I wouldn't be surprised if the capability to run these is already baked into the current Xeon SP hardware and just not enabled or perhaps requires a new BIOS to work.
Oh wow, thats cool to get confirmation there's going to be 3D XPoint DIMMs next year! I know we're nowhere close to a consumer product, which is where I'd end up seeing it AND it may not make any sense in a desktop or laptop, but it is nice to see something really new and different come out of the research labs.
The current M.2 Optane Memory effectively acts as your cache drive. Moving them to a DIMM slot could speed them up a whole lot. If their price was halved and their speed was 10x you would potentially have a good product. Problem still remains the top end of the market don't want low speed DIMMS and the bottom half of the market often don't care or even know how slow their HDD based system is.
It's just a thought, but if prices fall far enough, 3D XPoint DIMMs could end up serving those bottom-end consumers well by unifying RAM and storage into a single pool. Take those recent Atom laptops with 2-4GB of RAM and 32GB of eMMC storage for instance and just replace all that with 64GB of Optane DIMMs and the end user might have a faster system that costs less because the internals are simpler.
I can totally see it. Atom performance is sh!t, it will be a brilliant decision to make it even sh1ttier by denying it ram and using something orders of magnitude slower instead. Consumerism boner on!
Hey intel, give this guy a job. No wonder your mobile device strategy failed - you lak teh genios.
Well Atom was pretty crap. But this thing might actually turn out OK. I mean I'm not an engineer but I doubt Intel would just throw a product out there without some sort of expectation it will do OK. And if not, try try again,
The NV-Dimms pair DRAM with SLC flash. So a 32gb module actually has 32gb dram and 32gb flash. When the power is lost, a battery/capacitor dumps the dram contents to flash.
The process of loading and unloading the flash is something 3d Xpoint does not need to do. The xpoint dimms will be non-volatile without any need for flash, it will be fast without any need of DRAM. What the impact that will have on reliability and performance, I'm not sure. but its very different tech, so I expect the reliability and performance to be different from nv-dimms (better or worse in either category I'm not sure).
Absolutely, nothing nicer than seeing something new that doesn't make any sense. It's the hallmark of the human intellect.
It will pretty much require custom and highly specific software to be written to make adequate use of it. Regular software will not only run abysmally, but will eat through it in record time too. It might come in a DIMM factor, but it is definitely not something anyone anywhere would want to use as a drop-in substitute of RAM. It will struggle to find use even in the enterprise sector. "Consumer product" - sure thing. Just because ;)
Intel hyped the hell out of it, and the nvme flavor couldn't even deliver 5% of the expectations. Being so oddly quiet on the dimm version in times when hype is everything most likely means we are set for an even bigger "meh". Lets wait and see how much faster it will be unhindered by them slow pcie interfaces.
I read exactly what I said on another website and it sounded to the point. I'm not an engineer or even work with this but sometimes I still have a feeling I'm 10 times better that the ones you see at these big companies.
That's not me. Its one of my many "fans" pretending to be me ;)
If you examine the screen name you will find that the "i" in the copycat is not the Latin script "i" but Ukrainian "i", giving a pretty good idea who this sad loser is.
It won't struggle to find a use in the enterprise sector. It's an obvious replacement for DRAM in servers that depend more on memory capacity than performance. That's super useful in HPC. Since you're a moron and need your hand held: the more memory you have, the bigger the simulations you can run.
If you had read about the products you criticize, like any intelligent person would, you'd embarrass yourself a lot less often. You should try doing that sometime.
Comedy gold right here - you say "servers that depend more on memory capacity than performance" and immediately after than you say "That's super useful in HPC".
Do you realize that HPC stands for HIGH PERFORMANCE COMPUTING? Yeah, it just makes perfect sense, using a working memory solution that is completely crippled when it comes to high performance in a high performance computing scenario.
You struggle even with common sense. Calling yourself intelligent all the while embarrassing yourself by posting complete and utter nonsense... You are a stereotypical AT commenters - a clueless wannabe that seeks to cultivate illusion of intelligence by cheering at stuff you know nothing about.
I feel like I'm surrounded by idiots. If the forums here were curated I'd probably be left alone. Nobody will ever buy this, I stake my name to this. It's completely useless.
I have no idea how these will turn out, but you lack imagination or are being purposefully blind f you cant see the value in something like this. Of course general purpose SW won't handle this well. Nobody will buy these and not use them with customized SW packages. Large in-memory databases wouldn't adapt to this technology?
That last device is optane. Yes, it's faster than the dram nvme device. That's my point. All these devices, including that Samsung nvme, are running at the limits of nvme. Now, yes, those are best case numbers but my point is that we can't yet say what the medium access times are because they are hidden within the noise and granularity of nvme.
I love how Intel introduced and then deprecated one of the 3D XPoint, a.k.a. "persistent memory" instructions long before the product even started shipping:
You know they've been working toward this for many years now. They've patched the Linux kernel for NVDIMM, written ACPI rules, CPU instructions and many other things. If they discovered in testing that one of their instructions was a bad idea, it is good that they found out early.
Zan, I think that's a bad reading of the situation. To me the deprecation is more problematic.
The deprecation forces the memory controller to be part of the power-fail-safe domain. Is the idea that future memory controllers will ALWAYS have enough capacitor capacity on the die/package/motherboard that they will be able to force out all pending writes when the system loses power? (And that means enough power not just to run the memory controller, but also to push out the writes meaning enough power to keep the nvDIMMs alive.)
What are the implication of this for non-server systems? In particular consider (a) non-Intel (ie AMD) systems (b) small systems (Surface sorts of things, or Atom and Quark type designs). How onerous are these power storage requirements? Do they prevent the use of nvDIMMs in x86 small systems more or less forever (at least until a substantial technology reconfiguration)?
That might have to do with your affinity for idiocy. Because it wouldn't be hyper in any way - performance will be total garbage compared to ram, and endurance will be like a couple of months at best if you use it as ram. Last but not least, because it not intended to be used as ram.
But you go ahead and call it hyper ram, because like that other guy, the pinnacle of appreciation for you is liking mindless stuff :P
"frustration"? Your reading comprehension inadequacy is only surpassed by your ability to create strawmen. My dog can lick his balls also, that doesn't make him superior to me.
Seriously though, at one point I thought you might provide interesting technical contradictory points, and then I would read others responses to try and determine if validity one way or the other. But, now all it is snide comments mostly not relevant to any serious discussion and when someone makes a technical rebuttal all you do is ad hominem attacks and deflection. I don't want to argue w/you because that is pointless, I just want to block your comments so I can read the serious ones.
I believe everyone with an interest in servers has been putting support for persistent memory into their processors. Not all that much is required but it is like he problem in systems to ensure a flush really does put data somewhere that would survive a power failure rather than some cache that gets lost. There is a tension there that as often gone for speed rather than safety.
The point is that it would be possible to cut down the power requirements for a smartphone whilst making it look like it has a larger memory. You'd want to only read or write as necessary when starting up or closing down and keep as little memory as possible powered up in between uses.
Yeah, I'm sorry about that. Since Optane and 3D XPoint are being launched alongside one another and are using the same underlying technology, I've been making the mistake of interchanging the two terms when talking about them both. :(
With Intel previously announcing support in Purley 2, I would assume this will remain a HEDT-only product for a gen or two before it starts to make its way to consumer platforms. If ever, that is; the DIMM variant isn't a good fit for client workloads, explicit drives or transparent caches fit that niche better.
The more amusing point is the second slide and its single Xeon entry in the compute pool. (Even though there's plenty of space there for more logos.) That must make Altera and Nervana feel good! And how about you, Raja? Looking forward to more topnotch support like this from Intel PR are you?
Intel can talk and talk about how they love all their kids, but it seems that, when push comes to shove, there's always one kid that gets a LOT more love than all the others...
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
58 Comments
Back to Article
extide - Tuesday, November 14, 2017 - link
I wouldn't be surprised if the capability to run these is already baked into the current Xeon SP hardware and just not enabled or perhaps requires a new BIOS to work.PeachNCream - Tuesday, November 14, 2017 - link
Oh wow, thats cool to get confirmation there's going to be 3D XPoint DIMMs next year! I know we're nowhere close to a consumer product, which is where I'd end up seeing it AND it may not make any sense in a desktop or laptop, but it is nice to see something really new and different come out of the research labs.Amandtec - Wednesday, November 15, 2017 - link
The current M.2 Optane Memory effectively acts as your cache drive. Moving them to a DIMM slot could speed them up a whole lot. If their price was halved and their speed was 10x you would potentially have a good product. Problem still remains the top end of the market don't want low speed DIMMS and the bottom half of the market often don't care or even know how slow their HDD based system is.PeachNCream - Wednesday, November 15, 2017 - link
It's just a thought, but if prices fall far enough, 3D XPoint DIMMs could end up serving those bottom-end consumers well by unifying RAM and storage into a single pool. Take those recent Atom laptops with 2-4GB of RAM and 32GB of eMMC storage for instance and just replace all that with 64GB of Optane DIMMs and the end user might have a faster system that costs less because the internals are simpler.ddriver - Wednesday, November 15, 2017 - link
I can totally see it. Atom performance is sh!t, it will be a brilliant decision to make it even sh1ttier by denying it ram and using something orders of magnitude slower instead. Consumerism boner on!Hey intel, give this guy a job. No wonder your mobile device strategy failed - you lak teh genios.
PeachNCream - Wednesday, November 15, 2017 - link
Someone seems kinda grumpy-wumpy today.ddrіver - Wednesday, November 15, 2017 - link
Well Atom was pretty crap. But this thing might actually turn out OK. I mean I'm not an engineer but I doubt Intel would just throw a product out there without some sort of expectation it will do OK. And if not, try try again,rep_movsd - Saturday, November 18, 2017 - link
Today? ddriver can only shit when commenting on an Intel praising article, that makes him grumpy from constipation.Lord of the Bored - Thursday, November 16, 2017 - link
Look at it the other way: How much faster is a slow DIMM going to be than friggin' eMMC?IGTrading - Thursday, November 16, 2017 - link
We honestly believe this tech makes no sense in DIMM applications since we already have these : https://www.anandtech.com/show/12029/micron-announ...So what's the point ?! What's the advantage ?
Hixbot - Friday, November 17, 2017 - link
The NV-Dimms pair DRAM with SLC flash. So a 32gb module actually has 32gb dram and 32gb flash. When the power is lost, a battery/capacitor dumps the dram contents to flash.The process of loading and unloading the flash is something 3d Xpoint does not need to do. The xpoint dimms will be non-volatile without any need for flash, it will be fast without any need of DRAM.
What the impact that will have on reliability and performance, I'm not sure. but its very different tech, so I expect the reliability and performance to be different from nv-dimms (better or worse in either category I'm not sure).
Capt Calamity - Wednesday, November 29, 2017 - link
512GB per DIMM is the point. Or 2 to 4 TB of persistent memory. Also persistent memory.ddriver - Wednesday, November 15, 2017 - link
Absolutely, nothing nicer than seeing something new that doesn't make any sense. It's the hallmark of the human intellect.It will pretty much require custom and highly specific software to be written to make adequate use of it. Regular software will not only run abysmally, but will eat through it in record time too. It might come in a DIMM factor, but it is definitely not something anyone anywhere would want to use as a drop-in substitute of RAM. It will struggle to find use even in the enterprise sector. "Consumer product" - sure thing. Just because ;)
Intel hyped the hell out of it, and the nvme flavor couldn't even deliver 5% of the expectations. Being so oddly quiet on the dimm version in times when hype is everything most likely means we are set for an even bigger "meh". Lets wait and see how much faster it will be unhindered by them slow pcie interfaces.
PeachNCream - Wednesday, November 15, 2017 - link
Sure that's really nice! -.oddrіver - Wednesday, November 15, 2017 - link
I read exactly what I said on another website and it sounded to the point. I'm not an engineer or even work with this but sometimes I still have a feeling I'm 10 times better that the ones you see at these big companies.PeachNCream - Wednesday, November 15, 2017 - link
Why not apply for a job at one of those places?ddriver - Wednesday, November 15, 2017 - link
That's not me. Its one of my many "fans" pretending to be me ;)If you examine the screen name you will find that the "i" in the copycat is not the Latin script "i" but Ukrainian "i", giving a pretty good idea who this sad loser is.
ddrіver - Thursday, November 16, 2017 - link
That's my keyboard layout actually...CajunArson - Wednesday, November 15, 2017 - link
What's that smell? Oh, DDriver heard the news and needs a new diaper.Again.
ddriver - Wednesday, November 15, 2017 - link
Someone give this poor guy a medal :DIII-V - Wednesday, November 15, 2017 - link
It won't struggle to find a use in the enterprise sector. It's an obvious replacement for DRAM in servers that depend more on memory capacity than performance. That's super useful in HPC. Since you're a moron and need your hand held: the more memory you have, the bigger the simulations you can run.If you had read about the products you criticize, like any intelligent person would, you'd embarrass yourself a lot less often. You should try doing that sometime.
ddriver - Wednesday, November 15, 2017 - link
Comedy gold right here - you say "servers that depend more on memory capacity than performance" and immediately after than you say "That's super useful in HPC".Do you realize that HPC stands for HIGH PERFORMANCE COMPUTING? Yeah, it just makes perfect sense, using a working memory solution that is completely crippled when it comes to high performance in a high performance computing scenario.
You struggle even with common sense. Calling yourself intelligent all the while embarrassing yourself by posting complete and utter nonsense... You are a stereotypical AT commenters - a clueless wannabe that seeks to cultivate illusion of intelligence by cheering at stuff you know nothing about.
ddrіver - Thursday, November 16, 2017 - link
I feel like I'm surrounded by idiots. If the forums here were curated I'd probably be left alone.Nobody will ever buy this, I stake my name to this. It's completely useless.
Capt Calamity - Wednesday, November 29, 2017 - link
You are an idiot, and obviously a troll. No food for you.FwFred - Wednesday, November 15, 2017 - link
I have no idea how these will turn out, but you lack imagination or are being purposefully blind f you cant see the value in something like this. Of course general purpose SW won't handle this well. Nobody will buy these and not use them with customized SW packages. Large in-memory databases wouldn't adapt to this technology?tuxRoller - Wednesday, November 15, 2017 - link
The very fastest devices run close to the limits of the platform. DRAM-based nvme devices perform very similarly to optane.https://lkml.org/lkml/2015/11/6/454
That last device is optane. Yes, it's faster than the dram nvme device. That's my point. All these devices, including that Samsung nvme, are running at the limits of nvme. Now, yes, those are best case numbers but my point is that we can't yet say what the medium access times are because they are hidden within the noise and granularity of nvme.
bathotropic - Tuesday, November 14, 2017 - link
I look forward to having real-time targeted ads that update with every key stroke. Thanks Intel for your big data enhancements.londedoganet - Wednesday, November 15, 2017 - link
What are you on?Amandtec - Wednesday, November 15, 2017 - link
Why would you look forward to ads? If you are being sarcastic consider putting a /s at the end of your post to enable the sarcasm font.Spoelie - Wednesday, November 15, 2017 - link
it's sad that this post needs a special tag to be recognized as sarcasm - especially while it should be blatantly obviousinighthawki - Wednesday, November 15, 2017 - link
I know that interpreting sarcasm in text is sometimes difficult, but come on man, this one is really quite obvious.Elstar - Tuesday, November 14, 2017 - link
I love how Intel introduced and then deprecated one of the 3D XPoint, a.k.a. "persistent memory" instructions long before the product even started shipping:https://software.intel.com/en-us/blogs/2016/09/12/...
Zan Lynx - Wednesday, November 15, 2017 - link
You know they've been working toward this for many years now. They've patched the Linux kernel for NVDIMM, written ACPI rules, CPU instructions and many other things. If they discovered in testing that one of their instructions was a bad idea, it is good that they found out early.name99 - Wednesday, November 15, 2017 - link
Zan, I think that's a bad reading of the situation. To me the deprecation is more problematic.The deprecation forces the memory controller to be part of the power-fail-safe domain. Is the idea that future memory controllers will ALWAYS have enough capacitor capacity on the die/package/motherboard that they will be able to force out all pending writes when the system loses power? (And that means enough power not just to run the memory controller, but also to push out the writes meaning enough power to keep the nvDIMMs alive.)
What are the implication of this for non-server systems? In particular consider
(a) non-Intel (ie AMD) systems
(b) small systems (Surface sorts of things, or Atom and Quark type designs).
How onerous are these power storage requirements? Do they prevent the use of nvDIMMs in x86 small systems more or less forever (at least until a substantial technology reconfiguration)?
Hurr Durr - Wednesday, November 15, 2017 - link
Hyperam!ddriver - Wednesday, November 15, 2017 - link
Hype? Shut your blasphemous foul mouth! It will revolutionize everything!Samus - Wednesday, November 15, 2017 - link
I actually like the name HyperRam ;)ddriver - Wednesday, November 15, 2017 - link
That might have to do with your affinity for idiocy. Because it wouldn't be hyper in any way - performance will be total garbage compared to ram, and endurance will be like a couple of months at best if you use it as ram. Last but not least, because it not intended to be used as ram.But you go ahead and call it hyper ram, because like that other guy, the pinnacle of appreciation for you is liking mindless stuff :P
ironargonaut - Wednesday, November 15, 2017 - link
I know this is one of the most asked questions on Anandtech, but how do I block ddriver again? His drivel has finally passed my tolerance level.ddrіver - Wednesday, November 15, 2017 - link
Is this frustration because I am superior? I don't want to tickle my own balls but I could teach every single engineer at Intel some lessons.ironargonaut - Thursday, November 16, 2017 - link
"frustration"? Your reading comprehension inadequacy is only surpassed by your ability to create strawmen. My dog can lick his balls also, that doesn't make him superior to me.ironargonaut - Thursday, November 16, 2017 - link
Seriously though, at one point I thought you might provide interesting technical contradictory points, and then I would read others responses to try and determine if validity one way or the other. But, now all it is snide comments mostly not relevant to any serious discussion and when someone makes a technical rebuttal all you do is ad hominem attacks and deflection. I don't want to argue w/you because that is pointless, I just want to block your comments so I can read the serious ones.ddrіver - Saturday, November 18, 2017 - link
Great that nobody cares about what you want then.iwod - Wednesday, November 15, 2017 - link
Does it work with Any DIMM? Like with EPYC or ARM64 processors?Dmcq - Wednesday, November 15, 2017 - link
I believe everyone with an interest in servers has been putting support for persistent memory into their processors. Not all that much is required but it is like he problem in systems to ensure a flush really does put data somewhere that would survive a power failure rather than some cache that gets lost. There is a tension there that as often gone for speed rather than safety.tsk2k - Wednesday, November 15, 2017 - link
Okay, but do I need this in my gaming rig?Dmcq - Wednesday, November 15, 2017 - link
No, but I can definitely see it being used in smartphones.PeachNCream - Wednesday, November 15, 2017 - link
Oooh that would be a great place for Optane to go!Dmcq - Wednesday, November 15, 2017 - link
Optane is the SSD version. That would be too slow for instant on for smartphones.Intel do sell Optane for gaming and it is fine for that purpose.
Dmcq - Wednesday, November 15, 2017 - link
The point is that it would be possible to cut down the power requirements for a smartphone whilst making it look like it has a larger memory. You'd want to only read or write as necessary when starting up or closing down and keep as little memory as possible powered up in between uses.PeachNCream - Wednesday, November 15, 2017 - link
Yeah, I'm sorry about that. Since Optane and 3D XPoint are being launched alongside one another and are using the same underlying technology, I've been making the mistake of interchanging the two terms when talking about them both. :(ddriver - Wednesday, November 15, 2017 - link
Badly. Desperately even.tsk2k - Wednesday, November 15, 2017 - link
How many FPS do I get in Crysis 3?Lord of the Bored - Thursday, November 16, 2017 - link
Not until it comes with RGB lighting.edzieba - Wednesday, November 15, 2017 - link
With Intel previously announcing support in Purley 2, I would assume this will remain a HEDT-only product for a gen or two before it starts to make its way to consumer platforms. If ever, that is; the DIMM variant isn't a good fit for client workloads, explicit drives or transparent caches fit that niche better.name99 - Wednesday, November 15, 2017 - link
The more amusing point is the second slide and its single Xeon entry in the compute pool. (Even though there's plenty of space there for more logos.)That must make Altera and Nervana feel good!
And how about you, Raja? Looking forward to more topnotch support like this from Intel PR are you?
Intel can talk and talk about how they love all their kids, but it seems that, when push comes to shove, there's always one kid that gets a LOT more love than all the others...
name99 - Wednesday, November 15, 2017 - link
OK, I get it. Those (Nervana, Altera, Raja) all live in the Accelerator pool. That's the grand theory.