Optane SSDs available at Newegg - $2 per GB!

[VirtualLarry]

https://www.newegg.com/promotions/i...ntel/18-0594/homepage_300x120.jpg&icid=438852

 

[cbn]

That appears to be the only place the 800p is on sale at this point.

So far no reviews either.

P.S. When I click on the item link (even for the 58GB, which is still in stock) there is no product page.

 

[BonzaiDuck]

That appears to be the only place the 800p is on sale at this point.

So far no reviews either.

P.S. When I click on the item link (even for the 58GB, which is still in stock) there is no product page.

Yes -- confirmed -- the item page can't be found at the Egg. However, they're selling a 900P with 3D Xpoint as a PCIE device ready to put in the slot.

Looking at the Egg Specs, Seq-read/seq-write are 2500/2000 MB/s.

I may not be keeping up with the technical innovations. Why would I covet the Intel 3D XPoint 900P over my 960 Pro or even a 960 EVO? This is not a rhetorical question: a curious mind wants to know . . . .

 

[cbn]

Why would I covet the Intel 3D XPoint 900P over my 960 Pro or even a 960 EVO? This is not a rhetorical question: a curious mind wants to know . . . .

Take a look here:

https://www.anandtech.com/show/12136/the-intel-optane-ssd-900p-480gb-review/5

It is particularly strong in QD1 random read.

With that noted, the 800p only has PCIe 3.0 x 2 interface....so the Sequential Read will be slower than 960 Pro and 960 EVO. (Sequential write will predictably also be slower than the 250GB 960 EVO's psuedo SLC cache....but sustained write on the 960 EVO will be slower)

 

[IntelUser2000]

I may not be keeping up with the technical innovations. Why would I covet the Intel 3D XPoint 900P over my 960 Pro or even a 960 EVO? This is not a rhetorical question: a curious mind wants to know . . . .

It may not justify the extra price but it is the fastest drive. Even high end SSDs like the 960 Pro stutter for a few seconds when you do things like large file erase. That's because its overwhelming the internal DRAM buffer. It also does not suffer from performance degradation when the drive is close to full, and doesn't need TRIM because there's no distinction between clean and dirty state anymore.

Back to the OP:

Nice find!

On the page the specs are 1200MB/s, 275MB/s. However if you search for the drive you can see other numbers. One, is 1350/500. The other is 1200/600. Sisoftware leaks also show write speeds in the range of 500MB/s.

Since the marketing material is saying stable performance across capacity points, perhaps the Newegg number is a typo?

The $129 pricing may be preliminary too. A Canadian site has the 60GB(58GB? Who knows?) at $140, and 120GB and $265. That equals to USD of ~$110 and $206.

 

[BonzaiDuck]

Thanks for the replies.

Take a look here:

https://www.anandtech.com/show/12136/the-intel-optane-ssd-900p-480gb-review/5

It is particularly strong in QD1 random read.

With that noted, the 800p only has PCIe 3.0 x 2 interface....so the Sequential Read will be slower than 960 Pro and 960 EVO. (Sequential write will predictably also be slower than the 250GB 960 EVO's psuedo SLC cache....but sustained write on the 960 EVO will be slower)

That article explains it all. Now -- it makes sense. Not enough sense for me to give up my 960 Pro/EVO for an Optane in the current NVME usage for my system, but I can see where all this leads.

 

[cbn]

Product page at Newegg is now working.

Specs say Sustained Sequential Read/Write: up to 1450 / 640 MB/s.

 

[cbn]

One thing I have been wondering about Optane:

How does it compare to other NVMe SSDs for paging out during heavy browsing sessions?

Also, for say a system that is meant primary for web browsing (not CPU or GPU heavy tasks) and comes with only one 8GB stick of RAM how would adding a 16GB or 32GB Optane compare to adding another 8GB DDR4 stick?

8GB DDR4 RAM starts at $81
16GB Optane memory starts at $37
32GB Optane Memory starts at $60

(I am thinking the RAM would be the better choice for the time being, but with NVDIMM-P the choice between X amount of DRAM and X amount of 3DXPoint should get a lot harder....especially if Intel eventually uses smaller dies for the 3DXpoint)

 

[BonzaiDuck]

One thing I have been wondering about Optane:

How does it compare to other NVMe SSDs for paging out during heavy browsing sessions?

Also, for say a system that is meant primary for web browsing (not CPU or GPU heavy tasks) and comes with only one 8GB stick of RAM how would adding a 16GB or 32GB Optane compare to adding another 8GB DDR4 stick?

8GB DDR4 RAM starts at $81
16GB Optane memory starts at $37
32GB Optane Memory starts at $60

(I am thinking the RAM would be the better choice for the time being, but with NVDIMM-P the choice between X amount of DRAM and X amount of 3DXPoint should get a lot harder....especially if Intel eventually uses smaller dies for the 3DXpoint)

I'm not even completely confident of what I might assert in this response.

By "Optane memory," I'm assuming that the only devices relevant to your 2nd and 3rd options are M.2 sticks.

So your question is really about whether adding double or quadruple the size of such a device -- greater capacity but lower speed -- would trump adding RAM of lesser capacity and higher speed. And while the tradeoff there is real, it depends on the amount of data defining the tradeoff. And that would mean your question is subjective insofar as a usage profile is concerned. (?)

In my own case, I'm using an NVME M.2 for caching slower storage devices. The best I can muster in terms of performance is limited by the M.2 device's spec. A performance benchmark would show this:

relevant to an ~ 500 MB/s SATA source, and the NVME spec'd at ~ 3,000 MB/s.

If I double my RAM instead, or even just use the RAM I have, caching the SATA source might show in excess of 12,000 MB/s in a benchmark. But it depends on the amount of data read by the benchmark (or in a real world operation), and in fact, simply using a portion of a 2x8GB kit of RAM (I have 32GB) will show that result for tests of 1GB, 2GB, etc. to the limit of the amount of RAM used.

Aren't they eventually planning to replace volatile RAM with 3D Xpoint which allows memory data to be persistent? Or is that a misconception of mine? To be worth it, the speed would have to rival volatile RAM speed.

I can't be sure I'm not just talking smack here, because the only way I can keep up with all this these days is through generalized remarks about 3D Xpoint developments. I'm busy enough tuning and tweaking my system as it is, because every new Creators build update changes something to show up as red-bang event-log errors.

 

[IntelUser2000]

If I double my RAM instead, or even just use the RAM I have, caching the SATA source might show in excess of 12,000 MB/s in a benchmark.

RAMdisk is a waste of RAM because the latency is nowhere near taking data from RAM.

Take a look at this graph from PC Perspective: https://www.pcper.com/image/view/81097?return=node/67578

Optane achieves 10us latencies while RAMdisk is at 3us. That's not a noticeable difference because IO performance is limited by applications coding for platter HDDs, while not being fast enough to be like real DRAM.

Arguably the transfer rates are far greater, but you need the hassle of moving your data over to your RAMdisk every time the system needs to reboot.

Aren't they eventually planning to replace volatile RAM with 3D Xpoint which allows memory data to be persistent? Or is that a misconception of mine? To be worth it, the speed would have to rival volatile RAM speed.

If you look maybe 15-20 years into the future, yeah.

In the near future, it'll only be for servers. Cascade Lake SP's implementation tells us that its a requirement to have at least 1 DRAM DRAM module per Intel DIMM module. It's a hybrid approach so the DRAM can act as a write/endurance buffer.

How does it compare to other NVMe SSDs for paging out during heavy browsing sessions?

Also, for say a system that is meant primary for web browsing (not CPU or GPU heavy tasks) and comes with only one 8GB stick of RAM how would adding a 16GB or 32GB Optane compare to adding another 8GB DDR4 stick?

I have hands-on experience with a system like this. It's a Windows Vista system with 2GB RAM and Intel 520 SSD. You can easily tell when the system is paging. When the system boots and loads up its very close to 2GB. It's slower than an HDD system with sufficient memory.

What you are saying may benefit over an SSD system if you moved to a pure Optane drive, but using it as Optane Memory, you'd only get an SSD-like experience.

 

[cbn]

What you are saying may benefit over an SSD system if you moved to a pure Optane drive, but using it as Optane Memory, you'd only get an SSD-like experience.

So maybe ChromeOS or (ideally) some other small footprint Linux distro on a small optane drive + X amount of RAM vs. ChromeOS or (ideally) some other small footprint Linux distro on a small eMMC drive + amount of RAM?

 

[BonzaiDuck]

RAMdisk is a waste of RAM because the latency is nowhere near taking data from RAM.

Take a look at this graph from PC Perspective: https://www.pcper.com/image/view/81097?return=node/67578

Optane achieves 10us latencies while RAMdisk is at 3us. That's not a noticeable difference because IO performance is limited by applications coding for platter HDDs, while not being fast enough to be like real DRAM.

Arguably the transfer rates are far greater, but you need the hassle of moving your data over to your RAMdisk every time the system needs to reboot.



If you look maybe 15-20 years into the future, yeah.

In the near future, it'll only be for servers. Cascade Lake SP's implementation tells us that its a requirement to have at least 1 DRAM DRAM module per Intel DIMM module. It's a hybrid approach so the DRAM can act as a write/endurance buffer.



I have hands-on experience with a system like this. It's a Windows Vista system with 2GB RAM and Intel 520 SSD. You can easily tell when the system is paging. When the system boots and loads up its very close to 2GB. It's slower than an HDD system with sufficient memory.

What you are saying may benefit over an SSD system if you moved to a pure Optane drive, but using it as Optane Memory, you'd only get an SSD-like experience.

RAM Cache is not the same as RAM Disk. There are discussions and criticisms per Windows own cache and the use of RAM under a program like PrimoCache, but someone else here ran some serious benchies concluding a mild advocacy for the software -- certainly for gamers.

With that software, you have the choice under various situations -- RAID, AHCI, NVME or a mix of these configurations with more than one controller simultaneously as desired -- to cache slower media like SATA HDDs or even SSDs to either NVME, RAM, or a two-tiered combination thereof. Of course, you could cache one or more HDDs or HDD RAID arrays to an SATA SSD, but now we have NVME.

Recently, Maximum PC featured an evaluation of RAM configurations and choices, together with a little primer about latencies and speed. They suggested that 16GB was about as much as anyone could use, but they didn't mention caching. Frankly, I went forward and purchased a second kit of 3200 CL-14 TridentZs, partly from curiosity. Works for me! I cache my NVME (Pro) drive to RAM, and an SATA SSD to a smaller NVME (EVO).

Even G.SKILL support wouldn't guarantee full-spec usage of two identical but separate RAM kits, but it's been through the ringer with HCI Memtest, and there's no problem with it. [On that count, they said the would support the RAM clocked down 200Mhz, but I'm running at full spec.]

Well -- folks could lose patience with me -- but I posted the Anvil benchmark with only NVME_SSD-caching. Here it is with a second tier of RAM-caching, which Romex support vouches as a perfectly fine configuration:

I would expect the random reads and writes to be much better with an Optane NVME.

Just as an afterthought, even with 16GB RAM with maybe 6GB allocated to PrimoCache, I would get similar benchies. But raising the allocation to 16GB, there would be few instances wherein large file sizes or benchtest settings would overwhelm the cache.

 

[IntelUser2000]

I would expect the random reads and writes to be much better with an Optane NVME.

It won't be.

Remember what I told you about RAMdisk? The situation is the same. You still need to load data into RAM to take advantage of it, whether its done dynamically as in case of a RAMdisk, or pre-loaded as in case of RAMdisk.

The Random Read 4K picture you linked states the RAM caching results in 4.3us, 235K IOPS, or 919MB/s. That's in the range of 3us I stated above and the result PCPer got. If you read their article they said it was the best result.

Caching limits the total, maximum throughput to performance of the cache, minus the overhead. If you RAM-cache an Optane device, you'd still get mostly the same figures, because the caching is done by RAM.

To tell you how fast DRAM is without such limitations, a modern Intel system gets 50-60ns in latency. That's about 100x faster. The RAM-cache seems to be going through the NVMe interface as well.

RAM-caching an Optane 900P is questionable. Because the 900P is already limited by decades-old, HDD-oriented coding techniques. Making it faster won't help.

 

[whm1974]

Am I the only one who thinks that Optane is better used for NVMe SSD instead of memory once prices go down and sizes go up?

 

[BonzaiDuck]

. . . . If you RAM-cache an Optane device, you'd still get mostly the same figures, because the caching is done by RAM. . . . .

Actually, I stand corrected and you are right. A better-performing source device, in this case an NVME which has better 4K performance than what you'd expect from non-Optane NVME drives, would not make that much of a dent because -- as you say -- the performance is mostly determined by the caching device and not the source.

I've seen this in a comparison of a source-drive SSD compared to a good HDD cached to RAM for either source. Can't believe I fooled myself with that one . . .

 

[cbn]

With the Optane memory 32GB being $60 now, I wonder how long till the 58GB 800p is $99?

(Both 32GB Optane memory and 58GB 800p share the same controller. The main difference is that 32GB Optane has two packages with one die each and the 58GB 800p has two packages with two dies each).

P.S. With that noted, I have to admit I would have been a lot happier to see the Carson Beach 64GB Optane SSD even with premium pricing. (Carson Beach, unlike Stony Beach and Brighton Beach, is PCIe 3.0 x 4)

 

[cbn]

Am I the only one who thinks that Optane is better used for NVMe SSD instead of memory once prices go down and sizes go up?

I still like the idea of systems using the Optane memory with Hard drive (particularly if Intel ever allows Pentium and Celeron to do this....20th Anniversary Celeron?).

But actually of even more interest is using 58GB Optane 800p SSD (perhaps with larger page file?) in systems where there is only 8GB RAM.

 

[BonzaiDuck]

I still like the idea of systems using the Optane memory with Hard drive (particularly if Intel ever allows Pentium and Celeron to do this....20th Anniversary Celeron?).

But actually of even more interest is using 58GB Optane 800p SSD (perhaps with larger page file?) in systems where there is only 8GB RAM.

IntelUser2000 slowly made me aware of a blind spot I have in imagining how 3D XPoint will fit into the classic pyramidal hierarchy of computer storage devices: Expensive, low capacity, high speed storage at the top of the pyramid; cheap, high capacity and low speed storage at the bottom. At the top, we have CPU registers, L1, L2 and L3 cache. At the bottom, the lowly electromechanical hard disk with capacity in terabytes.

So one can ask questions that I still didn't have clarified for me by an Intel video promoting 3D Xpoint.

What is the advantage of having non-volatile memory, such as you'd anticipate with an application of 3D Xpoint?

What is the comparative speed of DRAM versus 3D Xpoint devices?

How does 3D XPoint fit into the pyramid? Or will it replace DRAM?

The 900P SSD is just 3D Xpoint used at a lower point in the pyramid like any NVME drive of 3D NAND.

Because I quickly came on board with the idea of using NVME as a caching device for slower storage, I couldn't much imagine how it would be used otherwise. 3D Xpoint fulfills three things: density, speed, and non-volatile persistence.

But I'm trying to imagine how a computer would work with a persistent RAM replacement. 3D Xpoint seems to muddy the waters of the old von Neumann machine and its components. Or does it?

 

[VirtualLarry]

I still like the idea of systems using the Optane memory with Hard drive (particularly if Intel ever allows Pentium and Celeron to do this....20th Anniversary Celeron?).

I know that I've commented about this before, but it seems to me, that "Optane Memory" (cache) is most suitable to budget rigs, that would still utilize a HDD for storage purposes, rather than go all-SSD like a more expensive system. I really think that Intel missed their target market completely with Optane.

 

[BonzaiDuck]

I know that I've commented about this before, but it seems to me, that "Optane Memory" (cache) is most suitable to budget rigs, that would still utilize a HDD for storage purposes, rather than go all-SSD like a more expensive system. I really think that Intel missed their target market completely with Optane.

Do you think they're trying to alter the paradigm? It's still not clear to me how it fits in, other than as an NVME device. And -- sure -- the accelerated HDD idea is all I had in mind for any NVME devices that weren't holding the boot-system volume.

And I wonder if Intel already has some thoughts about wider uses. They just aren't talking about it.

 

[cbn]

How does 3D XPoint fit into the pyramid? Or will it replace DRAM?

If placed on a DIMM the latency will higher than DRAM, but it will be persistent.

If used on a NVDIMM-P it will work alongside DRAM:

https://www.anandtech.com/show/11238/ddr5-to-double-bandwidth-over-ddr4-specification-due-next-year

Another specification that JEDEC plans to finalize in 2018 is the NVDIMM-P that will enable high-capacity memory modules featuring persistent memory (flash, 3D XPoint, new types of storage-class memory, etc.) and DRAM. The capacity of today’s NVDIMM-Ns is limited to the capacity of regular server DRAM modules, but the NVDIMM-P promises to change that and increase capacities of modules to hundreds of GBs or even to TBs. The NVDIMM-P is currently a work in progress and we are going to learn more about the tech in June.

What is the advantage of having non-volatile memory, such as you'd anticipate with an application of 3D Xpoint?

Here is a video where I found some pretty good info on the topic. (I didn't include comments from Oracle yet, but if you listen to the video he was pretty excited)

Of course, that talk is about server and workstation.

 

[IntelUser2000]

Am I the only one who thinks that Optane is better used for NVMe SSD instead of memory once prices go down and sizes go up?

Memory as in DRAM-replacement or Memory as in how Optane Memory works?

It's still not clear to me how it fits in, other than as an NVME device.

A pure Optane NVMe device makes the least sense to me. Aiming for the extreme section of the market usually ends up being much more successful. Optane Memory, while being too small for an SSD, will work because the absolute cost is low, and can be paired with HDD based systems that are still quite common. Intel DIMMs, make sense because it fully unleashes the capabilities of the 3D XPoint media. Even if its much higher priced than NAND SSDs, its still dirt cheap compared to DRAM and has the performance to back it up.

Optane SSDs need to be high enough capacity so it can fit Windows and most applications. So the price point is quite high making it unreachable to most people, while the media is hampered by the NVMe interface.

What is the advantage of having non-volatile memory, such as you'd anticipate with an application of 3D Xpoint?

What is the comparative speed of DRAM versus 3D Xpoint devices?

3D Xpoint seems to muddy the waters of the old von Neumann machine and its components. Or does it?

The 3D XPoint media in Optane devices are lot faster than the NVMe interface. The low queue depth random performance is identical between a $44 Optane Memory device and a $600 Optane SSD 900P. The differences are capacity and sequential performance and the latter is due to not having enough parallelism.

You'll see Optane-based DIMMs on servers by end of this year. In fact it was officially announced last year: https://newsroom.intel.com/news/int...pcoming-intel-xeon-processor-scalable-family/

It was supposed to be coming with Skylake, but it had some issues and they delayed it.

Picture of early sample Intel DIMM: http://www.legitreviews.com/wp-content/uploads/2016/01/intel-optane-512gb-ssd.jpg

Yes, that's a 512GB sized DIMM. Intel DIMMs can't work alone. It requires at least 1 DRAM slot per 1 Intel DIMM slot. DRAM is used for endurance and write caching. For reads, its expected to be not significantly slower. Intel DIMMs are the product that was expected out of 3D XPoint ever since its introduction back in 2015. It's not just capacity and non-volatility, but being significantly cheaper per GB than DRAM.

For PCs though, I expect it won't affect us for a decade. The amount of changes required are immense. We aren't talking about new Windows service packs or, firmware updates, or software versions. We're talking changing the mindset of the programmers that were used to one way of doing things for 50 plus years.

 

[BonzaiDuck]

Perhaps, then, we'd have a decade to prepare ourselves. Basically, it looks as though you're saying that the Intel DIMMs move a caching arrangement higher in the pyramid, and that the combination would make up for any speed shortcomings of the 3D XPoint DIMMs.

But I'm still trying to wrap my brain around the persistence aspect. With that type of capacity -- density -- you wouldn't need any slower storage devices except for uploading OS, programs and data into the XPoint DIMMs?

Maybe I should just check myself into a nursing home and call Dr. Kevorkian . . . .

 

[IntelUser2000]

I think Cascade Lake's Intel DIMM support is a great start.

Market-wise, I think the biggest benefit will be towards enterprise though, and in a more traditional way. The Intel DIMMs are somewhat flexible and it can work as NV storage just as it can work as large capacity DRAM. Or I believe even a combination of the two.

For YEARS they've been trying various buffer DIMMs and buffer chips to increase memory capacity because that's what enterprise requires. Hence the split into Xeon E5 and E7's. The buffer chips add latency and cost, but give you twice the capacity, at 3TB with 128GB DIMMs.

With Intel DIMMs, they can get the capacity benefit, but at lower cost and less power used due to not using buffer chips. Since buffer chips slowed it down anyway, 3D XPoint DIMMs won't be a loss there either. E7 platforms were used for in-memory compute applications, and persistence will make it even better. That should trigger an upgrade cycle for enterprise.

Outside of Enterprise, the persistency and lower cost should allow new usage models.

 

[whm1974]

I'm wondering if AMD and Intel are planning on release CPUs and chipsets with higher amounts of PCIe lanes so consumer level hardware can use 8x PCIe NVMe SSDs. PCIe v5 x8 should make for a really fast SSD for 4K video editing/rendering and tasks that need really high performance storage?