Overclock.net banner

1 - 20 of 21 Posts

·
Registered
Joined
·
619 Posts
Discussion Starter #1
Intel's new 3D NAND SSDs are finally here, and they look like scorchers! However, they had better be cheap! 3D NAND is still more expensive than SSDs with regular NAND which doesnt make sense to me, they are supposed to be cheaper.
Quote:
The DC P3520 offers up to 365,000 random read IOPS and 22,000 random write IOPS, which is similar to the Intel DC P3500. The DC P3520 suffers a slight performance reduction due to the loss of parallelism that stems from denser 3D NAND (fewer packages).
Intel also announced its DC P3520 SSDs, but did not share any performance data other than to note that it will be the performance-oriented model. We unearthed a document in August of last year that correctly predicted the announcement of Intel's DC P3608 series, and it also included information on the DC P3520.


http://www.tomsitpro.com/articles/intel-3d-nand-dual-port-nvme-ssds,1-3217.html?_ga=1.125106633.2116075509.1436276413
 

·
Banned
Joined
·
2,915 Posts
Quote:
Originally Posted by tp4tissue View Post

where's cross point ... lame.....
biggrin.gif
I'm waiting for the same, they can pull my Intel 520s out of my cold dead hands, not dropping a single penny until XPoint and 8-lanes.
biggrin.gif


Definitely NOT buying that Intel 750, nope nevah.

Give us Optane 3D XPoint NOW.
tongue.gif
 

·
Premium Member
Joined
·
5,973 Posts
Quote:
Originally Posted by Computurd View Post

Intel's new 3D NAND SSDs are finally here, and they look like scorchers! However, they had better be cheap! 3D NAND is still more expensive than SSDs with regular NAND which doesnt make sense to me, they are supposed to be cheaper.



http://www.tomsitpro.com/articles/intel-3d-nand-dual-port-nvme-ssds,1-3217.html?_ga=1.125106633.2116075509.1436276413
You're paying for technology investment and new tools vs fabs that have many depreciated tools.

Crosspoint isn't a NAND/SSD replacement. You're bound to always be disappointed if you only read headlines.
 

·
Registered
Joined
·
391 Posts
Quote:
Originally Posted by mmonnin View Post

You're paying for technology investment and new tools vs fabs that have many depreciated tools.

Crosspoint isn't a NAND/SSD replacement. You're bound to always be disappointed if you only read headlines.
While true fundamentally, this isn't Crosspoint. It's 3D VNAND directly competing with Samsung's 3D VNAND, it's stated beyond the headlines
tongue.gif
In addition to the fact these are datacenter SSDs, I'd assume not intended for purchase by regular consumers and I doubt they're going to be are/available via regular retail channel. Assuming the comparison made is to consumer SSDs

Personally, whilst the price isn't lower right now, Samsung's 3D VNAND SSDs have an extremely reasonable price regardless and I'd call the benefits worth an extra $20 over the very cheapest consumer SATA SSDs, that feature flash made on very small process nodes (undermining their reliability/data retention hugely compared to something massively increasing reliability).

For what it's worth to counter some incorrect and misleading claims made in the article, this quarter Samsung is actually releasing their 3rd generation of VNAND which was demonstrated over 5 months ago, announced over the 6 months ago. Proclaimed advantages of Intel is density, which is actually matched by Samsung; 256Gb MLC NAND packages though I'm unsure about the TLC. Their first drive for enterprise releasing this quarter, also happens to be over 1 million IOPS random 4k read and over 300k IOPS random 4k write. Compared to Intel's "performance-oriented" drive, 3x the random reads and 15x faster writes (as measured independently 5 months ago). Not exactly fit competition in the performance category.
 

·
Premium Member
Joined
·
13,266 Posts
Quote:
Originally Posted by mmonnin View Post

You're paying for technology investment and new tools vs fabs that have many depreciated tools.

Crosspoint isn't a NAND/SSD replacement. You're bound to always be disappointed if you only read headlines.
Xpoint will initially be a DDR4 drop in, but later on it will be a NAND replacement.
 

·
!
Joined
·
9,163 Posts
Quote:
Originally Posted by Sin0822 View Post

Xpoint will initially be a DDR4 drop in, but later on it will be a NAND replacement.
I'd expect it to be in a DIMM form factor for servers, but for the 200 (Union Point) series chipset which will be the first supported platform for consumers, I'd expect it to replace NAND rather than DRAM, at least for the first generation or two. I'd love to be wrong and not have to buy DDR4 and just go straight from DDR3 to this Xpoint magic where your DRAM is 256GB+ and everything is basically constantly loaded, but I don't see that happening for at least ~2 years for consumers.
 

·
Premium Member
Joined
·
13,266 Posts
Quote:
Originally Posted by juano View Post

I'd expect it to be in a DIMM form factor for servers, but for the 200 (Union Point) series chipset which will be the first supported platform for consumers, I'd expect it to replace NAND rather than DRAM, at least for the first generation or two. I'd love to be wrong and not have to buy DDR4 and just go straight from DDR3 to this Xpoint magic where your DRAM is 256GB+ and everything is basically constantly loaded, but I don't see that happening for at least ~2 years for consumers.
Intel said that it would be in DDR4 DIMMs for "next generation Xeons", that was back at IDF and they said it would launch mid 2016 for the server CPU DIMMs. It would be electrically compatible with DDR4, but only work with these Xeons. I assume they were referencing Broadwell-EP since it fits the general time frame and it wasn't released back then. I was at IDF and i atteneded many of their XPoint seminars, some interesting points other than speed numbers:

  • Optane (SSD) will use NVMe protocols and come in modern and already available form factors (assume m.2, u.2, or PCI-E slot).
  • Data retention without power should be roughly the same as modern SSDs.
  • Thermals limits same as modern NAND and DRAM.
  • Longevity initiatives will be much like today's SSDs, partitioning and the such.
  • We will see it in both DRAM and Storage markets by end of this year (not sure if this is still true but I haven't heard anything otherwise).
  • It will use less power than NAND.
 

·
!
Joined
·
9,163 Posts
Quote:
Originally Posted by Sin0822 View Post


  • We will see it in both DRAM and Storage markets by end of this year (not sure if this is still true but I haven't heard anything otherwise).
I just hope that they meant that for consumers as well. Would make Kaby Lake a lot more interesting to have access to memory that would be truly different rather than just "a faster SSD", which I'm not super interested in.
 

·
Premium Member
Joined
·
13,266 Posts
Quote:
Originally Posted by juano View Post

I just hope that they meant that for consumers as well. Would make Kaby Lake a lot more interesting to have access to memory that would be truly different rather than just "a faster SSD", which I'm not super interested in.
Well you have to understand that DRAm will still be faster than XPoint, XPoint is meant to increase memory density on the enterprise side since there is a much greater demand for high density DDR4 banks, but NAND is too slow, that is where Xpoint comes in. Right now there are some stacked DRAM solutions coming out to compete with Xpoint. The increased demand for visualization inside organizations, and the increase in CPU core counts, DRAM needs to keep up. Most corporations want to centralize converged servers in the basement, which will provide VMs to replace desktops. With a decent backbone (I would prefer 10G over 1Gbit) the system should workout as a desktop replacement, put a thin client (weak GPU with a network connection) on every desk to replace the desktop and just plug into the network when you need a computer. It reduces waste since it can scale the servers dynamically based on demand, centralizes hardware and software, reduces troubleshooting and desktop staff costs, allows for much simpler updates, and possibly better licensing agreements. The issue is you don't want to hurt PC performance compared to a modern desktop, you want to maintain something similar, so you see more CPUs with many cores coming out, and with those CPUs you need an increase in DRAM. Instead of providing each VM only 2GB of RAM, imagine being able to bring them up to 4 or 6GB, something usable.

Anyways, I am hoping we see Xpoint in SSDs, since it would make NAND almost obsolete. Has anyone noticed how Samsung is releasing fast drive after fast drive, nothing out there beats the 950 pro in the consumer market, heck nothing really beat the SM951, but they released the 950 Pro and now the SM961, they are obviously moving what they can before Intel and Micron deal a blow their way. Samsung is an interesting company, they value the fact that they offer superior products, so when that competitive advantage is being threatened their strategy has to change, and Samsung has been known for crazy and abrupt strategy shifts in the past.
 

·
Registered
Joined
·
391 Posts
Quote:
Originally Posted by Sin0822 View Post

Anyways, I am hoping we see Xpoint in SSDs, since it would make NAND almost obsolete. Has anyone noticed how Samsung is releasing fast drive after fast drive, nothing out there beats the 950 pro in the consumer market, heck nothing really beat the SM951, but they released the 950 Pro and now the SM961, they are obviously moving what they can before Intel and Micron deal a blow their way. Samsung is an interesting company, they value the fact that they offer superior products, so when that competitive advantage is being threatened their strategy has to change, and Samsung has been known for crazy and abrupt strategy shifts in the past.
I've indeed noticed how rapid they are also, it's rather insane. It honestly feels like R&D are if anything, so fast that they're ready with an advancement but are slowed on the release of it, due to the time it takes for mass production to market.

I'd not even thought of their tendency to do anything to hold onto market leadership, that's a very interesting scenario you put forward if Samsung do indeed want to maintain leadership in the storage market... the rapid shifts in direction have been exactly what's needed to both bring a superior product to market along with them maintaining market leadership in the past, so we could have some pretty interesting results if they feel truly threatened.

They really are quite interesting. I can't say I ever foresaw Samsung suddenly leading the SSD market but I'm glad it happened, they've not stopped their advancement despite lacking competition and it's been great to see it. I can barely believe the rate at which they're pushing VNAND forward though, just wow. Not sure if Intel will make a dent with theirs but it could mean something. I can imagine Xpoint definitely being a mass storage contender in the future maybe, how soon I wish I knew..
 

·
Premium Member
Joined
·
13,266 Posts
Quote:
Originally Posted by CoD511 View Post

I've indeed noticed how rapid they are also, it's rather insane. It honestly feels like R&D are if anything, so fast that they're ready with an advancement but are slowed on the release of it, due to the time it takes for mass production to market.

I'd not even thought of their tendency to do anything to hold onto market leadership, that's a very interesting scenario you put forward if Samsung do indeed want to maintain leadership in the storage market... the rapid shifts in direction have been exactly what's needed to both bring a superior product to market along with them maintaining market leadership in the past, so we could have some pretty interesting results if they feel truly threatened.
There is a really good case study on Samsung done by Harvard Business School, may favorite part is when their president or leader at the time (i forgot his position) was so upset with the quality of their products that he took all the factory workers outside along with their products, put them (products) in a pile, and burned 10s of millions of dollars worth of electronics in front of all those who made them, while preaching higher quality like an evangelical preacher. After that their quality has been quite good, and from the financial stats, you could see they were charging more than the competition for the same DRAM, and they were selling more of it at the higher price since it was less prone to failure (a big issue with DRAm at that time). Samsung as a company took a huge bet even entering the technology market, I think they were originally a textiles manufacturer.

Haha they even mention it in their outreach material:
http://www.samsung.com/common/aboutsamsung/download/performance/Samsung_Profile_2013.pdf
"MARCH 1995
The bonfire is lit, burning defective products
and sending a message to every employee."
 

·
Robotic Chemist
Joined
·
3,010 Posts
Quote:
Originally Posted by Sin0822 View Post

Well you have to understand that DRAm will still be faster than XPoint, XPoint is meant to increase memory density on the enterprise side since there is a much greater demand for high density DDR4 banks, but NAND is too slow, that is where Xpoint comes in. Right now there are some stacked DRAM solutions coming out to compete with Xpoint. The increased demand for visualization inside organizations, and the increase in CPU core counts, DRAM needs to keep up. Most corporations want to centralize converged servers in the basement, which will provide VMs to replace desktops. With a decent backbone (I would prefer 10G over 1Gbit) the system should workout as a desktop replacement, put a thin client (weak GPU with a network connection) on every desk to replace the desktop and just plug into the network when you need a computer. It reduces waste since it can scale the servers dynamically based on demand, centralizes hardware and software, reduces troubleshooting and desktop staff costs, allows for much simpler updates, and possibly better licensing agreements. The issue is you don't want to hurt PC performance compared to a modern desktop, you want to maintain something similar, so you see more CPUs with many cores coming out, and with those CPUs you need an increase in DRAM. Instead of providing each VM only 2GB of RAM, imagine being able to bring them up to 4 or 6GB, something usable.

Anyways, I am hoping we see Xpoint in SSDs, since it would make NAND almost obsolete. Has anyone noticed how Samsung is releasing fast drive after fast drive, nothing out there beats the 950 pro in the consumer market, heck nothing really beat the SM951, but they released the 950 Pro and now the SM961, they are obviously moving what they can before Intel and Micron deal a blow their way. Samsung is an interesting company, they value the fact that they offer superior products, so when that competitive advantage is being threatened their strategy has to change, and Samsung has been known for crazy and abrupt strategy shifts in the past.
What is special about 3DXpoint is not really its capacity, that is good but more than matched by NAND. It is special because it doesn't require power to store data and It is bit addressable and bit erasable. This is good because it keeps the data when off, it is non-volatile, but also because it takes a lot less power to have 1TB of it 'on' in your server. You can also access the data like RAM so you do not need complex wear leveling and pre-fetch methods which add latency and cause write amplification. 3DXpoint has much higher endurance than standard NAND, but only up to good SLC, not invincible like DRAM.

3DXpoint is not going to be used as RAM for hosting VMs, that would still be giving your virtual desktop users RAM that is 10x slower than DRAM. It is for the giant data sets used in Deep Learning and other big data data analysis. It is a way to get terabytes of data very close to the CPU with low power usage and great random access behavior.

I too wait for 3DXpoint but not in the hope that NAND will be obsolete. 3DXpoint really would be best used as a new type of memory. Something like 'active computer memory' with an OS designed to be "always on" and programs using DRAM more as an active short-term cache for data manipulation. However this gets complicated and requires redesigning the OS and all applications so it might be a while.
wink.gif


For now, we consumers get faster storage.
tongue.gif
 

·
Premium Member
Joined
·
13,266 Posts
Quote:
Originally Posted by Asmodian View Post

What is special about 3DXpoint is not really its capacity, that is good but more than matched by NAND. It is special because it doesn't require power to store data and It is bit addressable and bit erasable. This is good because it keeps the data when off, it is non-volatile, but also because it takes a lot less power to have 1TB of it 'on' in your server. You can also access the data like RAM so you do not need complex wear leveling and pre-fetch methods which add latency and cause write amplification. 3DXpoint has much higher endurance than standard NAND, but only up to good SLC, not invincible like DRAM.

3DXpoint is not going to be used as RAM for hosting VMs, that would still be giving your virtual desktop users RAM that is 10x slower than DRAM. It is for the giant data sets used in Deep Learning and other big data data analysis. It is a way to get terabytes of data very close to the CPU with low power usage and great random access behavior.

I too wait for 3DXpoint but not in the hope that NAND will be obsolete. 3DXpoint really would be best used as a new type of memory. Something like 'active computer memory' with an OS designed to be "always on" and programs using DRAM more as an active short-term cache for data manipulation. However this gets complicated and requires redesigning the OS and all applications so it might be a while.
wink.gif


For now, we consumers get faster storage.
tongue.gif
Actually, what makes it special as a DRAm replacement is it's density and yes it's non volatile parameters. What makes it special as a NAND replacement is it's speed. However, Intel is looking to target almost everything, including VMs since they are mentioned many times in the presentation below.

Here is Intel's presentation:
https://hubb.blob.core.windows.net/e5888822-986f-45f5-b1d7-08f96e618a7b-published/984faf16-2ba4-4c1d-81dd-e339b59c93d7/DCWS007%20-%20SF15_DCWS007_100f.pdf?sv=2014-02-14&sr=c&sig=dpDAOPEkriLzx2XgOaJeofS6eXAqufmNz08n%2F2Vr%2FcM%3D&se=2016-04-05T18%3A30%3A02Z&sp=rwd

They see Xpoint as being another tier of memory between DRAM and NAND to meet higher density needs.

Anyways with this technology it would be wrong of either of us to say it's only meant for this or that, i was just giving an example for one of the possible most common uses for servers today and tomorrow, and how Xpoint could help. Everything is virtualized these days.
 

·
Robotic Chemist
Joined
·
3,010 Posts
Virtualized servers with large data sets, yes.

Giving your users 6GB of RAM instead of 2GB for their virtual desktop environments, no.
 

·
Premium Member
Joined
·
13,266 Posts
Quote:
Originally Posted by Asmodian View Post

Virtualized servers with large data sets, yes.

Giving your users 6GB of RAM instead of 2GB for their virtual desktop environments, no.
Actually, they are already doing it with DRAM, but the cost savings would be immense, not to mention the fact that the bottleneck is still in the network.
 

·
Premium Member
Joined
·
13,266 Posts
Quote:
Originally Posted by mmonnin View Post

Intel and Micron said xpoint will create a new market and will not replace DRAM or NAND.
Quote:
Originally Posted by mmonnin View Post

Intel and Micron said xpoint will create a new market and will not replace DRAM or NAND.
We aren't arguing replacing DRAM or NAND, we are discussing application scenarios. Xpoint falls in between the two, and as such will be used for a wide variety of different applications. His argument is that XPoint will be used for large data sets, and he is correct, in the tiered hierarchy approach you have CPU cache levels, then DRAM, then XPoint, and then NAND. XPoint's non volatile properties make it very useful for a wide range of applications. I am arguing that based on the potential cost savings, huge amounts of DRAM (3D stacked DRAM using TSVs) will cost significantly more than equivalent amounts of XPoint (which Intel also says will be the case), so XPoint will enter pretty much any market where huge amounts of DRAM are required. Virtualization is one of those applications, and it will end up in converged systems where the servers are used for a multitude of things depending on demand. We need faster and higher density storage for most everything, not only large data sets. I am not limiting the scope of the product before it is even launched. For Intel's plan to produce XPoint in a DDR4 form factor, they require the latest Xeons to differentiate between DRAM and XPoint, so it wont replace DRAM, but it will heavily compliment it.

On the consumer storage side of things, it is possible that XPoint will enter the market as an NAND alternative. Just like SSDs entered the market as a HDD alternative. SSDs still haven't totally replaced HDDs, while HDDs are slower they still have great cost/density benefits. We don't know if the Optane variant will be differentiated by the system, or just seen as general storage like current SSDs are seen, but from what Intel is saying Optane will act as general storage. I don't see why it wouldn't just replace NAND as system storage for the consumer other than cost.
 
1 - 20 of 21 Posts
Top