Intel Demos 3D XPoint Optane File Copy At 2 GB/s

[Megalith]

Here’s Intel showing off their “crazy fast” memory, although the author of this article thinks the demonstration is suspect in some ways.

It is great to see XPoint / Optane technology being demonstrated again, but as far as demos go, this was not the best / fairest example that Intel could have put together. First of all, the 'NAND SSD' they are using is a Thunderbolt 3 connected external, which was clearly bottlenecked badly somewhere else in the chain (when was the last time you saw a 6 Gbit SATA SSD limited to only 283 MB/s?). Also, using SATA for the NAND example while using PCIe x4 NVMe for the Optane example seems a bit extreme to me.

 

[Galvin]

I'm sure resistive memory can go way beyond 2GB/s. The problem with that video is we have SSD's now that connect with M.2 that have speeds over 2GB/s. So I'm not all that impressed with the demo.

Be really impressed when we make a processor based off this stuff.

 

[tgom222]

Hmm, unimpressive.

Weren't they claiming a 100x performance boost?

Flash really isn't the problem anymore. It's the interface. We went from maxing out SATA III, skipped SATA express and on to Nvme on pci-e 3.0 x 4.

The problem is the first generation consumer drives on this new interface are already hitting 60-70 % of theoretical throughput.

Even if X-point is significantly faster, there isn't enough headroom on the interface to make a big difference.

 

[pxc]

Hmm, unimpressive.

Weren't they claiming a 100x performance boost?

Flash really isn't the problem anymore. It's the interface. We went from maxing out SATA III, skipped SATA express and on to Nvme on pci-e 3.0 x 4.

The problem is the first generation consumer drives on this new interface are already hitting 60-70 % of theoretical throughput.

Even if X-point is significantly faster, there isn't enough headroom on the interface to make a big difference.

It was a prototype and connected through a Thunderbolt 2 connection, which has about 2GB/s bandwidth each way. The performance target for the ****first**** Optane drive is 6GB/s IIRC.

The 1000x number isn't the maximum copy bandwidth. XPoint can read and write down to the single byte level, which is much faster than reading and writing a whole block like with NAND flash. That won't show on a file copy benchmark. In target applications use, Optane drives should have insane IOPS even compared to NAND flash.

 

[gxp500]

Hmm, unimpressive.

Weren't they claiming a 100x performance boost?

Flash really isn't the problem anymore. It's the interface. We went from maxing out SATA III, skipped SATA express and on to Nvme on pci-e 3.0 x 4.

The problem is the first generation consumer drives on this new interface are already hitting 60-70 % of theoretical throughput.

Even if X-point is significantly faster, there isn't enough headroom on the interface to make a big difference.

I don't think the interface is really the issue here, is anyone complaining "i wish i could move files at 10gb/sec"?
If you compare a sata ssd and an m2 ssd why doesn't windows/games load any faster on the m2 ssd? You have 4x the read speeds yet load times are identical. Where's the bottleneck here, cpu/chipset?

 

[aaronspink]

Hmm, unimpressive.

Weren't they claiming a 100x performance boost?

Flash really isn't the problem anymore. It's the interface. We went from maxing out SATA III, skipped SATA express and on to Nvme on pci-e 3.0 x 4.

The problem is the first generation consumer drives on this new interface are already hitting 60-70 % of theoretical throughput.

Even if X-point is significantly faster, there isn't enough headroom on the interface to make a big difference.

No we aren't at 60-70% of theoretical throughput, we are significantly below that. Looking at rather synthetic high queue depth sequential benchmarks is and has always been pointless. Any benchmark doing anything beyond QD=1 sequential performance tests is just wasting time and even then its hardly representative.

The vast majority of all data movement that matters is of an effectively random nature. Pretty much no one in the storage industry cares about sequential performance at the device level, they care about the random performance.

Also you are greatly confusing device and component performance metrics. Just getting to and back from a PCIe device takes several uS. Even if Xpoint has zero latency, the device performance would be limited by the transaction latency of PCIe. That's why one of the primary pushes with Xpoint is as memory augmentation off of the dram interfaces.

 

[drescherjm]

I don't think the interface is really the issue here, is anyone complaining "i wish i could move files at 10gb/sec"?
If you compare a sata ssd and an m2 ssd why doesn't windows/games load any faster on the m2 ssd? You have 4x the read speeds yet load times are identical. Where's the bottleneck here, cpu/chipset?

I say the biggest bottleneck is the NAND. M2 uses the same NAND as SATA just a lower latency controller and interface with more internal raid0 however there is only so much you can do for low queue depth 4k.

 

[wgm3446]

Should have used the 'middle out' compression method.