News Kioxia's Developing PCIe 5.0-Ready SSDs With New E3.S Form Factor

[bit_user]

High capacity HDDs are relatively slow because reading and writing smaller bits requires greater positional accuracy that can only be achieved at lower speeds.

If you want affordable high capacity HDDs, you'll have to put up with 100-200MB/s sequential transfer speeds.

Please don't post misinformation. There are plenty of high-capacity 7200 RPM drives.

Here's one where an 8-drive RAID6 achieved "2.311GB/s read and 869MB/s write". That works out to 289 MB/sec read, measured (not theoretical). The raw write speed is harder to work out, due to RAID, CIFS, and NAS getting in the way. Unfortunately, they didn't have a separate benchmark on single-drive performance.

https://www.storagereview.com/review/seagate-ironwolf-pro-16tb-nas-hdd-review​

(Or even all the way down to ~10MB/s when writing to SMR drives.)

Suffice to say: avoid SMR drives if you care about write speed and are doing pretty much anything other than copying huge files. However, as long as it's purely sequential, even SMR drives can write fast. And their read performance is as good as non-SMR drives.

 

[USAFRet]

I have 12 spinning drives in the house.

1 in a 12 year old low end Tosh laptop.
The other 11 are in or attached to the NAS - 51TB space.

All the other actual systems are SSD only.

 

[InvalidError]

Please don't post misinformation. There are plenty of high-capacity 7200 RPM drives.

In the context of 15000RPM multi-actuator HDDs, regular 7200RPM single-actuator HDDs are slow.

 

[bit_user]

In the context of 15000RPM multi-actuator HDDs, regular 7200RPM single-actuator HDDs are slow.

That comparison doesn't even make sense. The 15 kRPM drives are small and basically extinct, as you already pointed out. Platter densities have since increased past the point of canceling out their sequential performance advantage.

As for multi-actuator HDDs, let's be clear: it's 2. They're dual-actuator HDDs. We don't know how well it can scale, or whether 2 is the practical limit. And AFAICT they're based on 7200 RPM HDDs, so it's just a further improvement on that foundation, but at not insignificant added cost.

https://www.seagate.com/files/www-content/datasheets/pdfs/exos-2x14-DS2015-2-1912US-en_US.pdf

Still, I don't know what your point is. I'm not saying dual-actuator is bad, but it's pretty clearly a server/cloud-oriented technology, and its mere existence doesn't make single-actuator 7200 RPM drives any slower. Yes, it's there, if you need the performance and can afford the price premium.

Let's please not argue for the sake of argument. You said HDDs topped out at 100-200 MB/sec, and that was demonstrably incorrect. That was my only point and I've supported it with empirical data (which you should maybe try, if you want to minimize the risk of making further false statements).

 

[InvalidError]

That comparison doesn't even make sense. The 15 kRPM drives are small and basically extinct, as you already pointed out. Platter densities have since increased past the point of canceling out their sequential performance advantage.

15000RPM HDDs weren't meant for sequential workloads. The main motivation behind ludicrous spindle speeds was reduced worst-case rotational latency for random read/writes like high traffic databases. For most everyday uses, read/write latency is a far greater component of perceived performance than bandwidth as it still is with SSDs today: SSDs with higher random IOPS yield better overall user experience than SSDs with higher sequential bandwidth.

 

[bit_user]

15000RPM HDDs weren't meant for sequential workloads.

But sequential was @Kamen Rider Blade 's point! That user was clamoring for HDDs with more sequential throughput. We all know that SSDs rule at random performance.

 

[InvalidError]

But sequential was @Kamen Rider Blade 's point! That user was clamoring for HDDs with more sequential throughput. We all know that SSDs rule at random performance.

True, another reason why his desire for 15000RPM HDDs was misplaced.

 

[Kamen Rider Blade]

If I want low latency & Random IOPS, I'll go Optane for my OS Drive, that's the only point where low latency and IOPs really matter for me as a normal end consumer.

If I need bulk Sequential for Game loading, there's TLC / MLC M.2 NVME SSD's that can do the job.

But you got to remember, I'm comparing QLC/PLC sequential performance to HDD's.
If you up the performance of a HDD, the performance value of QLC/PLC startings getting shakier.
Whatever it takes to boost HDD performance.
Be it Multi-Actuator or 15K RPM 2.5" drives.

 

[bit_user]

Whatever it takes to boost HDD performance.
Be it Multi-Actuator or 15K RPM 2.5" drives.

"Whatever it takes"? So, you're willing to pay any price?

Well, I think we're all in agreement that you can scratch 15k RPM drives from your list - those appear to be gone and not coming back. However, the major technology you left off your list is EAMR (HAMR/MAMR), which should further boost bits per track, and therefore benefit transfer rates.

I doubt there's a scenario where HDDs ever catch SSDs, in linear transfer rates, but it's not inconceivable if HDDs can continue to increase density at current rotational speeds. I don't think EAMR will get us there, but it's set to take us to at least 80 TB/drive:

https://www.anandtech.com/show/1548...-denko-develops-hamr-platters-for-hard-drives

 

[Kamen Rider Blade]

"Whatever it takes"? So, you're willing to pay any price?

Well, I think we're all in agreement that you can scratch 15k RPM drives from your list - those appear to be gone and not coming back. However, the major technology you left off your list is EAMR (HAMR/MAMR), which should further boost bits per track, and therefore benefit transfer rates.

I doubt there's a scenario where HDDs ever catch SSDs, in linear transfer rates, but it's not inconceivable if HDDs can continue to increase density at current rotational speeds. I don't think EAMR will get us there, but it's set to take us to at least 80 TB/drive:

Seagate has claimed in their FAQ over their Multi-Actuator tech that it only costs the space of 1x platter to perform "Multi-Actuator".

That doubling of throughput / Read/Write is HUGE for them and can greatly benefit normal HDD users.

If they can get Triple, Quadruple, Pentuple, etc Actuators going, the throughput can be greater =D.

Each extra Actuator being a "Effective Multiplier" can lend to huge gains in HDD Sequential Read/Write throughput. Who doesn't want faster HDD's?

 

[USAFRet]

Each extra Actuator being a "Effective Multiplier" can lend to huge gains in HDD Sequential Read/Write throughput. Who doesn't want faster HDD's?

Noise, heat, power, price...

"normal HDD users"
More and more, "normal users" are going completely solid state. Any spinning drives may live in a NAS box, accessed across the LAN. Where the LAN interface negates any 'uberfast' spinning drives.

In my QNAP, there is a 480GB SSD and a selection of spinning drives. Due to the LAN interface in the middle, read/write from any of the other systems is the same, to either type of drives in the NAS.

With the presence of solid state that is much faster, I'm not seeing a compelling case for "Triple, Quadruple, Pentuple, etc Actuators " on a spinning drive.

 

[InvalidError]

If they can get Triple, Quadruple, Pentuple, etc Actuators going, the throughput can be greater =D

And then you end up with a 2TB HDD that costs more than a 2TB SSD, is still nowhere near as fast, uses several times more power primarily due to aerodynamic drag across the multiple head racks and that many more times more likely to have a head crashing into a platter and physically destroying your data.

 

[USAFRet]

"If we just build a double wall heat box, we can make the steam engine relevant again!"

 

[spongiemaster]

That doubling of throughput / Read/Write is HUGE for them and can greatly benefit normal HDD users.

If they can get Triple, Quadruple, Pentuple, etc Actuators going, the throughput can be greater =D.

Each extra Actuator being a "Effective Multiplier" can lend to huge gains in HDD Sequential Read/Write throughput. Who doesn't want faster HDD's?

Multiplying the throughput will result in negligible performance increases without a comparable decrease in average random access time. What makes SSD's so much faster than rotational drives is that access to data anywhere on the drive is nearly instantaneous compared to the average access time of platters. If you put a SATA SSD on a SATA 2 port and limit its throughput to about 265MB/s which right in line with the maximum sustained transfer rates of todays high capacity 7200 RPMs, the SSD is still going to kill it in benchmarks because throughput is almost never the limiting aspect of storage performance.

Because of the significantly higher sector density of today's 7200RPM drives vs the last generation of 15k drives, raw throughput is higher with a 7200 RPM drive. If you were to short stroke the 7200RPM drive to the maximum capacity of 15k drives (going from 16TB to 600GB would be a 96% reduction in platter usage) you probably would be able to reduce the average seek time enough to overcome the 2ms average latency advantage of the 15k drive while also keeping the heads at the very outer most edges of the platter with the highest throughput. The result would be that the 7200 rpm drive would not only have higher peak sustained transfers, it would even have comparable if not higher IOPS as well. Higher RPM is not the be all end all.

 

[bit_user]

Seagate has claimed in their FAQ over their Multi-Actuator tech that it only costs the space of 1x platter to perform "Multi-Actuator".

That's probably per-actuator.

If they can get Triple, Quadruple, Pentuple, etc Actuators going, the throughput can be greater =D.

Each extra Actuator being a "Effective Multiplier" can lend to huge gains in HDD Sequential Read/Write throughput.

The best-case is that you get linear scaling W.R.T. # of actuators. It's not like exponential, as you seem to imply.

A bigger concern might be how each actuator needs to compensate for the motion of the others. It seems quite likely they do, and could well be the greatest challenge in scaling up.

And at some point, it becomes cheaper and simpler to just have more drives than to try to pack even more actuators per drive.

Who doesn't want faster HDD's?

It's all about tradeoffs. Cost, power, reliability, longevity...

 

[InvalidError]

It's all about tradeoffs. Cost, power, reliability, longevity...

You omitted the single biggest one with regards to HDDs: cost per bit. It'll be quite a while before SSDs break even with no-frills HDDs on $/TB, though they are closing in for the killing blow on the sub-TB class.