3D XPoint memory chip samples "just around the corner"

[HEXUS]

However it may be 12 to 18 months before mass production commences.

Read more.

[Michael H]

3D XPoint looks very exciting for video production at 4k and beyond. Imagine being able to record RAW onto a 3D XPoint module, drop it straight into an editing rig and access footage at near RAM speeds.

[qasdfdsaq]

3D XPoint looks very exciting for video production at 4k and beyond. Imagine being able to record RAW onto a 3D XPoint module, drop it straight into an editing rig and access footage at near RAM speeds.

With SATA or NVMe your interface is the limitation already, with DIMMs .... well you have to invent some sort of hot-pluggable DIMM interface first. And that'd require new mainboards, new processors, new sockets, and new RAM. Which isn't going to happen anytime soon.

[l8oss]

micron coming backkk

[lumireleon]

....delivering more than 95,000 IOPs at a 9ms latency, last time I checked the Intel SSD 750 can do 290,000 IOPs write but the details of latency 'milliseconds' is what I don't have.

[MrJim]

....delivering more than 95,000 IOPs at a 9ms latency, last time I checked the Intel SSD 750 can do 290,000 IOPs write but the details of latency 'milliseconds' is what I don't have.

I think the figure of 95,000 IOPS is per chip - SSDs comprise many chips...

[enemys]

Shouldn't the latency be in microseconds (µs), instead of milliseconds (ms)? It's just a typo, but a hugely important one.

[miniyazz]

My understanding was that SSDs had latency around the order of 0.1ms, not the 73ms mentioned in the article, with HDDs worse by a factor of 100 (latencies of 10-20ms). 73ms would be really slow (think running your operating system from a DVD drive slow) unless I'm missing something here.

[whatif]

Ah, presales hype and spin.
Until they hit the market, we will probably be seeing a lot of typos and unreferenced figures.
It is just like how journalists can take only part of a sentence out of a politicians speech and use that to make a point. But how do we know the context that part of a sentence had in the rest of the speech, apart from how the reporter presents it.

[qasdfdsaq]

Shouldn't the latency be in microseconds (µs), instead of milliseconds (ms)? It's just a typo, but a hugely important one.

No. Milliseconds is correct. We're talking about enterprise workloads here, with dozens of commands in a queue, not one command at a time like on desktops. Some SSDs can have response times over one full second under these loads.

Intel's P3608 datacentre NVMe SSD averages over 20 milliseconds in these workloads.

http://www.storagereview.com/intel_s...vme_ssd_review

....delivering more than 95,000 IOPs at a 9ms latency, last time I checked the Intel SSD 750 can do 290,000 IOPs write but the details of latency 'milliseconds' is what I don't have.

Under best case conditions, perhaps. In reality it does 55,000 IOPS read and 8,000 IOPS write in IOMeter 4K random tests.

My understanding was that SSDs had latency around the order of 0.1ms, not the 73ms mentioned in the article, with HDDs worse by a factor of 100 (latencies of 10-20ms). 73ms would be really slow (think running your operating system from a DVD drive slow) unless I'm missing something here.

What you are missing is this is not a consumer desktop product so you cannot compare it to your consumer desktop products under consumer desktop loads.