Intel's 3D XPoint-Powered Optane DC P4800X 'Cold Stream' NVMe SSD Leaks

[Guest]

Multiple leaks in Chinese-language media apparently outline the performance specifications Intel's first 3D XPoint-powered SSD, the Optane P4800X Series Cold Stream PCIe Add-In-Card.

Intel's 3D XPoint-Powered Optane DC P4800X 'Cold Stream' NVMe SSD Leaks : Read more

 

[BoredSysAdmin]

We were promised a revolution in SSD, what we got is are mild improvements

 

[problematiq]

This is more of a "slow start, fast finish" style of launch. After a/some fab(s) get finished I bet the gap will widen out quite a bit.

 

[Geekwad]

We were promised a revolution in SSD, what we got is are mild improvements

The drive writes per day? I know it may not apply to you, but it's impressive.

 

[dhemp]

I figured 3 year warranty with drive writes per day and total petabytes written numbers. Also I get a resizing video(ad?) in the text on firefox that makes reading the articles terrible! I'm seriously considering an adblocker just to stop those. Otherwise good article; good to see latencies moving toward the dram scale!

 

[dstarr3]

We were promised a revolution in SSD, what we got is are mild improvements

This is data center gear. There've been huge performance gains in recent generations in terms of functionality that the enterprise sector needs. Consumer SSDs haven't seen this kind of "revolution" because consumers don't need to write terabytes a day. And speed isn't such a concern since most consumer SSDs are bottlenecked by the SATA III interface anyway.

 

[George_180]

I've been reading Tom's Hardware for a long time. It has good information, good reviews, etc., but to be honest, all these auto-starting videos are getting REAL old. Not just auto-starting, but on this page, there is one that auto-sizes as I try to scroll down the page, basically causing me to lose my place. Then, it goes back and forth between the center of the page to the bottom right. I thought that REAL tech sites understood long ago that this nonsense was a no-no. If it keep up, I'll move on to a different tech site. This one is getting too annoying to visit.

 

[problematiq]

We were promised a revolution in SSD, what we got is are mild improvements

This is data center gear. There've been huge performance gains in recent generations in terms of functionality that the enterprise sector needs. Consumer SSDs haven't seen this kind of "revolution" because consumers don't need to write terabytes a day. And speed isn't such a concern since most consumer SSDs are bottlenecked by the SATA III interface anyway.

SATA interface SSD's are starting to fade away in desktops. We are seeing a jump to M.2 and PCI-E 4x which has much better bandwidth. The problem with getting faster and faster SSD's is that, while we have bandwidth room on the interface, we are going to choke up the pipe between the north bridge and the CPU. The latest chipset series sped up that bus from 4 lanes of DMI 2.0 to 4 lanes of DMI 3.0 for a total pipe of 3.938 GB/s, yes that's right, DMI 3.0 speeds are the same as PCI-E 3.0 o.o .

 

[gigabob]

Wait for the price. In an enterprise server, with this endurance, I expect a full suite of RDMA tools for data base acceleration. This seems....limited.

 

[teamninja]

I bet it will cost consumers 5k for the first ones out if not more

 

[littleleo]

If these specs are real and not fake news it will depend on the pricing as to how much it will effect the channel. If it is cheaper than the current brands/models yet has this level of performance I'd expect resellers to sell the heck out of these and only go to the other brands because of shortages. Other SSD brands will be forced to by the Intel/Micron 3D Xpoint stuff or try and develop their own version (good luck with that).

 

[teamninja]

lolololol Littleleo no lol what are you even saying Intel is currently being bottleneck by their own interface and connectors. There are many out there currently with similar speeds with the exception of drive writes per day they maybe lower then intel Nothing big will really change

 

[Ewitte]

The latency is nearly 1000 (read)/ 250 (write) times faster with MUCH higher endurance what isn't a revolution in that? The reason PCIe drives currently do not feel much faster than SATA drives is qd1 performance is still the same at 10000-12000 iops I've seen these shown over 90000.

 

[Shujee]

Wait a minute. I may have missed the obvious, but isn't it that Samsung 960 series SSDs (http://www.samsung.com/semiconductor/minisite/ssd/product/consumer/ssd960.html) have BETTER read/write speeds (3500/2100 MB/s compared to 2400/2000 MB/s of this monster)? I mean what's the point here?

(Just talking about speed. I know other factors such as endurance and latency are equally important).

 

[laststop311]

I would love a little 375GB m2 optane in 1 of the 2 m2 slots on all boards now. Perfect size for full operating system insteall + your most used applications like chrome and winrar and your productivity appls like microsoft office, photoshop video editing software and if u game have steam installed on there and like hearthstone and league of legends and a couple other of ur heavily played game. Use a 2nd 1TB samsung 960 m2 to store pretty much all of your other games that you even consider playing. Than have a couple 2.5" sata III el cheapo 2TB drives for 500 each to hold your torrenting activities this way you never see the disc overloaded message in your utorrent.

 

[laststop311]

PROBLEMATIQ not all m2 slots are run thru the PCH. Some have their lanes directly connected to CPU. so there is no DMI latency or bandwidth problem.

 

[computerguy72]

Shujee - this device is more enterprise oriented but I would still bet in every day usage scenarios the p4800x would crush a 960. That 4k performance is an order of mag higher. I have two 960's in raid 0 right now and my average rate is nowhere near the max. The negligible latency and near infinite endurance could expand it's use cases considerably as well. For instance near instant cold boots.

 

[remosito]

500k random writes at those latencies? My Database Server would love a raidable version of this

 

[Nintendork]

Many of the enterprise class nand-SSD's (not from intel) have an endurance of 4-8PB.

 

[Nintendork]

@EWITTE
You're wrong.

It's not 1000/250x faster.

It's 100-250microseconds (0.1-0.25ms) the typical 2-6ms Qos QD16 latency

That's 20x less latency in that scenario.

 

[bit_user]

We were promised a revolution in SSD, what we got is are mild improvements

This is data center gear. There've been huge performance gains in recent generations in terms of functionality that the enterprise sector needs.

I disagree. With the trend towards smaller lithography and more bits per cell, NAND-based SSDs are actually getting slower. This is true of both consumer & enterprise drives.

That's why we need fundamentally different technology, like 3D XPoint. Had Intel not made such lofty claims at the outset, everyone would now be singing their praises for delivering such big improvements in performance & endurance.

Anyway, while I'm in awe of these specs, I will reserve my most effusive praise for the publication of testbench data on low QD read performance. That's the most perceptible performance attribute, for most users.

 

[bit_user]

Many of the enterprise class nand-SSD's (not from intel) have an endurance of 4-8PB.

Please compare apples to apples. You really need to differentiate between read vs. write-oriented enterprise SSDs. Size also matters.

Since this is 375 GB, let's compare with Intel's 400 GB data center drives.

Model      PBW    DWPD
----------------------
DC P3500   0.22    0.3
DC P3600   2.19    3.0
DC P3700   7.30   10.0

Source: http://ark.intel.com/compare/80996,82846,79624

You also need to take into account cost. I think Intel has been close to the best PBW/$, in the industry, at least for a given size & class of drive. Feel free to post corrections, if you've got em.

 

[Shujee]

Shujee - this device is more enterprise oriented but I would still bet in every day usage scenarios the p4800x would crush a 960. That 4k performance is an order of mag higher. I have two 960's in raid 0 right now and my average rate is nowhere near the max. The negligible latency and near infinite endurance could expand it's use cases considerably as well. For instance near instant cold boots.

Thanks. I see your point. I was talking from a desktop user's perspective (maybe because I'm building a gaming machine these days, lol). If I have to spend twice as much as 960 on a disk, I'd expect a performance gain of maybe 5-10x. If my rig boots in say 5 seconds with 960, I won't spend another $500 to make it 4 or 3 seconds. But that might be just me.

 

[jasonf2]

RAM Drives only hit 6000-7000 so I think it is ridiculous to somehow expect first gen stuff on pci 3.0nvme to be cranking out some order of magnitude increase of Mbps figure. 2500 up and down with lowered latency is pretty impressive out of a 350 GB drive. There is going to have to be some interface rework before we are seeing the full potential of this stuff. And yea this is going to be expensive. I think the biggest challenge that Xpoint is going to have is that unlike early SSDs the upgrade isn't going to provide a magical performance upgrade to three year old machines that make them seem like new machines.

 

[problematiq]

PROBLEMATIQ not all m2 slots are run thru the PCH. Some have their lanes directly connected to CPU. so there is no DMI latency or bandwidth problem.

M.2 use's 4 lanes of PCI-E. If you are using 16 lanes for a graphics card using say a 200 series board you are already using all the lanes direct to your CPU in which case it uses the 24 connected to your PCH.