Seagate introduces HDDs as fast as SSDs

Well, as fast as a SATA SSD, under certain benchmark conditions. But interesting, nonetheless.


Point of note…

It should be noted that those are transfer rates. Spinning media can’t match an SSD for random reads and writes. Read/write IOPS for Exos 2X18 are listed at 304/560 while a SATA SSD’s read/write IOPS can top 100k/90k.

Still, it clobbers the single actuator-based HDDs. Seagate’s 20TB Exos X20 also has a SATA3 interface and 7,2000 rpm spin speed, but its maximum transfer rate is just 270MBs, half the speed of the Exos 2X18.

On a related note…
I wonder when there’ll be better 8TB SSD’s available?

The only one that’s been available for a couple of years now is the Samsung 870 QVO SATA 2.5" (MZ-77Q8T0), but there doesn’t seem to be signs of any others. :person_shrugging:

Unfortunately it’s rubbish for mass read/write usage. Two thoughtful Amazon reviewers:

This SSD is a “Cache + QLC SSD”. This consists of a total of 3 different memory types. In terms of write rates, 2 fast stores (cache) and 1 slow storage (storage).

1 GB LPDDR4 cache: 530 MB/s write rate
42 GB SLC cache: 530 MB/s write rate

8000 GB QLC storage: 80 MB/s write rate

The data is written first to the cache and then to storage. If a lot of data needs to be written and therefore the cache is full, the write rate breaks down to 80 MB/s.


Samsung calls this technology 4bit MLC but what it really is QLC (quad-level cell). In short SLC flash (single-level cell) is better than MLC flash (multi-level cell) but more expensive. MLC is better than TLC (triple-level cell) and TLC is better than QLC (quad-level cell). QLC is lot slower compared to TLC.

The advertised speed of 550 MB/s read and 520 MB/s write can only be achieved through SLC cache which is only 42 GB on the 1TB model. After the cache is full, performance drops significantly, with sequential writes of only 80 MB/s. This is the same speed a good hard disk can offer for a lot less than what you are paying for this QLC SSD. Not value for money at all.

You are better off spending extra and get yourself TLC ssd like the crucial mx500 which offers speed above 500 MB/s CONSISTENTLY. That all being said, I would buy this QLC SSD if it was lot cheaper than it’s TLC competitors. But it is not really cheap and I value performance.

1 Like

It will be very interesting to see what the reliability of these new Seagate drives are. Double the actuators means more moving parts.

True, although the voice-coil motors used in hard drives are extremely reliable. I think it is far more likely that something else (head crash, blown capacitor, etc.) will fail before the actuator.

My biggest question about this is what system software will need to do in order to take proper advantage of such a device.

If you have a SAS drive, then each actuator assembly acts as a separate drive. So you can treat your 18TB drive as two 9TB drives and leave it at that. Maybe run RAID-0 across the two LUNs in order to get (effectively) a double-speed 18TB drive.

If you have a SATA drive, however, LUNs aren’t available, so the drive partitions itself, with the first 9TB belonging to one actuator and the second 9TB belonging to the other. If you format it as one large partition, your file system is going to need to distribute the data evenly to the two halves in order to maximize throughput.

Or you may have to partition it into two 9TB partitions (first- and second-half), so each partition belongs to a separate actuator. Then either treat them as separate normal-speed drives or use RAID-0 to stripe data across the two in order to get double-speed behavior.

Either way, it’s going to be a bit weird for system administrators until some best common practices are developed for this kind of hardware.

1 Like

I’ve been thinking about this a bit more and it occurs to me that if the two actuators’ data are being used to form a striped data set (e.g. RAID-0), then the heads are going to pretty much move in unison together.

If that ends up doubling the data-rate (by combining data from each actuator over the SATA interface), then why can’t drive makers to that today with a single actuator?

If a drive has 4 platters (8 heads), then why can’t they today stripe data across all 8 heads. Effectively creating a system where blocks are identified (internally) by track and sector, with each block’s data striped across all the heads. This should multiply the data throughput by the number of heads, compared to each block existing entirely on a single surface (and would probably cause all but the slowest drives to quickly hit the SATA bandwidth limit).

If they’re not already doing this, I wonder why not. And I don’t think they are, because if they were, then there would be no speedup from using dual actuators.

I wonder if I should go and apply for a patent now. :grin:

1 Like