Should SATA get updated specs? Example SATA 4.0 @ 36Gbps, 48Gbps or 64Gbps.

[Selaya]

Why would you need a premium SATA drive? Even the most basic ones max out the SATA 3 speed. Reliability isn't a concern, either, because they're cheap.

look at the reviews, a good sata ssd can rougly match or even beat a bad nvme, but a bad sata ssd like a qvo's quite a bit slower.

 

[chrcoluk]

The SATA connector could be fitted with more connectors like they did with USB and still remain backward compatible. They could also expand the edge connector out with more connections. For example, see below:
View attachment 247849
This is just a quick mock-up, but something like this would preserve backward compatibility while providing for greatly expanded bandwidth potential.

Problem with that, boards would surely come with less slots to fit the wider connectors, I personally wouldnt consider that worth it, the more I think about it the more I think there is no demand for it. Unless maybe the extra part could be chosen to be used as a SATA port in itself at the legacy speed. That would give the best of both worlds.

SATA is not a bottleneck to HDD's, and SATA SSD's are bottlenecked by it, but its a high enough bottleneck that for most people they wont care. Most of the visible performance boost from NAND comes from quicker seeks. The few use cases that require actual super high sequential speeds are served by NVME.

 

[lexluthermiester]

Problem with that, boards would surely come with less slots to fit the wider connectors, I personally wouldnt consider that worth it

Agreed. Which is why I think the same SATA connector, fitted with more contact pins, would be the better solution.

 

[Chrispy_]

SATA and AHCI are such a bodge/kludge for SSDs. Yes, they work, but even the concept of NCQ is designed for rotating platters. The fact that SATA SSD controllers can use NCQ in reverse to work around some of the limitations of the SATA and AHCI protocol is a bonus, but definitely not how things should be or an efficient process!

Far better to phase out SATA SSDs and leave SATA 3.0 with its 550MB/s bandwidth per port for mechinical drives. No mechanical drive has really warranted more than SATA 2.0 to date. The 16TB enterprise drives I use in SANs and NASes are pushing 250MB/s on the outer edges of the platters. Realistically you can bifurcate a SATA 3.0 port into two drives and see zero performance penalty.

Keeping 2.5" SATA drives around as way to put an SSD into an old laptop or legacy PC has been great for the past decade and because we're talking ancient hardware that predates NVMe, there's no point developing better SATA SSDs. The SATA interface is so rarely a real-world bottleneck in these older machines that just moving the OS & applications off mechanical drives is plenty to shift the overall performance bottleneck so far away from the storage than there's no point improving it.

If mechanical drives enter the market that can comfortably exceed 550MB/s then there's always 12Gb/s SATA Express.

 

[Count von Schwalbe]

if pcie ribbon extensions exist for graphics cards then one would thinks they could make a cord for an nvme interface and have 6 of them on a mobo instead of sata ports

either way I’d say yes. Sata is in need of a revision. I’d guess it’s just gonna stay a usb 3.0 type a interface forever??

The modern Thunderbolt cables would work quite well, a smaller connector and very fast speeds. The trouble is, most CPU's don't have the lanes and it requires dedicated hardware. OTOH, SATA also requires dedicated hardware...

Thunderbolt 3 with a scaling mechanism to split PCIe lanes as more drives were added could work well though.

 

[Valantar]

The modern Thunderbolt cables would work quite well, a smaller connector and very fast speeds. The trouble is, most CPU's don't have the lanes and it requires dedicated hardware. OTOH, SATA also requires dedicated hardware...

Thunderbolt 3 with a scaling mechanism to split PCIe lanes as more drives were added could work well though.

Thunderbolt is really, really expensive though, it requires very expensivecontrollers and cabling. And it's hardly suitable for internal usage anyway - there's likely too much RF noise inside of a PC for that to be reliable. Why not just use native PCIe?

 

[Count von Schwalbe]

Thunderbolt is really, really expensive though, it requires very expensivecontrollers and cabling. And it's hardly suitable for internal usage anyway - there's likely too much RF noise inside of a PC for that to be reliable.

I was under the impression that it used a particular type of USB-C cable, but maybe I am wrong. AFAIK the reason the microcontrollers are expensive is because it is an Intel/Apple proprietary standard, which could be changed. Dunno about RF, I would have thought a USB cable would have been more shielded than a PCIe riser or a SATA cable.

 

[Aquinus]

I was under the impression that it used a particular type of USB-C cable, but maybe I am wrong. AFAIK the reason the microcontrollers are expensive is because it is an Intel/Apple proprietary standard, which could be changed. Dunno about RF, I would have thought a USB cable would have been more shielded than a PCIe riser or a SATA cable.

I think the part you're missing is that Thunderbolt cables have a chip on both ends of the cable to alter the signal to something that can actually travel over a wire for some reasonable distance (unlike PCIe alone.) This is a big reason why they tend to be pretty expensive. It's also why the connector gets warm. It's also why you can get fiber optic thunderbolt cables that can be run 100 meters (and those cost a metric crap ton.) Thunderbolt is expensive all the way around.

Why not just use native PCIe?

Distance. Thunderbolt can be run further distances than PCIe alone by itself.

 

[Frick]

Probably no. But a new standard for "external" (as in not connected to slots on the motherboard) storage units like HDD's and slightly faster speeds? Sure. Maybe even USB drives, but internal.

Thunderbolt is really, really expensive though, it requires very expensivecontrollers and cabling. And it's hardly suitable for internal usage anyway - there's likely too much RF noise inside of a PC for that to be reliable. Why not just use native PCIe?

Is there noticable RF inside a computer case?

I was under the impression that it used a particular type of USB-C cable, but maybe I am wrong. AFAIK the reason the microcontrollers are expensive is because it is an Intel/Apple proprietary standard, which could be changed. Dunno about RF, I would have thought a USB cable would have been more shielded than a PCIe riser or a SATA cable.

Intel has loosened the grip on Thunderbolt a bit, but that happened fairly recently. Plus the stuff above.

 

[Count von Schwalbe]

I think the part you're missing is that Thunderbolt cables have a chip on both ends of the cable to alter the signal to something that can actually travel over a wire for some reasonable distance (unlike PCIe alone.) This is a big reason why they tend to be pretty expensive. It's also why the connector gets warm. It's also why you can get fiber optic thunderbolt cables that can be run 100 meters (and those cost a metric crap ton.) Thunderbolt is expensive all the way around.

Distance. Thunderbolt can be run further distances than PCIe alone by itself.

Probably no. But a new standard for "external" (as in not connected to slots on the motherboard) storage units like HDD's and slightly faster speeds? Sure. Maybe even USB drives, but internal.

Intel has loosened the grip on Thunderbolt a bit, but that happened fairly recently. Plus the stuff above.

Yes, I thought the PCIe was more "native" to Thunderbolt. USB4, however, will support PCIe tunneling, and the cables should be cheaper than TB. Also, the controllers will be integrated into the chipset anyways, so why not? Even if there were USB "decoders" at the far end of the line to convert to SATA or PCIe (m.2 or u.2 connector).

 

[oxrufiioxo]

I'd rather a board come with more Nvme slots than an updated sata spec. I stopped using sata when boards starting being equipped with 3+ Nvme slots

For me it would only be appealing if the extra bandwidth came at no cost to the bandwidth of pcie which I find doubtful. Even now using x amount of nvme disables sata ports on most boards equipped with 3+.

.

 

[Aquinus]

Yes, I thought the PCIe was more "native" to Thunderbolt. USB4, however, will support PCIe tunneling, and the cables should be cheaper than TB. Also, the controllers will be integrated into the chipset anyways, so why not? Even if there were USB "decoders" at the far end of the line to convert to SATA or PCIe (m.2 or u.2 connector).

Thunderbolt requires both a controller on the motherboard and the connected device as well as the mux/demux chips on the cable. You could totally do something like that inside the PC, but I don't think that you're going to like how much the cables are going to cost, particularly if they require additional shielding from being inside a chassis. USB 4 is basically just TB3 to be honest.

 

[ARF]

The discussion came up in another thread and it seemed like it might be interesting to find out what everyone thinks about the subject.
Would you like an updated spec for the standard SATA connector?

No..

I agree with @Valantar

I see no reason for an updated SATA standard. The current spec is plenty for HDDs, and NVMe handles anything else. m.2 is vastly more practical, and there's u.2 if you need off-board mounting - which is easily adapted from any m.2 slot or PCIe slot. The lack of extra controllers is a boon for efficiency and makes for simpler chipsets and motherboards, and the inherent flexibility and scalability of PCIe is great. Thr only drawback is the increased need for PCIe controllers, but faster SATA would require that as well.

+1.

There is no need for SATA. Maybe it should be left to die off...

 

[Valantar]

I was under the impression that it used a particular type of USB-C cable, but maybe I am wrong. AFAIK the reason the microcontrollers are expensive is because it is an Intel/Apple proprietary standard, which could be changed. Dunno about RF, I would have thought a USB cable would have been more shielded than a PCIe riser or a SATA cable.

Yes, a particular kind with bespoke chips in each connector and very high requirements for shielding and signal integrity, as well as quite short maximum lengths.

As for the controller, TB3 is integrated into USB4, so that could theoretically be used, but you're still adding a relatively expensive high speed I/O controller just to transfer an already existing PCIe signal which could be transferred natively with a riser or u.2 cable. Why overcomplicate things by adding unnecessary hardware?

Distance. Thunderbolt can be run further distances than PCIe alone by itself.

To a degree - PCIe 3.0 can be run impressive distances with just passive risers. 4.0 needs tons of shielding and very high quality cabling though. TB3 tops out at, what, 3m, though in real life more like 2m due to there being no available cables? Active/optical overcomes that, but then you're looking at cables costing several hundred dollars, which kind of defeats the purpose.

Is there noticable RF inside a computer case?

Tons. The main reason PC cases are metal is to contain the RF noise they emit to avoid it interfering with other devices. High frequency signaling through board traces or wiring gives off noise; coils and power conversion gives off noise, your CPU and GPU operating at several GHz will give off noise, etc.

 

[Deleted member 24505]

I like NVME but please upgrade SATA speed, is what i vote for.

I have 4x gen 4 M.2 slots on my board but still use a 2TB WD green for backups.

 

[Chrispy_]

I have 4x gen 4 M.2 slots on my board but still use a 2TB WD green for backups.

You do realise that your 2TB WD green's sustained write speed is lower than the SATA3 bandwidth, right?
Apart from a brief period before the miniscule SLC cache runs out, the 2TB model tops out at ~430MB/s in a best-case, empty drive scenario.

I'm assuming you're talking about a 2TB WD Green SSD. If you are talking about spinning rust, even the SATAII interface at 3Gb/s is complete overkill, because the 2TB WD mechanical drive can't even saturate the original SATA1 bandwidth.

 

[lexluthermiester]

You do realise that your 2TB WD green's sustained write speed is lower than the SATA3 bandwidth, right?

You're missing the point. SSD's are limited by SATA3 speeds and there is only so much NAND you can fit on a 2280 NVMe PCB. The standard 2.5" and 3.5" form factor drives have MUCH more internal volume for NAND to be built into. The only reasonable way forward for inexpensive mass storage is to increase the SATA spec and make drives that fit into the existing 2.5/3.5" form factor. NVMe is a great answer for primary boot drives. It's a shitty answer for expanded internally mounted storage.

SATA needs updating! That updating needs to be done with the existing connector to maintain backward compatibility.

 

[eidairaman1]

You're missing the point. SSD's are limited by SATA3 speeds and there is only so much NAND you can fit on a 2280 NVMe PCB. the standard 2.5" and 3.5" form factor drive have MUCH more internal volume for NAND to be built into. The only reasonable way forward for inexpensive mass storage is to increase the SATA spec and make drives that fit into the existing 2.5/3.5" form factor. NVMe is a great answer for primary boot drives. It's a shitty answer for expanded internally mounted storage.

SATA needs updating! That updating need to be done with the existing connector to maintain backward compatibility.

Remote mounting is still an advantage as well

 

[Valantar]

You're missing the point. SSD's are limited by SATA3 speeds and there is only so much NAND you can fit on a 2280 NVMe PCB. the standard 2.5" and 3.5" form factor drive have MUCH more internal volume for NAND to be built into. The only reasonable way forward for inexpensive mass storage is to increase the SATA spec and make drives that fit into the existing 2.5/3.5" form factor. NVMe is a great answer for primary boot drives. It's a shitty answer for expanded internally mounted storage.

SATA needs updating! That updating need to be done with the existing connector to maintain backward compatibility.

You can get 4TB TLC m.2-2280 NVMe drives, or 8TB with QLC (meaning that if anyone was willing to pay for it, you'd get 6TB TLC drives too). Cost is the main roadblock here, not capacity. And, conversely, nobody in the world would be able to afford a 2.5" drive stuffed to the gills with flash. (And for those who can, they already exist, with u.2 interfaces.) NVMe already does what you're proposing an updated SATA standard would do, but with more flexibility and no need for new controllers.

 

[lexluthermiester]

You can get 4TB TLC m.2-2280 NVMe drives, or 8TB with QLC (meaning that if anyone was willing to pay for it, you'd get 6TB TLC drives too). Cost is the main roadblock here, not capacity.

And beyond 8TB? That is HDD territory for now but it's not going to be long before affordable SSD's of larger capacities start hitting the market and the NVMe form factor is not going to get the job done, nor should it be expected to. We need an update to SATA bandwidth and we need it soon.

 

[Tetras]

For me, it is kind of a no-brainer, since the drives (and I bet the controllers, too) are capable of being much faster than they are. I prefer to have compact, single-sided drives in a m.2 format. As manufacturers add more m.2 and drives get bigger (i.e. when 250GB isn't the smallest size ) I expect SATA to die. ATX 12VO may possibly have a role in killing it, because of the complexity it adds to a board to have SATA, but that's another thread entirely.

 

[Selaya]

And beyond 8TB? That is HDD territory for now but it's not going to be long before affordable SSD's of larger capacities start hitting the market and the NVMe form factor is not going to get the job done, nor should it be expected to. We need an update to SATA bandwidth and we need it soon.

by that time we probably have 352-layer NAND, you should easily be able to cram like 16tb of those on an m.2 if not more

your argument is invalid.

 

[Aquinus]

To a degree - PCIe 3.0 can be run impressive distances with just passive risers. 4.0 needs tons of shielding and very high quality cabling though. TB3 tops out at, what, 3m, though in real life more like 2m due to there being no available cables? Active/optical overcomes that, but then you're looking at cables costing several hundred dollars, which kind of defeats the purpose.

I actually think it's less than that for a passive TB3 cable, something like 1.3m. Those 3m cables are actually probably active. As I said, you can tell when the connector ends get warm because of the chips to handle converting the signals and whatnot on each end. Nobody wants to spend $80 on a cable, but that's what a 3m TB3 cable is likely going to cost you. Either way, it's not a great solution for an in-chassis interconnect. It's a far better solution for connecting things like docks or monitors with USB 3.1 or 3.2 hubs.

And beyond 8TB? That is HDD territory for now but it's not going to be long before affordable SSD's of larger capacities start hitting the market and the NVMe form factor is not going to get the job done, nor should it be expected to. We need an update to SATA bandwidth and we need it soon.

What's wrong with U.2 beyond cost again? Mind you, it's cost of the flash, not using U.2 that's the problem.

 

[ThrashZone]

Hi,
Cost of U2 is more than enough to laugh at it

 

[Deleted member 24505]

You do realise that your 2TB WD green's sustained write speed is lower than the SATA3 bandwidth, right?
Apart from a brief period before the miniscule SLC cache runs out, the 2TB model tops out at ~430MB/s in a best-case, empty drive scenario.

I'm assuming you're talking about a 2TB WD Green SSD. If you are talking about spinning rust, even the SATAII interface at 3Gb/s is complete overkill, because the 2TB WD mechanical drive can't even saturate the original SATA1 bandwidth.

I did not put it in because of the speed of it, it is just a backup device, so the speed of it matters not. I have 1x gen3 256gb for boot and 2x Gen 4 WD black SN850 for games so you really think i am bothered about a slow HDD for backup only. Possibly the only device i have the could benefit from faster SATA interface is a 2TB SATA SSD.

Some people still use big SATA HDD so they "might" benefit from a interface update, but not sure.