News UltraRAM Demos Prototype Chip, Secures Funding to Validate Commercial Potential

[Admin]

UltraRAM technology and business progress continue apace. We visited the company's lab and saw a working prototype in action.

UltraRAM Demos Prototype Chip, Secures Funding to Validate Commercial Potential : Read more

 

[RichardtST]

This would be nice. 1000+ years storage time. Good luck to them!

 

[peachpuff]

This bs sounds familiar.... optane anyone?

 

[edzieba]

This bs sounds familiar.... optane anyone?

Like with Ovonic (3DXP/Optane) and other phase-change memory, it works. Anyone claiming it to be 'bs' is incompetent at best.
The problem, as always, is making it work in an economically viable manner, which means production at scale. For Optane, the same Micron malfeasance that meant it took two decades for Chalcogenide process memory to go from design to initial production hobbled the expansion needed to drop prices, and this implementation could just as well be hobbled by a similar lack of investment (or continued patent trolling from Micron).

The loss of 3DXP is disappointing, because even today it beats the pants off of NAND-based SSDs when it comes to low-queue-depth mixed read and write workloads, which is the typical client use that direct impacts local machine performance. Lets hope this does not see the same fate as other PCM technologies.

 

[Kamen Rider Blade]

Like with Ovonic (3DXP/Optane) and other phase-change memory, it works. Anyone claiming it to be 'bs' is incompetent at best.
The problem, as always, is making it work in an economically viable manner, which means production at scale. For Optane, the same Micron malfeasance that meant it took two decades for Chalcogenide process memory to go from design to initial production hobbled the expansion needed to drop prices, and this implementation could just as well be hobbled by a similar lack of investment (or continued patent trolling from Micron).

The loss of 3DXP is disappointing, because even today it beats the pants off of NAND-based SSDs when it comes to low-queue-depth mixed read and write workloads, which is the typical client use that direct impacts local machine performance. Lets hope this does not see the same fate as other PCM technologies.

3DXP / Optane is literally perfect as a OS drive where most reads are QD1 and Random.

It's only ok as a "Mass Storage" drive where sustained linear speeds are important.

But normal NAND Flash can do "Bulk Storage" that is high-speed or Linear
HDD can do "Bulk Storage" that is cheap on the price per GB/TB

Most people don't need "Bulk Storage" for the OS Drive, they need Optane.

 

[Francis412]

This would be nice. 1000+ years storage time. Good luck to them!

Here is a data sheet on EEPROMS https://ww1.microchip.com/downloads/en/DeviceDoc/21073j.pdf

 

[photon123]

This bs sounds familiar.... optane anyone?

Actually this is nothing like Optane, completely different technology. Sounds more like a variation of flash.

 

[InvalidError]

Most people don't need "Bulk Storage" for the OS Drive, they need Optane.

Most people don't need Optane. They only need a good-enough all-rounder to do everything on.

 

[jeremyj_83]

This bs sounds familiar.... optane anyone?

Optane isn't bs. The major problem it had was for RAM it was too slow and storage it was too small for really anything other than cache drives in HCI arrays.

 

[gg83]

This bs sounds familiar.... optane anyone?

I get what you're saying. A new type of memory with 100000000x the speed. Then it's gone. Octane lasted a few years and that was backed by Intel. This would need a ton of funding.

 

[TJ Hooker]

Like with Ovonic (3DXP/Optane) and other phase-change memory, it works. Anyone claiming it to be 'bs' is incompetent at best.

It's not that 3DXP doesn't work, it's that actual Optane products never achieved the lofty claims made back when it was in early development.

Also, UltraRAM does not appear to be PCM.

 

[Li Ken-un]

Octane lasted a few years and that was backed by Intel.

It lasted only a few years because it wasn’t backed by Intel. That was along with some strategic blunders like tying Optane to proprietary platform support. Rather than making Optane more accessible, thus moving Optane towards profitability, they positioned Optane to drive sales of Xeon.

 

[usertests]

This bs sounds familiar.... optane anyone?

Many have tried, many have failed. Crossbar RRAM comes to mind rather than XPoint/Optane. Intel didn't make claims this big.

If it's not vaporware, the industry will gobble it up, so good luck. However, I'm not sure "4,000x endurance of NAND" is enough for a complete DRAM replacement. I don't see density mentioned so it might not be a NAND killer either.

 

[Diogene7]

I have concern about the read/write endurance claimd of 10E7, which for a DRAM usage seems low I think, no ?

However, for storage purposes, with added benefits of extremely low latency (1ns) and byte adressability, and 1000+ years storage time, it would be excellent if, and only if, density could come close to NAND Flash at a reasonable cost…

 

[Sluggotg]

Looks very impressive but so did, "Cold Fusion", "Room Temperature Super Conductors" etc. I hope they can pull it off. I am a bit skeptical about it. We will find out in a year or two from now.

 

[palladin9479]

It always boils down to economics. Optane was discarded because it was not economical at scale. Lots of great cool tech ends up in the bit bucket.

 

[peachpuff]

Optane isn't bs. The major problem it had was for RAM it was too slow and storage it was too small for really anything other than cache drives in HCI arrays.

It was bs, intels original claims and its actual product were world's apart, nowhere near the performance that they were originally claiming.
But if intel couldn't sell it and make it work then good luck to these folks.

 

[Geef]

If this stuff works then it might go down like this. Big IF

Imagine ---

Your PC has a big switch on the front. 🦯

Click! ⚡ Its off.

Click! ⚡ Its back on after a second or two of stuff getting power, but your OS was already loaded and ready to go in the RAM.

Think of it like when you first switched between HDD and SSD for main drive. Fast load time.

 

[InvalidError]

If this stuff works then it might go down like this. Big IF

Imagine ---

Your PC has a big switch on the front. 🦯

Click! ⚡ Its off.

Click! ⚡ Its back on after a second or two of stuff getting power, but your OS was already loaded and ready to go in the RAM.

Think of it like when you first switched between HDD and SSD for main drive. Fast load time.

Enable Standby, use that instead of shutdown/hibernate and you can already get all of this today at the expense of 1-2W to keep the DRAM in self-refresh mode. Just don't plug peripherals with heaps of always-on RGB into always-on USB ports.

BTW, with a write endurance of only 10 million cycles, you don't want to use this stuff as a DRAM replacement for frequently modified data: if you have a statically/globally allocated memory block in a game engine that gets updated at 100fps, you can expect the memory to burn out in ~28h if you left such software run continuously. This stuff would need to get used as some sort of resident-in-memory asset store that gets written to ~1000X less frequently than DRAM is.

 

[Hooda Thunkett]

Faster than DRAM. Longer term storage than Flash. Less power than either. Does it come with a puppy and a pizza and ice cream party for every purchaser when it debuts?

 

[usertests]

Faster than DRAM. Longer term storage than Flash. Less power than either. Does it come with a puppy and a pizza and ice cream party for every purchaser when it debuts?

Density and cost are unknown, write endurance isn't enough to replace DRAM.

If the properties are just right, it could replace NAND in all SSDs. If it's wrong, it could slide into the XPoint/Optane tier, or directly into the trash bin.

 

[exophase]

The big question mark is cost. Gallium and Indium are very expensive materials. That can be acceptable for thin enough layers but the 2um GaSb (Galium Antimonide, about 36% Gallium by mass) buffer layer stands out as potentially problematic, and that's before knowing what the planar density of the memory is like.

 

[Geef]

Thousand year storage? So stick an M.2 drive of it in a time capsule. Wait no, use a USB stick of it and fill it with whatever the internet thinks is important at the time...

1000 years later...

"Hmm lets open this, ah good we're still using USB-A type drives... WTF? 4k pictures and video of movie stars doing what?!?"

The internet grins evilly.

 

[edzieba]

It's not that 3DXP doesn't work, it's that actual Optane products never achieved the lofty claims made back when it was in early development.

It was bs, intels original claims and its actual product were world's apart, nowhere near the performance that they were originally claiming.

The 'original claims' were the performance projections for final development (i.e. physical limits of the process technology). People just never read the slides, picked the highest number and ignored everything else (including the subsequent slides with real-world benchmarks), and somehow demanded that the first release match the theoretical process limits in ignorance of all evidence and common sense.

 

[Li Ken-un]

It was bs, intels original claims and its actual product were world's apart, nowhere near the performance that they were originally claiming.

Part of it was also due to the limitations of the protocols at the time. In M.2 and U.2 SSD form factor, Optane was bottle-necked by PCIe. While an order magnitude faster than NAND, it wasn’t as fast as promised. The SSDs had latency on the order of thousands of nanoseconds (5 microseconds for the P5800X and about 10 for the prior generations) while the DIMMs had latency on the order of hundreds of nanoseconds (0.35 microseconds). Now we’ve got CXL to overcome some of the latency limitations, but it came too late for Optane and the last of the lowest latency Optane form factor will be stuck with Intel’s aging proprietary platforms.

There is also the fact Optane was new, and they might have taken a very conservative approach. Rather than promising that their product would be—let’s say—thousands of times more endurant than NAND, they released products with warranties that they were absolutely sure they could back.

I don’t doubt that Intel’s raw media could live up to their original claims, but the media was ahead of its time and hamstrung by other technology. And Pat Gelsinger—unfortunately—did not have any patience for Intel’s non-core (pardon the pun) businesses.