Not All First Generation PCIe 5.0 SSDs Will Offer the Same Performance

[Chrispy_]

It is relevant to video editors that work with the very high end stuff- RAW data of 6K-8K video.
Those files weight by the hundreds of GB to TB per file (it`s about 100-120 GB per minute).
Anyone else may pass.

You're not wrong, but if you're regularly working on footage from a $7,500-$50,000 camera, you can probably afford a workstation with 1TB RAM and 25GB/s network storage so that you don't need to be copying each file onto a scratch disk constantly.

Our company only works with 4K at the moment but I apply video-editing principles to workstations that work with site survey pointclouds. They're typically in the 500GB+ range per building and we deal with campuses occasionally.

NVMe SSDs are great for short projects but they rapidly get overwhelmed by this sort of stuff. When you're dealing with large datasets, you don't look at the headline figures, you look at the minimum sustained transfer rates. Reads are usually always high which is great for scrubbing but if you have to sit there and copy a stream the controller will overheat and throttle, your SLC cache will fill up, and it always comes down to the raw NAND speed which even PCIe 3.0 x2 is often fast enough for:

It's hinted at in this article, but newer NAND is coming which is the thing the video editing (and other large dataset) industries are most eagerly antipating, not PCIe 5.0. I still think that the make-or-break application that will justify PCIe 5.0 (and 4.0 for that matter) is DirectStorage but that's still a few years out in reality - even If it's launched tomorrow it will be like RTX where there aren't (m)any games that take full advantage of it for a good couple of years.

 

[Tsukiyomi91]

this is why chasing for big numbers will either be impossible or just close enough to it but at increased cost just to get there. I'll be sticking with PCIe Gen4 and saturating the heck out of it.

 

[diopter]

Even Gen 4 drives still have some room for improvement. Let the apes subsidise the development cost of GEN 5 drives early on.

 

[TheDeeGee]

There are plenty of third party options otherwise, but yes, we're getting to a point where it's starting to get hard to cool the controllers.
I ended up getting one of these for my KC3000 SSD, as it was throttling when used with the motherboard heatsinks.

HR-09 2280 – Thermalright

thermalright.com

The large heatsinks are an issue.

Maybe time for a motherboard redesign with M.2 slots on the back, plenty of space for a heatsink behind the motherboard tray.

 

[TheLostSwede]

The large heatsinks are an issue.

Maybe time for a motherboard redesign with M.2 slots on the back, plenty of space for a heatsink behind the motherboard tray.

That would require new cases though.
So would my idea, where they'd hang off the front edge of the motherboard and use up that empty space in most cases, but then again, that would only work on standard ATX boards.
That heatsink wasn't too big and works really well to cool the drive, even when stress testing it.
The flat bits of aluminium that came with the board, not so much.

 

[thestryker6]

I'm still curious what people are doing that anything above a good PCIe 3.0 drive makes a real difference. 10gbe is roughly 1.2GB/s after overhead which is less than half PCIe 3.0 a prior poster mentioned 120GB/m video which is closer to 2/3 of PCIe 3.0. Sustained reads/writes are a problem for every current SSD (due to caching) and plagues them no matter the interface as does random read/write performance. We need something that rewrites the book for consumer NAND storage overall not just higher sequential burst numbers.

 

[Panther_Seraphin]

Gen 5 makes sense in the Datacenter as the are able to saturate PCI-E 4 already. But in the consumer space its very much overkill!!!

 

[jigar2speed]

Are you telling us that there is a full chance companies can commit scam and legally get away with it ?

I guess i am skipping PCIe 5.0 SSD in that case.

 

[TheLostSwede]

Are you telling us that there is a full chance companies can commit scam and legally get away with it ?

I guess i am skipping PCIe 5.0 SSD in that case.

No scam, they simply can't get hold of fast enough NAND flash, so the controller ends up being bandwidth starved.
As such, it's a bad idea getting this first generation of drives, much like many early PCIe 4.0 drives.

 

[evernessince]

Well, I see no reason to store terabytes of music, movies and photographs on an SSD, unless I'm editing in Premiere. In fact, I'm quite curious about HAMR/MAMR hard drives and really want a couple of 20tb ones once they hit the consumer markets.

22TB drives are already on the market. 20TB have been for awhile now.

Gen 5 makes sense in the Datacenter as the are able to saturate PCI-E 4 already. But in the consumer space its very much overkill!!!

Heck there's only a small bump for PCIe 3.0 drives over SATA drives in consumer workloads. PCIe 4.0 and 5.0 bring the average person zero benefit (literally). It's not the sequential speed that's going to improve performance for consumers, it's latency and random reads/writes. Something that drive manufacturers haven't really improved on much.

The large heatsinks are an issue.

Maybe time for a motherboard redesign with M.2 slots on the back, plenty of space for a heatsink behind the motherboard tray.

This wouldn't be compatible with every case. You'd be better off going the daughter board route at that point.

 

[Space Lynx]

Can't wait until Noctua DH15's are needed to cool them NVME drives.

We can have one on our CPU, one on our GPU, one on each NVME, and suddenly we have six DH15's in our PC's. "All Noctua" suddenly has a whole new meaning.

it would be a sexy monstrosity. i am in.

 

[Dristun]

22TB drives are already on the market. 20TB have been for awhile now.



Heck there's only a small bump for PCIe 3.0 drives over SATA drives in consumer workloads. PCIe 4.0 and 5.0 bring the average person zero benefit (literally). It's not the sequential speed that's going to improve performance for consumers, it's latency and random reads/writes. Something that drive manufacturers haven't really improved on much.



This wouldn't be compatible with every case. You'd be better off going the daughter board route at that point.

Lol, I'm behind the times. Thanks, cool! Need to see if they're available here.

 

[Shou Miko]

Why doesn't this topic surprise me anymore

Serious it's been like this even with HDD different performance levels.

 

[Chrispy_]

Sustained reads/writes are a problem for every current SSD

It's mostly just writing that is the problem.

Even shit-tier, DRAMless, QLC, budget SSDs can generally sustain decent read speeds. Maybe don't expect high IOPS and low latency but they can sequentially stream gigabytes a second for practically their entire capacity.

 

[RJARRRPCGP]

Just like how the early SATA SSDs had different speeds at the drive level.

 

[RogueSix]

The large heatsinks are an issue.

Maybe time for a motherboard redesign with M.2 slots on the back, plenty of space for a heatsink behind the motherboard tray.

ASUS has already had mainboards with at least one m.2 slot on the back of the board a while ago but it seems like they discontinued putting it there for some reason. I don't think you really need a heatsink on the board's back side for normal use. It will be much cooler than the slot under the graphics card. The back of the board is usually very cool compared to everything on the front.

I believe it would be great if mainboard makers would revisit the concept because, yes, thermals are going to be an increasing problem with faster SSDs. The top PCIe 4.0 drives already get close to the thermal limits even when "only" gaming if they are installed in the top slot under a RTX 3090.
The 25mm width might help a little, but in the long run, I figure we will need a redesign with riser cards or a switch to PCIe cards with active cooling and more than 16 PCIe slot lanes on the boards so that a SSD card won't take away lanes from the GPU.

 

[TheLostSwede]

Just like how the early SATA SSDs had different speeds at the drive level.

It was very different back then, as we had half a dozen controller makers.
All of the first generation of PCIe 4.0 SSD's were based around a single controller from Phison and the same appears to be true for the first generation of PCIe 5.0 SSD's.
The competition is simply not keeping up with Phison when it comes to releasing new products.
As far as the drive makers are concerned, it comes down to who can source what NAND flash, the ones that get the good stuff, will have the faster drives.

 

[evernessince]

Lol, I'm behind the times. Thanks, cool! Need to see if they're available here.

18TB is the current price / performance winner if you are looking at buying now. There's a steep up-charge for 22TB drives and a smaller but definitely noticeable up-charge for 20TB.

I would not buy above 18TB unless you absolutely need the space now. The price of the higher capacity drives tend to drop pretty sharply until they go under $280 and then their value starts retaining better.

I do really wish there we could see larger consumer facing SSDs than 8TB and with reasonable price tags.

 

[Frank_100]

im more interested what kind of cooling components attachments provided by motherboard manufacturers
to combat the massive heat output by the drives at max speed... pci 3.0 drives already get pretty hot and needs an aluminium heatsink on.

It may be time for a redesign of the fishtank pc.

Maybe not he whole machine, just the motherboard, memory, SSD's, & CPU.

 

[AsRock]

News flash, not every thing is equal.

 

[evernessince]

ASUS has already had mainboards with at least one m.2 slot on the back of the board a while ago but it seems like they discontinued putting it there for some reason. I don't think you really need a heatsink on the board's back side for normal use. It will be much cooler than the slot under the graphics card. The back of the board is usually very cool compared to everything on the front.

I believe it would be great if mainboard makers would revisit the concept because, yes, thermals are going to be an increasing problem with faster SSDs. The top PCIe 4.0 drives already get close to the thermal limits even when "only" gaming if they are installed in the top slot under a RTX 3090.
The 25mm width might help a little, but in the long run, I figure we will need a redesign with riser cards or a switch to PCIe cards with active cooling and more than 16 PCIe slot lanes on the boards so that a SSD card won't take away lanes from the GPU.

They stopped doing that because 1) It's not compatible with the majority of cases that people own 2) The back of the board has to rely entirely on passive dissipation. Even with hotter air, actively cooled is far superior. The only exception might be an M.2 slot that's placed right under your GPU exhaust. Then again that's a combination of GPU choice and board design so part of that is on the user.

 

[nguyen]

I still haven't found valid reason to buy PCIe gen4 NVME, now gen5 is coming out?

 

[R0H1T]

I mean how do you expect them to continue making (insane) profits? The regular low(er) end SSD prices have sunk below ground level ~ Silicon Power A55 2 TB SATA SSD Now $103 at Amazon

 

[tetrapak]

No 15GB\sec - no buy. I can't stand those turtle like 10-12GB\sec drives.

Tru dat. For a regular home usage, there’s no difference even between sata and PCIe 3.0 , they are speaking about 15GB\sec.

 

[evernessince]

I mean how do you expect them to continue making (insane) profits? The regular low(er) end SSD prices have sunk below ground level ~ Silicon Power A55 2 TB SATA SSD Now $103 at Amazon

That's nice but unfortunately the savings per TB don't translate to the higher capacity models. 8TB (Samsung is the only company with 8TB consumer facing models) is still $87.50 / TB and that's considering it's QVO which has lower endurance.

The rate at which SSD price per TB has been decreasing has been extremely slow and that's in light of the fact that endurance continues to decline. Someone is going to have innovate in the field because continuously reducing endurance is not a sustainable path towards cost effective larger capacity SSDs. File sizes continue to increase so you cannot expect customers to buy SSDs of raising capacity but continuously diminishing endurance (mind you this will play over 10+ years).