher latencies (lower IOPS). So for streaming and sustained data pull access, it's not much of a big deal. But when you're dealing with lots of files (as in, millions of small files), random access, SQL queries etc, it's a substantial performance hit. I've experienced this first-hand with a ZFS array involving lz4 compression.

With things like SQL you move the storage compression up in to the database engine, the data is actually still compressed just not at the storage layer. Ideally you actually want compression to always be done at the application layer but few actually do it or to a competent degree that does anything meaningful.

 

Still for enterprise storage arrays you do want to turn on compression as it actually gives higher IOPs and throughput but these have much more optimized firmware, software and hardware to deal with it, all that built in cost etc.

 

As for games the content is still better off compressed simply due to how they store it currently and also have multiple copies of them for certain graphics settings, optimization for HDDs. Once HDDs are firmly dead and considered gone and not worth thinking about, even mildly, we should actually see some drastic changes to game data and install sizes.

[StDragon]

9 minutes ago, leadeater said:

With things like SQL you move the storage compression up in to the database engine, the data is actually still compressed just not at the storage layer. Ideally you actually want compression to always be done at the application layer but few actually do it or to a competent degree that does anything meaningful.

 

Still for enterprise storage arrays you do want to turn on compression as it actually gives higher IOPs and throughput but these have much more optimized firmware, software and hardware to deal with it, all that built in cost etc.

 

As for games the content is still better off compressed simply due to how they store it currently and also have multiple copies of them for certain graphics settings, optimization for HDDs. Once HDDs are firmly dead and considered gone and not worth thinking about, even mildly, we should actually see some drastic changes to game data and install sizes.

To be brutally honest, even with SQL in the application layer, lots of RAM will be thrown at it for caching. And depending on how well the DB is optimized, the cache-miss ratio will be kept to a minimum. As for enterprise storage, how much of that latency is reduced with cache vs ASIC hardware? I figure if you're going to be running uber expensive enterprise gear, chances the cost of cache is a minuscule in overall pricing. It's low-hanging fruit not to invest is more read-cache RAM.

[leadeater]

24 minutes ago, StDragon said:

As for enterprise storage, how much of that latency is reduced with cache vs ASIC hardware? I figure if you're going to be running uber expensive enterprise gear, chances the cost of cache is a minuscule in overall pricing. It's low-hanging fruit not to invest is more read-cache RAM.

That's part of the flow for storage compression. Our Netapp 8200's for example have 128GB RAM, 16GB NVRAM, 1TB NVMe read cache per controller and all data that lands on SSD is compressed. SSDs do actually perform significantly better with compressed data but you have to avoid weak links in the chain otherwise it's no better or worse.

 

Far as I know roll your own ZFS implementations don't do as good of a job at inline compression, one part get what you pay for and one part nature of hardware flexibility. Even so ZFS compression is still really good a lot of the time, just not always. 

 

Quote

4130 Mbit/s / 260 020 IOPS @2.7ms (ORT = 1.04 ms)

http://spec.org/sfs2014/results/res2017q3/sfs2014-20170908-00021.html (without SSDs)

[StDragon]

18 minutes ago, leadeater said:

Far as I know roll your own ZFS implementations don't do as good of a job at inline compression, one part get what you pay for and one part nature of hardware flexibility. Even so ZFS compression is still really good a lot of the time, just not always. 

Specifically, I was referring to a TrueNAS unit (M series) by iXsystems. They're nice units for what they are, but there's no ASIC hardware acceleration involved regarding compression. If I recall the CPUs are Intel Xeons. Your Netapp is most likely superior kit.