3D XPoint

[john3850]

Intel Is Changing The Definition Of The PC

It is a good read.

http://seekingalpha.com/article/3966724-intel-changing-definition-pc?isDirectRoadblock=true&uprof=

 

[beginner99]

The issue this guy forgets is cost and a HDD which we should not forget most of these desktop/client PCs use, will cost way, way less per GB. This is a new tech and intel will milk it in the enterprise sector first.

Second issue is production capacity. It's just not available to replace HDD storage. Not even close.

Same could have been said for SSDs yet it tokk very long for them to get more popular and come down in price and it still makes 0 sense to store media collections on SSDs.

The last issue is endurance. The amount of read and writes to RAM is just not in anyway comparable to HDD/SSD. Even if it is 10k times better than flash, it will not be enough. Not even close.

Intel and Micron are claiming that 3D XPoint provides up to a thousand times higher endurance than NAND. Assuming that the numbers are relative to modern (15-20nm) MLC NAND, the endurance should be in the order of a few million P/E cycles; though the marketing materials are claiming up to tens of millions of write cycles. If we assume 3 million write cycles (1000x of what modern MLC has), a 256GB 3D XPoint based drive would have a total write endurance of 768 petabytes. That's equivalent to 420TB per day for five years, or 4.9GB per second. For storage applications that currently rely on NAND, 3D XPoint will eliminate any potential endurance concerns, but it's not durable enough to challenge DRAM in that front since DRAM endurance is essentially infinite. Whether 3D XPoint provides enough endurance to replace DRAM ultimately depends on the application, but especially in certain enterprise workloads there's a need for DRAM.

http://www.anandtech.com/show/9470/...-1000x-higher-performance-endurance-than-nand

 

[Hellhammer]

There are a few critical errors in the article:

What stands out from the new structure is the Non-Volatile Memory Solutions Group. They specifically characterized the function as "non-volatile" which means that they aren't targeting DRAM at all.

That's because Intel hasn't been in the DRAM business for decades, and the NVM Group has been there for years. Entering the business now wouldn't make any sense given that DRAM revenues are more or less flat and there's been more and more consolidation lately.

Given the cost structure, if 3D XPoint read speeds are "near or equivalent to DRAM read speeds," and all writes go into the speedy eDRAM cache, then the need for standalone DRAM has just vanished.

The eDRAM cache is used by the IGP. Sure Intel could use it for IO caching as well, but the capacities are small (less than what SSDs have) and DRAM is volatile, hence making it vulnerable to power losses.

All in all, 3D XPoint is only talk at this point. Until the first product ships and we know all the essential specs (price, real-world performance and endurance, performance and endurance in long-term etc.), it's impossible to realistically estimate the impact that 3D XPoint may have.

The author also seems to think that Intel and Micron will be the only company with this sort of technology, which is not true. Everyone is preparing their own, similar technologies and while Intel may have a lead in the beginning, the competition will get fierce as others enter the market segment.

 

[IntelUser2000]

Although its an old article...

http://semiaccurate.com/2012/09/07/intel-to-do-away-with-dram-in-pcs/

That's their ultimate goal. No, it doesn't look like Optane Gen 1 would do it. And for PCs its even further away from that. At least for servers we'll see DRAM based Optane with Purley.

The eDRAM cache is used by the IGP. Sure Intel could use it for IO caching as well, but the capacities are small (less than what SSDs have) and DRAM is volatile, hence making it vulnerable to power losses.

If they are in planning to use it for a serious server platform, then it isn't all hype. By the way, the Purley platform would likely use similar strategy. 128GB DIMM as a "cache" for 512GB Optane DIMMs. The server guys would throw away the platform if endurance and performance are so bad. Of course, I am not saying it couldn't go the other way, but things suggest they likely have a good idea what they are doing for Purley.

We can theoretically expand that to the eDRAM. Of course we don't know whether Intel will keep using eDRAM like they do now or use it for a real purpose like as a cache for Optane DIMMs. Based on their graphics investment(or lack thereof) their long term purpose isn't probably to sacrifice gamer folk.

 

[jpiniero]

The author also seems to think that Intel and Micron will be the only company with this sort of technology, which is not true. Everyone is preparing their own, similar technologies and while Intel may have a lead in the beginning, the competition will get fierce as others enter the market segment.

It sounds like to really get the hyped performance out of 3DXP you have to use the DIMM interface/memory controller. A PCIe/NVMe 3DXP isn't going to be substantially better than the competition except maybe on endurance. Intel could of course prevent others from using the DIMM interface.

 

[Jaepheth]

Anyone know the news on this tech? Wasn't it supposed to be on the market by now?

 

[FFFF]

Anyone know the news on this tech? Wasn't it supposed to be on the market by now?

Probably sampling to enterprise clients right now. It will be at least 3-5 years until it will be available at reasonable prices for consumers.

 

[IntelUser2000]

Kabylake will bring Optane support. Don't know if that means NVMe or, hopefully DIMMs, though the latter seems less likely.

Tomshardware is saying about the rumor that Kabylake might support Optane DIMMs. That would be a game changer. It would make the platform worth it even if the CPU is 30% slower. That does make more sense in saying "Kabylake has Optane support", but no one but Intel knows if they are ready with such tech.

Possibly then Optane NVMe would be for non-Kabylake owners.

By the way, the "1000x speed" claim is entirely for latency if anyone is confused. Even HBM doesn't do 100x over NAND. 1/1000-1/500 latency they are talking about compared to NAND is what allows it to be a future RAM replacement. It really is a once in 30 years opportunity to revamp the PC.

 

[R0H1T]

Kabylake will bring Optane support. Don't know if that means NVMe or, hopefully DIMMs, though the latter seems less likely.

Tomshardware is saying about the rumor that Kabylake might support Optane DIMMs. That would be a game changer. It would make the platform worth it even if the CPU is 30% slower. That does make more sense in saying "Kabylake has Optane support", but no one but Intel knows if they are ready with such tech.

Possibly then Optane NVMe would be for non-Kabylake owners.

By the way, the "1000x speed" claim is entirely for latency if anyone is confused. Even HBM doesn't do 100x over NAND. 1/1000-1/500 latency they are talking about compared to NAND is what allows it to be a future RAM replacement. It really is a once in 30 years opportunity to revamp the PC.

It would mean very little if NVMe Optane drives were to be priced at 10x the current cheapest (TLC) alternative. Price is key & unless they get close to the current prices of NVMe drives the Optane series will always be relegated to a niche &/or server segment.

Not sure what you mean by "30% slower" but again if the price for Optane isn't affordable the upgrade to this (Kaby) platform excuse isn't very compelling.

It'd be great if storage & memory were to converge at some point in time, but endurance (& capacity) will always be an issue & until the time that's sorted out DRAM will not be replaced. Then there's the (big) matter of Optane being manufactured just by IMFT, their competitors are also looking at alternatives like memristor & there the issue of standard(s) will crop up.
Intel loves proprietary, TB being a prime example, & if they're not able to beat the likes of HP, Sandisk, Samsung with an exclusive piece of tech on all the major fronts of price, capacity i.e. volume & openness the Optane will remain two notches below SSD, HDD, DRAM et al in becoming a ubiquitous computing solution for everyone but the richest of their clientèles.

 

[ShintaiDK]

Server segment will push 3D XPoint prices down. So whatever initial price/volume etc hitting the end user in the first wave is irrelevant.

Everyone in the server segment is standing ready with their wallets for buying Purley with 3D XPoint for their DB servers etc. And cost is the least of their worries.

 

[john3850]

http://www.pcgamer.com/intels-crazy-fast-3d-xpoint-optane-file-copy-test-approaches-2gbs/

 

[StrangerGuy]

I'm sure it will make sense in big iron servers, but for mainstream consumers its gonna be another tech with next to no real world benefit despite massive increases in price because the software programming itself has become the real bottleneck. Even early 2000s games don't load instantly on a RAMdisk and no faster than on a SSD, for example.

 

[shabby]

http://www.pcgamer.com/intels-crazy-fast-3d-xpoint-optane-file-copy-test-approaches-2gbs/

Samsung: lol

So this optane thing better be cheaper than those 950 pro's if intel wants to regain its ssd business, samsung already has 3.2gbps ssd's waiting for release http://www.anandtech.com/show/10168/samsung-shows-off-sm961-and-pm961-ssds-oem-drives-get-a-boost