Looks cool I guess, but I guess I'm still not particularly interested. Kind of need to know if the pricing will be competitive.
From the onset, it seems like a mistake. Pairing relatively consumer-unwanted Optane with relatively consumer-unwanted QLC NAND storage for an accelerated QLC M.2 ssd at 256/512/1TB sizes. Could it work out and be a great product? Yeah, at the right price. But given that both these technologies are in relative infancy compared to the TLC + DRAM technologies we have on tried-and-true and now pretty affordable M.2 drives, these things need to be priced really well to move in volume. And given Intel's at the reigns here, I don't expect great pricing to happen here. *shrugs*
I'd rather see a unified single-controller drive with transparent caching and enough 3D XPoint that it doesn't need any DRAM. But I don't expect Intel to produce such a product, if only because it would be a cross-platform drive instead of tied to Intel chipsets. Micron might, if their in-house controller team starts delivering real products.
They can do this. Aside from arbitrary reasons(locking out AMD), I don't think they are ready yet. They just got Optane working. A controller that works with both is a significant change and may take 2-3 years to do so if they do it fast. It's a chip after all. You have to design it, put it into silicon, test to make sure it works properly, and validate it.
QLC isn't that cheap at the moment and the proof that it will become significantly cheaper than TLC is still not made. TLC was cheaper than MLC the moment it came out (and had some performance issues down the line, that I never noticed with my original 840). QLC is not cheaper than any comparable TLC drive and the yield issues seem to be significant. QLC was also not marketed as a play to replace performance TLC drives, it was meant to substitute magnetic drives. I don't need more than a medium size SLC cache for an HDD substitute.
Don't get me wrong, I'm still interested in these kinds of products, as I was in hybrid HDDs. They could work. Coming from Intel though gives me a lot of pause, since they like to price themselve out of competition. And hybrid HDDs never really caught on because SSD prices dropped fast enough and the performance uplift wasn't large enough. Something that can easily happen here as well.
Capacity vs price is still not right. Once that happens, I will be all over this. Consumers SSDs are plenty fast except when it does randoms such as doing a Windows update or installation where Optane has an advantage.
I personally see it as a pointless product M.2 SSD are falling in price as practically every brand is doing them from Gigabyte to Corsair, these would of been usefull 5-8 years ago were now i dont think they have much point, if the put Intel Optane in a HDD for example 5400RPM/7200RPM i think that will make a massive difference i know Seagate do the SSHD i am suprised WD haven’t bothered to try make one yet
"the H10 is Optane caching for QLC NAND, with both 3D XPoint and flash memory on the same M.2 module"
That's really awesome. That could mitigate the low endurance (TBW) of QLC, by having all writes first go to the XPoint and later be folded over to the QLC. That might reduce write amplification, and therefore, extend endurance somewhat.
"This suggests the Optane Memory H10 will appear to the host system as two different NVMe SSDs, and will likely require PCIe port bifurcation support to operate a PCIe x4 M.2 slot as two PCIe x2 links."
Well, bummer. That just spoils it.
If Intel develops a QLC SSD with an XPoint cache that reduces write amplification and extends endurance, while transparently appearing to the system as a single, ordinary SSD, then I'm sold.
"If Intel develops a QLC SSD with an XPoint cache that reduces write amplification and extends endurance, while transparently appearing to the system as a single, ordinary SSD, then I'm sold."
They will, but how real-time performance would be better than a good TLC drive with a hefty SLC cache (such as the evo plus)? It would be worth it only if (A) steady state performance is higher and (B) price is lower. I highly doubt this, after having seen the terrible performance that QLC drives have.
Intel should, if they can, package 3D XPoint for on-device storage for high-end consumer devices: , GoPro's & drones, cell phones, hand-held consoles, etc. TLC over the years, especially for 4K/8K video write/delete/write cycles, can't be healthy. And you can't beat the random performance of Optane at low QD for phones & hand-held consoles.
Write endurance isn't really a problem for sequential video recording; TLC or MLC is adequate depending on the use case, and cheaper than 3D XPoint. But I would absolutely love to have it in my phone. I'd rather have 64GB of 3D XPoint than 512GB of flash in my phone.
On video recording: really? Maybe I'm taking Samsung's 64L TLC endurance lightly, but it seems just 2 years of ~2 hours of 40GB/hour (roughly 100Mbps) per day for a 64GB card. Maybe a bigger issue than I suspected. And,eh, I forgot that people still use MLC. Samsung actually did just that for their continuous-recording microSD Card, which uses 64L MLC.
But, yes. In phones/tablets, it'd be killer...This is how you could justify your prices, Apple. ;)
One issue with Xpoint, aside from price of course, is that it uses significantly more power, at idle and usage, than NAND drives. So it could be a problem on smaller devices with lower capacity batteries. Although, it still is an NVMe 1.3 device thus can have very low power states.
Actually the Optane Memory M10 devices are efficient enough in laptops that it gets decent battery life despite being used as a cache.
You don't need a ton of throughput in a phone anyway, so it might just work. Though if you want to go for 64GB 3D XPoint, its better to wait until it works as a complement to RAM, so its like having 64GB of RAM. Then you can have nearly instantaneous boot and loads and get away with having small about of RAM, maybe 1-2GB.
It's 60% less than an average NVMe M.2 Drive on load. While Idle power is 40% more than the average now, it could certainly be optimized for lower idle draw. They have no inherent reason for a 1.2W idle draw and Intel has swaths of experience in reducing idle draw (the entire Ultrabook market).
Optane still does not make sense. RAM and Storage are the same and should be packaged on the same die. Right now we are supposed to pay double and triple for slow products on a bus. An external bus means SLOW. A Ryzen 16 core with 1tb of near shared graphics/cache/ssd would be very interesting. Optane is just old, useless and value adding only to Intel.
If the H10 had been TLC instead of a QLC module, I would have ordered it as soon it came on the market. But QLC ... meh, even with optane caching. You only need caching when your primary storage sucks, and it's limited even then, since you are dependent on cache hits.
I think that the modules with DRAM + 3D X-point on the same module (as planned for Intel Cascade Lake server processors) to lower overall latency.
Ex : 8GB LPDDR-4x / LPDDR-5 + 64GB 3D X-point on the same memory module
Unfortunately, as of January 2019, there is no timeline yet for this kind of product for consumers, and I guess that there is also the need of new protocol like Gen-Z to be natively supported by the processor...
I think something along this kind of idea with will certainly come to market, but unfortunately it will take a few years (somewhere between 2020 and 2025 I would think)...
Just run half of the maximum DRAM on your Mainboard. To me this is as useless as NVMe M.2s. I still us one because my 1Us only house a pair of SSDs. Intel should concentrate on their last area of dominance, single core performance. Once that disappears all of these nifty proprietary technologies will disappear like farts in the wind.
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.
27 Comments
Back to Article
JoeyJoJo123 - Tuesday, January 8, 2019 - link
Looks cool I guess, but I guess I'm still not particularly interested. Kind of need to know if the pricing will be competitive.From the onset, it seems like a mistake. Pairing relatively consumer-unwanted Optane with relatively consumer-unwanted QLC NAND storage for an accelerated QLC M.2 ssd at 256/512/1TB sizes. Could it work out and be a great product? Yeah, at the right price. But given that both these technologies are in relative infancy compared to the TLC + DRAM technologies we have on tried-and-true and now pretty affordable M.2 drives, these things need to be priced really well to move in volume. And given Intel's at the reigns here, I don't expect great pricing to happen here. *shrugs*
Billy Tallis - Tuesday, January 8, 2019 - link
I'd rather see a unified single-controller drive with transparent caching and enough 3D XPoint that it doesn't need any DRAM. But I don't expect Intel to produce such a product, if only because it would be a cross-platform drive instead of tied to Intel chipsets. Micron might, if their in-house controller team starts delivering real products.IntelUser2000 - Tuesday, January 8, 2019 - link
They can do this. Aside from arbitrary reasons(locking out AMD), I don't think they are ready yet. They just got Optane working. A controller that works with both is a significant change and may take 2-3 years to do so if they do it fast. It's a chip after all. You have to design it, put it into silicon, test to make sure it works properly, and validate it.drexnx - Tuesday, January 8, 2019 - link
that was my response too, "yeah, it's cool, but why bother?"ccHephaestus - Tuesday, January 8, 2019 - link
QLC is cheap for high capacity, but its draw back is performance,Optane is blazing fast, but its expensive for high capacity.
To me, It makes perfect sense to use one to supplement the other.
It has the same question any product has, whats the cost? whats the performance?
We will have to wait and see. Still a unique product either way
Death666Angel - Wednesday, January 9, 2019 - link
QLC isn't that cheap at the moment and the proof that it will become significantly cheaper than TLC is still not made. TLC was cheaper than MLC the moment it came out (and had some performance issues down the line, that I never noticed with my original 840). QLC is not cheaper than any comparable TLC drive and the yield issues seem to be significant. QLC was also not marketed as a play to replace performance TLC drives, it was meant to substitute magnetic drives. I don't need more than a medium size SLC cache for an HDD substitute.Don't get me wrong, I'm still interested in these kinds of products, as I was in hybrid HDDs. They could work. Coming from Intel though gives me a lot of pause, since they like to price themselve out of competition. And hybrid HDDs never really caught on because SSD prices dropped fast enough and the performance uplift wasn't large enough. Something that can easily happen here as well.
name99 - Monday, January 21, 2019 - link
The question is not why use cache on a QLC drive, it’s why is OPTANE cache better than good-old SLC cache?TelstarTOS - Sunday, February 24, 2019 - link
it isn't.zodiacfml - Wednesday, January 9, 2019 - link
Capacity vs price is still not right. Once that happens, I will be all over this.Consumers SSDs are plenty fast except when it does randoms such as doing a Windows update or installation where Optane has an advantage.
ScouserPcgamer - Tuesday, January 8, 2019 - link
I personally see it as a pointless product M.2 SSD are falling in price as practically every brand is doing them from Gigabyte to Corsair, these would of been usefull 5-8 years ago were now i dont think they have much point, if the put Intel Optane in a HDD for example 5400RPM/7200RPM i think that will make a massive difference i know Seagate do the SSHD i am suprised WD haven’t bothered to try make one yetMikewind Dale - Tuesday, January 8, 2019 - link
"the H10 is Optane caching for QLC NAND, with both 3D XPoint and flash memory on the same M.2 module"That's really awesome. That could mitigate the low endurance (TBW) of QLC, by having all writes first go to the XPoint and later be folded over to the QLC. That might reduce write amplification, and therefore, extend endurance somewhat.
"This suggests the Optane Memory H10 will appear to the host system as two different NVMe SSDs, and will likely require PCIe port bifurcation support to operate a PCIe x4 M.2 slot as two PCIe x2 links."
Well, bummer. That just spoils it.
If Intel develops a QLC SSD with an XPoint cache that reduces write amplification and extends endurance, while transparently appearing to the system as a single, ordinary SSD, then I'm sold.
CheapSushi - Tuesday, January 8, 2019 - link
Ah darn, I was excited about it but it does indeed kill the interest for me. I would love to use it for warm cache on my rigs.TelstarTOS - Sunday, February 24, 2019 - link
"If Intel develops a QLC SSD with an XPoint cache that reduces write amplification and extends endurance, while transparently appearing to the system as a single, ordinary SSD, then I'm sold."They will, but how real-time performance would be better than a good TLC drive with a hefty SLC cache (such as the evo plus)? It would be worth it only if (A) steady state performance is higher and (B) price is lower. I highly doubt this, after having seen the terrible performance that QLC drives have.
ikjadoon - Tuesday, January 8, 2019 - link
I've said it once and I'll say it again.Intel should, if they can, package 3D XPoint for on-device storage for high-end consumer devices: , GoPro's & drones, cell phones, hand-held consoles, etc. TLC over the years, especially for 4K/8K video write/delete/write cycles, can't be healthy. And you can't beat the random performance of Optane at low QD for phones & hand-held consoles.
Billy Tallis - Tuesday, January 8, 2019 - link
Write endurance isn't really a problem for sequential video recording; TLC or MLC is adequate depending on the use case, and cheaper than 3D XPoint. But I would absolutely love to have it in my phone. I'd rather have 64GB of 3D XPoint than 512GB of flash in my phone.ikjadoon - Tuesday, January 8, 2019 - link
On video recording: really? Maybe I'm taking Samsung's 64L TLC endurance lightly, but it seems just 2 years of ~2 hours of 40GB/hour (roughly 100Mbps) per day for a 64GB card. Maybe a bigger issue than I suspected. And,eh, I forgot that people still use MLC. Samsung actually did just that for their continuous-recording microSD Card, which uses 64L MLC.But, yes. In phones/tablets, it'd be killer...This is how you could justify your prices, Apple. ;)
CheapSushi - Tuesday, January 8, 2019 - link
One issue with Xpoint, aside from price of course, is that it uses significantly more power, at idle and usage, than NAND drives. So it could be a problem on smaller devices with lower capacity batteries. Although, it still is an NVMe 1.3 device thus can have very low power states.IntelUser2000 - Tuesday, January 8, 2019 - link
Actually the Optane Memory M10 devices are efficient enough in laptops that it gets decent battery life despite being used as a cache.You don't need a ton of throughput in a phone anyway, so it might just work. Though if you want to go for 64GB 3D XPoint, its better to wait until it works as a complement to RAM, so its like having 64GB of RAM. Then you can have nearly instantaneous boot and loads and get away with having small about of RAM, maybe 1-2GB.
ikjadoon - Wednesday, January 9, 2019 - link
Pardon? XPoint uses much less power than SSDs on load.https://www.pcper.com/reviews/Storage/Intel-Optane...
It's 60% less than an average NVMe M.2 Drive on load. While Idle power is 40% more than the average now, it could certainly be optimized for lower idle draw. They have no inherent reason for a 1.2W idle draw and Intel has swaths of experience in reducing idle draw (the entire Ultrabook market).
808Hilo - Thursday, January 10, 2019 - link
Optane still does not make sense. RAM and Storage are the same and should be packaged on the same die. Right now we are supposed to pay double and triple for slow products on a bus. An external bus means SLOW. A Ryzen 16 core with 1tb of near shared graphics/cache/ssd would be very interesting. Optane is just old, useless and value adding only to Intel.CheapSushi - Tuesday, January 8, 2019 - link
I really like the H10 with QLC. I'd buy up a few if available depending on price.JKJK - Tuesday, January 8, 2019 - link
If the H10 had been TLC instead of a QLC module, I would have ordered it as soon it came on the market. But QLC ... meh, even with optane caching. You only need caching when your primary storage sucks, and it's limited even then, since you are dependent on cache hits.Diogene7 - Wednesday, January 9, 2019 - link
I think that the modules with DRAM + 3D X-point on the same module (as planned for Intel Cascade Lake server processors) to lower overall latency.Ex : 8GB LPDDR-4x / LPDDR-5 + 64GB 3D X-point on the same memory module
Unfortunately, as of January 2019, there is no timeline yet for this kind of product for consumers, and I guess that there is also the need of new protocol like Gen-Z to be natively supported by the processor...
I think something along this kind of idea with will certainly come to market, but unfortunately it will take a few years (somewhere between 2020 and 2025 I would think)...
One
dsplover - Wednesday, January 9, 2019 - link
Just run half of the maximum DRAM on your Mainboard.To me this is as useless as NVMe M.2s. I still us one because my 1Us only house a pair of SSDs.
Intel should concentrate on their last area of dominance, single core performance.
Once that disappears all of these nifty proprietary technologies will disappear like farts in the wind.
shatteredx - Thursday, February 28, 2019 - link
Any word on a successor to the 905p?darkbasic - Wednesday, March 13, 2019 - link
+1Dotachin - Wednesday, April 17, 2019 - link
If the 128GB, PCIe 3x4 M15 goes for under $200, I want one + PrimoCache please!