Intel 8000p Optane offered in 16GB and 32GB sizes

[cbn]

Impressive performance considering the size:

http://www.guru3d.com/news-story/in...ered-in-16gb-and-32-gb-3d-xpoint-storage.html

We talked about Intel and Micron's 3D XPoint storage a couple of times already. 3D Xpoint memory technology is the successor to standard NAND flash memory with an ETA likely in 2017. Intel will call this Storage series the Intel Optane SSD and it will be available in form factors like M.2/NGFF, SATA-Express, PCI-Express (card).

They will makes use of the new NVMe protocol. An early prototype was already demonstrated at IDF and would offer up-to 5.5 times the throughput of NAND flash-based DC P3700 series SSDs, (engineering sample and thus early development). Compared to the queue depth of just 32 commands for AHCI NVMe offers command queue depth of a 65,535 commands. Micron is the co-developer of 3D Xpoint so you may expect Crucial branded SSD drives on this technology as well.

New information has been spotted (net necessarily exciting) as Intel will release two Optane Memory 8000p products based on the new technology at launch, a 16GB and 32GB 3D XPoint storage unit is planned based on M.2 2280 and 2241 form factors. The units would thus use PCI-E 3.0 with two lanes according to Taiwanese website Benchlife. The 32 GB model would reach 1600 MB/s on reads and 500 MB/s on writes whereas the 16GB model would reach 1400 MB/sec and 300 MB/sec. IOPS reads are 300K and 285K respectively with 120K and 70K on writes. To be honest if this is true, then the results do disappoint a little as even the NVMe based Samsung 960 Pro can already reach 3500 MB/sec and 2100 MB/sec on writes (we have a review on the Samsung 960 Pro 1TB next week btw).

Currently there is no information about the actual controller, however we do know that you'll need a Kaby Lake processor and compatible chipset to be able to run these Optane Memory 8000p units. I don't know, these might be this first models, but somehow I guess we all expected way more from the prognosed performance ? Next to that the 16GB and 32GB volume sizes are simply way too small to be interesting on the consumer end side of things. Have a peek at the leaked info by clicking on the thumbnail below.

 

[cbn]

https://www.pcper.com/news/Storage/Intel-Optane-XPoint-First-Gen-Product-Specifications-Leaked

Now I know there is a bunch of rambling around the net already. "Why so small?!?!". What I think we are looking at is Stony Beach - Intel's 'Application Accelerator" seen here:

What further backs this theory is that you'll note the PCIe 3.0 x2 link of that product in the above roadmap, which couples nicely with the upper end limits seen in the 32GB product, which is clearly hitting a bandwidth limit at 1.6 GB/s, which is the typical max seen on a x2 PCIe 3.0 link.

Now with the capacity thing aside, there is another important thing to bring up. First gen XPoint dies are 128 Gbit, which works out to 16 GB. That means the product specs for the 16GB part are turning in those specs *WITH ONE DIE*. NAND based SSDs can only reach these sorts of figures by spreading the IO's across four, eight, or more dies operating in parallel. This is just one die, and it is nearly saturating two lanes of PCIe 3.0!

 

[shabby]

Wake me up when i can buy this, until then im not impressed.
960pro review coming next week though.

 

[jpiniero]

Wake me up when i can buy this, until then im not impressed.

Well, according to the article you need a Kabylake processor AND a compatible mobo. Probably headed to laptop makers to create a very cheap hybrid drive.

 

[cbn]

Probably headed to laptop makers to create a very cheap hybrid drive.

I am also wondering if we see these being used as primary drives for the following laptops.

16GB for Kabylake Chromebook

32GB for Kabylake "Cloudbook"

 

[John Carmack]

Are they mistaken? Requiring a Kaby Lake CPU and chipset support for M.2 drives?

 

[cbn]

Considering the high Sequential Read performance per die (one 256 Gbit (16GB) Optane die can nearly saturate PCIe 3.0 x 2 on Seq. Read) I wonder what capacities we will see for the PCIe 3.0 x 4 Optane SSDs? (ie, Mansion Beach)

32GB, 48GB or 64GB at the low end?

And perhaps 128GB or greater for those that also want to saturate the Sequential write?

 

[VirtualLarry]

I assume that this is primarily for cache / acceleration of storage, and not storage in it's own right.

The sizes may be small, but if this is positioned as a RAM supplement / replacement...?

 

[cbn]

The sizes may be small, but if this is positioned as a RAM supplement / replacement...?

For "RAM Supplement" I wonder how well a 16GB or 32GB Optane SSD would work with a low cost 4GB RAM Kabylake laptop running Ubuntu? (Or Mint)?

Reason: Ubuntu is about 10GB less than Windows 10 64 bit in install footprint and thus would have more room for swap for any given Optane SSD size.

EDIT: Thinking about this more, perhaps if Optane is viable enough at the low end we might end up seeing the Kabylake equivalent of the 6W/4.5W Skylake 4405Y make into low cost laptops? (AFAIK Skylake 4405Y was never used in a laptop for whatever reason)

 

[TheELF]

For "RAM Supplement" I wonder how well a 16GB or 32GB Optane SSD would work with a low cost 4GB RAM Kabylake laptop running Ubuntu? (Or Mint)?

There is a version, planned for next year,that will plug straight into the(a) ddr4 slot,so cheap note/lap-tops won't have any real ram anymore.
Real curious about sizes and prizes though,if it's cheap it will be revolutionary having your whole system in mem.

 

[biostud]

there have been SSD caches too. Wouldn't mind a 32GB PCIe x4 cache for all my drives, or an SSD with optane cache.

 

[IntelUser2000]

Considering the high Sequential Read performance per die (one 256 Gbit (16GB) Optane die can nearly saturate PCIe 3.0 x 2 on Seq. Read) I wonder what capacities we will see for the PCIe 3.0 x 4 Optane SSDs? (ie, Mansion Beach)

32GB, 48GB or 64GB at the low end?

Capacities aren't much of a constraint as price. Capacity wise they can likely get it large as the largest non-stacked cell SSDs. They'll probably release big enough drive to be useful and competitive.

And perhaps 128GB or greater for those that also want to saturate the Sequential write?

The low sequential write is a bit strange. But its quite likely that Intel is going to use garbage-bin XPoint chips for SSDs/Caching products. I mean, people expect Intel's claims of 1000x performance/endurance to end up being in every single XPoint-based product, but that's not typical of Intel. They have more SKUs on CPUs and GPUs then people care to ever know. Are they going to sell one type of Optane product to people? No way. If its successful for them you'll see it covering all the way from $50 to $5000.

The dies going on future DRAM interface Optane products will be far better performance/endurance wise. I wouldn't be surprised if single top-bin XPoint die saturate PCIe 3.0 x2 or even x4. Of course, products based on that would be so much more costly which is why it makes sense to put it on a DIMM, which can unleash its full potential.

there have been SSD caches too.

Problem with SSD cache might be that the SSD cache itself needs a cache. Maybe not needing its own caches will make Optane caching products far more useful in reality.

Unless you are transferring files all the time, since the first decent SSD(Intel X25-M) the only real advancement for the vast majority of folks have been TRIM, when its properly implemented.

The real leap from SSDs would be:
-DRAM replacement, with full OS and application support. Which for consumers is realistically 2019+. 2019 probably for first consumer DRAM XPoint and 2022 for OS/application support
-If Optane caching is capable enough that Optane cache + HDD can roughly simulate low cost standalone SSDs

PCI Express Optane is not useful. Even 1/7 latency won't be noticeable. With proper implementation on DRAM interface, you can get the latency down to hundredth. The real usefulness would not be that your system is super snappy, but application/OS optimization to make a truly instant computing experience and relegate sleep modes and P states to legacy.

 

[dealcorn]

I was irked this product requires Kaby Lake until I heard the update from the scary smart, tin foil hat front. Brahmin Breezy sees this as DRAM replacement for ultra mobile markets. Trans-formative products are not necessarily high performance. http://seekingalpha.com/article/401...e93f33c3dbb5398f8e2edef2cf96bfa&uprof=46#alt1.

 

[cbn]

Knowing about RAM usage differences between the following browsers:

......And suspecting that a fresh install of Ubuntu (which today all the updates uses 5.1GB according to System Monitor) may take up even less space than ChromeOS in 2016 (which according to my research used 4.8GB (EDIT: 5.2 GB ) back in 2014) I have to think that maybe Ubuntu is better suited a low end Kabylake laptop with 16GB optane as primary storage.

The question is how well would Ubuntu's Firefox browser run compared to ChromeOS's Chrome browser if using a very weak Kabylake dual core? If the hypothetical 16GB Optane equipped laptop were using an atom quad core, then no doubt ChromeOS would win.....but with the dual big core I think Ubuntu has a much better chance now.

 

[IntelUser2000]

I was irked this product requires Kaby Lake until I heard the update from the scary smart, tin foil hat front. Brahmin Breezy sees this as DRAM replacement for ultra mobile markets. Trans-formative products are not necessarily high performance. http://seekingalpha.com/article/401...e93f33c3dbb5398f8e2edef2cf96bfa&uprof=46#alt1.

I can't imagine using something that's in the 10us range of latency to be a memory replacement - value system or not. Optane Memory 8000p is NVMe based, therefore likely caching to supplement slow storage, not slow memory. He does point out that a clever mechanism could adequately make it perform like a DRAM based system but cheaper. Perhaps that's the requirement of needing Kaby Lake. Maybe Intel will surprise us and we find out that we can indeed use Optane drives for supplementing DRAM, even those drives on the NVMe interface.

Another thing is Intel's roadmap itself. It's under the section, "System Acceleration". Same section where 40GB X25-M based controller and Z68's SSD-based acceleration was. To go from that to value DRAM-replacement is a leap.

He also suggests that eDRAM will be used as a side cache, and overall it'd end up saving cost. But the eDRAM itself is too expensive now to be in more than a handful SKUs.* So whatever cost savings brought on by low-DRAM + Optane is made up by the cost of eDRAM. The patents don't necessarily point to the things they'll do right away. It might be in the near future, like 5-6 years.

*It does sort of explain why Intel went for custom eDRAM route rather than HBM. People might have thought it was for small CPU performance boost and for their iGPUs. Maybe the relatively small, but very low latency nature compared to HBM/HMC is to act as a cache for Optane. That would mean far back as 2013 with Haswell GT3e they were readying for Optane.

 

[TheELF]

I can't imagine using something that's in the 10us range of latency to be a memory replacement - value system or not. Optane Memory 8000p is NVMe based

How much latency is the OS issuing the command to the nvme/ssd controller?How much latency for said controller to find the piece of data? How much for the data to get copied over to system ram?So that finally you get access to it wit low μs.
Optane is just an nvme replacement but 3dxpoint will skip all the steps I listed and the CPU will have access to the data with only the latency of the 3dxpoint "ram" .

 

[john3850]

What is happening between Intel and Microns 3DXPoint ram.
Micron may be buying Intel's stake in two IM Flash fabs.
http://www.eetimes.com/document.asp?doc_id=1261280

I read in few places that intel may buy out micron.
http://247wallst.com/technology-3/2...make-a-challenging-and-daring-bid-for-micron/

 

[VirtualLarry]

If Micron is at risk in any way from being swallowed by some bigger foreign state-back conglomerate (Samsung, LG, etc.), then by all means, Intel should buy them instead, and keep the technology centered in the USofA.
Micron would have been a much better acquisition than McAfee to be frank.

 

[IntelUser2000]

How much latency is the OS issuing the command to the nvme/ssd controller?How much latency for said controller to find the piece of data? How much for the data to get copied over to system ram?So that finally you get access to it wit low μs.
Optane is just an nvme replacement but 3dxpoint will skip all the steps I listed and the CPU will have access to the data with only the latency of the 3dxpoint "ram" .

8-10us is the expected latency for NVMe Optane drives. That's the number that you'll feel as a user. Software, interface, technology, everything.

I think they might still try, but if you are thinking of absolute value devices, they still use platter HDDs. You might use Optane to bridge the gap between Platter vs NAND, but Platter vs DRAM? The previous caching setups using NAND SSDs to supplement Platter drives never caught up to pure SSD setups. That's why I believe(in practice), Optane caching products like the Optane Memory 8000p will be used alongside platter HDDs. SSDs are still far more expensive.

Also, 3D XPoint is just the technical name for Optane. It's the same thing. You can call Optane SSDs 3D XPoint SSDs too. It's just the description for underlying technology.

 

[Arachnotronic]

If Micron is at risk in any way from being swallowed by some bigger foreign state-back conglomerate (Samsung, LG, etc.), then by all means, Intel should buy them instead, and keep the technology centered in the USofA.
Micron would have been a much better acquisition than McAfee to be frank.

Micron is losing money and can barely keep up technologically with Samsung and SK Hynix. Maybe not such a good fit for Intel