from the waiting-for-one-with-serial-port,-parallel-port,-and-a-Hercules-video-adapter dept.
Startup Unisantis Proposes DRAM Alternative
Singapore-based DRAM specialist Unisantis Electronics revealed at this week's (virtual) IEEE International Memory Workshop (IMW) advances in its work on dynamic flash memory (DFM) that it claims is a faster and denser technology than DRAM or other types of volatile memory.
[...] DFM is also a type of volatile memory, but since it does not rely on capacitors it has fewer leak paths, it has no connection between switching transistors and a capacitor. The result is a cell design with the potential for significant increases in transistor density and —because it not only offers block refresh, but as a Flash memory it offers block erase — DFM reduces the frequency and the overhead of the refresh cycle and is capable of delivering significant improvements in speed and power compared to DRAM.
Using TCAD simulation, researchers at Unisantis have proven that DFM has a substantial potential to increase density 4X compared to DRAM. The scaling of DRAM has almost stopped at 16Gb, according to recent IEEE ISSCC (International Solid-State Circuits Conference) papers.
Using a PCIe Slot to Install DRAM: New Samsung CXL.mem Expansion Module
In the computing industry, we've lived with PCIe as a standard for a long time. It is used to add any additional features to a system: graphics, storage, USB ports, more storage, networking, add-in cards, storage, sound cards, Wi-Fi, oh did I mention storage? Well the one thing that we haven't been able to put into a PCIe slot is DRAM – I don't mean DRAM as a storage device, but memory that actually is added to the system as useable DRAM. Back in 2019 a new CXL standard was introduced, which uses a PCIe 5.0 link as the physical interface. Part of that standard is CXL.memory – the ability to add DRAM into a system through a CXL/PCIe slot. Today Samsung is unveiling the first DRAM module specifically designed in this way.
[...] Samsung's unveiling today is of a CXL-attached module packed to the max with DDR5. It uses a full PCIe 5.0 x16 link, allowing for a theoretical bidirectional 32 GT/s, but with multiple TB of memory behind a buffer controller. In much the same way that companies like Samsung pack NAND into a U.2-sized form factor, with sufficient cooling, Samsung does the same here but with DRAM.
Related: Compute Express Link Specification (CXL) Version 1.0 Launched
Related Stories
CXL Specification 1.0 Released: New Industry High-Speed Interconnect From Intel
With the battleground moving from single core performance to multi-core acceleration, a new war is being fought with how data is moved around between different compute resources. The Interconnect Wars are truly here, and the battleground just got a lot more complicated. We've seen NVLink[0], CCIX[1], and GenZ[2] come out in recent years as offering the next generation of host-to-device and device-to-device high-speed interconnect, with a variety of different features. Now CXL, or Compute Express Link, is taking to the field.
This new interconnect, for which the version 1.0 specification is being launched today, started in the depths of Intel's R&D Labs over four years ago, however what was made is being launched as an open standard, headed up by a consortium of nine companies. These companies include Alibaba, Cisco, Dell EMC, Facebook, Google, HPE, Huawei, Intel, and Microsoft, which as a collective was described as one of the companies as 'the biggest group of influencers driving a modern interconnect standard'.
[...] While some of the competing standards have 20-50+ members, the Compute Express Link actually has more founding members than PCIe (5) or USB (7). That being said however, there are a few key names in the industry missing: Amazon, Arm, AMD, Xilinx, etc. Other standards playing in this space, such as CCIX and GenZ, have common members with CXL, and when questioned on this, the comment from CXL was that GenZ made a positive comment to the CXL press release - they stated that there is a lot of synergy between CXL and GenZ, and they expect the standards to dovetail rather than overlap. It should be pointed out that Xilinx, Arm, and AMD have already stated core CCIX support, either plausible future support or in products at some level, making this perhaps another VHS / Betamax battle. The other missing company is NVIDIA, who are more than happy with NVLink and its association with IBM.
[0] NVlink:
NVIDIA® NVLink™ technology addresses this interconnect issue by providing higher bandwidth, more links, and improved scalability for multi-GPU and multi-GPU/CPU system configurations. A single NVIDIA Tesla® V100 GPU supports up to six NVLink connections and total bandwidth of 300 GB/sec—10X the bandwidth of PCIe Gen 3.
(Score: 2, Interesting) by Anonymous Coward on Saturday May 22 2021, @05:41PM (13 children)
I do not see anything in TFA about getting around the limited erase cycles of flash. And without that, any damn loop variable that goes outside the cache, will kill the thing dead in seconds.
(Score: 1) by fustakrakich on Saturday May 22 2021, @05:44PM
Yeah, we should be going in the opposite direction away from flash to something more durable
La politica e i criminali sono la stessa cosa..
(Score: 2) by takyon on Saturday May 22 2021, @05:45PM (8 children)
They've either fixed that problem, or the "memory makers and foundries" will laugh them out of the room.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 1) by fustakrakich on Saturday May 22 2021, @06:33PM
If they fixed the problem, wouldn't that make big flashy headlines everywhere?
La politica e i criminali sono la stessa cosa..
(Score: 0) by Anonymous Coward on Saturday May 22 2021, @06:42PM (6 children)
The pie-in-the-sky advertising articles are written not for "memory makers and foundries" but for lay investors. The more ignorant, the better.
(Score: 2) by takyon on Saturday May 22 2021, @06:54PM (5 children)
There's research like this from a decade ago:
https://ieeexplore.ieee.org/abstract/document/6479008 [ieee.org]
It slowed down after 3D NAND took off and put a giant stacked band-aid on the endurance problem.
Also, 4x density compared to DRAM should mean that it is far less dense than normal NAND, something to think about.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 1, Informative) by Anonymous Coward on Saturday May 22 2021, @07:17PM (4 children)
"100M cycles", for RAM, is the literal nothing.
Any old 4096x4096 RGB image is 48M color values; loop over them twice, go buy new memory?
(Score: 2, Redundant) by takyon on Saturday May 22 2021, @07:38PM (3 children)
Your math is way off. 100 million cycles means each cell (or the full contents of the device) can be rewritten 100 million times.
One Intel Optane product [blocksandfiles.com] has an endurance of about 1.4 million cycles, enough for 770 drive writes per day for 5 years. 100 million cycles would be about 55,000 DWPD, or one drive write per ~1.6 seconds for 5 years.
But we don't know what the endurance of this "dynamic flash memory" is. Check back in 10 years for the next DFM update.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 0) by Anonymous Coward on Saturday May 22 2021, @10:10PM (2 children)
Your brainz are way off if you think persisting can make you not wrong.
A loop counter is a variable, it sits at a FIXED address and is written SINGLY. No nice remapping trickies for you, RAM isn't a block device.
(Score: 0) by Anonymous Coward on Saturday May 22 2021, @10:57PM
The day my compiler puts a loop variable in ram instead of register is the day I get a new compiler.
The day my store forwarding engine and/or L1 data cache fail to cache my loop that was put in ram is the day I get a new CPU.
The day my L2 fails to cache frequently used loop data, is the day I get a new CPU. (Note L2 generally isn't write through these days)
RAM is far away time wise compared to the actual register set.
If ram is a block device, the block are cache line sized, 64 Bytes, so remapping them does seem impractical.
Write logging data into a circular buffer with non-temporal stores and you will still blow through the write endurance pretty fast: orders of magnitude faster that you would get with full device writes with a system that remaps.
(Score: 1, Interesting) by Anonymous Coward on Sunday May 23 2021, @12:34AM
Maybe, maybe not. If a controller sits between memory and address bus, all bets are off so long as the controller maintains a consistent model (illusion?) for the CPU.
(Score: 2) by fakefuck39 on Saturday May 22 2021, @08:27PM (2 children)
You're thinking about it wrong. The issue is, the use case you have in mind is yours - your computer or cute little server. The use case for this however, is the same use case as Optane - in fact, this appears to be an Optane competitor.
This is a replacement for DRAM. Just like Optane is a replacement for DRAM. Guess what else you use terabytes and terabytes of DRAM for? Hint: it's not storing variables.
Take a large storage array, attached to a SAN. It has about 20TB of DRAM, which is used as cache for read and write IO. So your write a prefetched disk track to it once, then you read it a hundred times. Instead of getting 20TB of DRAM, I can get 12TB of Optane, and 10TB of DRAM. Much cheaper, same performance for a sustained IO peak. This is literally standard design of any enterprise storage array, and the technology this is designed for. No, it's not designed to keep the stack on the PC in your basement that's running some perl script you wrote.
Yes, if you buy a ferarri and move a piano with it, the racecar will be dead within seconds. The racecar is not a bad product. The guy using it to move a piano (you) just has a very small mind and cannot fathom any use case outside of what's right in front of him. I doubt you're actually willing to learn anything, but here, in case you can read. Here's a storage array that'll do 15mil IOPs, and could use storage like this. https://www.delltechnologies.com/en-gb/storage/powermax/powermax-8000-all-flash-storage-array.htm [delltechnologies.com]
(Score: 0) by Anonymous Coward on Saturday May 22 2021, @10:14PM (1 child)
The reason for not using a plain old mmap() in your scenario is - what, exactly?
(Score: 2) by fakefuck39 on Saturday May 22 2021, @10:45PM
you think a $10mil storage array with 5PB of data and 20TB of cache, running on 600 xeon cores, provisioning virtual disks to 2000 servers, doing 15mil IOPs, should, umm, mmap something? There are about 5mil lines of code on a Symmetrix class array just to handle caching. That's why. Because it's not designed by an idiot who builds PCs in the basement.
building a PC in your basement from parts off newegg does not give you knowledge of anything besides building a PC in your basement.
(Score: -1, Offtopic) by Anonymous Coward on Saturday May 22 2021, @05:48PM (1 child)
Niggers Tongue My Anus
(Score: 0, Insightful) by Anonymous Coward on Saturday May 22 2021, @05:57PM
We know it's you, acid andy. Fess up.
(Score: 2) by sjames on Saturday May 22 2021, @07:22PM (19 children)
Dynamic flash sounds like somebody has managed to combine the worst characteristics of DRAM and Flash into one craptacular technology.
They need to either try again or fire their marketing department.
(Score: 2) by fakefuck39 on Saturday May 22 2021, @08:33PM (18 children)
Yes, exactly like Optane, great for HCI and storage arrays, which already use Optane. Huge demand out there, widely used in the industry, is in most datacenters. You mean they need to fire their marketing department because you, someone with zero exposure to the enterprise scale systems running every aspect of your life, an uninformed user, can't think of a use case for this?
They are trying again. This is competition to a widely successful product that's used everywhere. What you need to try again with, is not commenting on shit you know zero about, yet feel are right about, and the huge companies and their marketing and engineers are wrong about. This technology is awesome, and reduces the cost of a storage array by a million bucks, and an HCI farm by a few hundred thousand. Which is why tech like it is and has been in use for years, but literally every company out there.
(Score: 2) by sjames on Saturday May 22 2021, @09:14PM (17 children)
It's actually the opposite of Optane. Also Optane is not nearly as in demand as you might think. Put the cool aid down. Further, Intel is targeting gaming. How much demand do you see for improved gaming performance in the enterprise?
(Score: 2) by takyon on Saturday May 22 2021, @09:22PM
Intel is focusing on enterprise storage with Optane, not gaming.
Intel Kills Off All Optane-Only SSDs for Consumers, No Replacements Planned (Updated) [tomshardware.com]
Intel® Optane™ Storage for Enterprise [intel.com]
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2, Flamebait) by fakefuck39 on Saturday May 22 2021, @10:00PM (13 children)
You are literally a moron, unwilling to learn and making shit up. I on the other hand, am literally an engineer who designs and sells the things I am talking about.
Optane is not about improved performance. Optane decreases performance, not increases it. Optane replaces RAM in storage arrays. Small amounts of Optane also replace large amounts of flash cache in HCI. Optane is for saving money, not for increasing performance.
Intel targeting gaming with Optane? You're on crack. Me on the other hand - I'm pumping out a couple of quotes per week for enterprise systems that have Optane.
>How much demand do you see for improved gaming performance in the enterprise?
zero.
>Intel is targeting gaming
I love it when election fraud retards talk tech. You're very entertaining. Have you gotten your 5G injection yet? Newsflash - when you make up blatant laughable bullshit, it works for rednecks and dumb people like yourself. Saying things like this outloud to an audience of people literally genetically superior to you, just makes you our clown.
(Score: 2) by sjames on Saturday May 22 2021, @10:19PM (2 children)
Look, I know times are tough all around, but drinking furniture polish isn't the answer. Seek help.
(Score: 2) by fakefuck39 on Saturday May 22 2021, @10:34PM
At least I don't have sex with parrots while the sheep watch. I'm not sure why you do that, but your pastor does not approve. He does approve when you have sex with your dad though. I'm not sure why, but it is what it is.
(Score: 0) by Anonymous Coward on Tuesday May 25 2021, @02:51AM
The autism is spilling out of your vagina. Have you tried Depends(R)?
(Score: 0) by Anonymous Coward on Saturday May 22 2021, @10:20PM (4 children)
You are literally a Russian troll. With that sorry occupation, whatever else you are, cannot be anything well-paid.
(Score: 2) by fakefuck39 on Saturday May 22 2021, @10:37PM (3 children)
you are free to your opinions, and I in fact may be a russian troll, or a goat fucker. In the real world, I am a well off enterprise storage and compute engineer, and you have autism, and are a highschool social reject trapped in an aged body. you have no way out, your brain defect is not curable, and you will live and die a very depressing life, hyperfocused on random shit, plotting and scheming about weird shit, and avoided by society. there's nothing you can do to change that.
knowing this fact, and knowing there are others like you, literally gives me a small level of happiness.
(Score: 2) by coolgopher on Saturday May 22 2021, @11:42PM (2 children)
Funny how despite his alleged autism his social skills are way higher than yours right now... Chill out mate, sit on the lawn, have a beer, enjoy the weather, take a break from the screen - it'll help rebalance you.
(Score: 0, Offtopic) by fakefuck39 on Sunday May 23 2021, @02:29AM
I'm sorry, you must be confused. This is a website with random accounts from all over the world posting anonymous random shit for their entertainment. There is no such thing as social skills here, because this is not socializing. I know, you also are a social reject, so let me explain: here's a though test you can administer, to know whether you are socializing. If you are not socializing, social skills do not apply.
1) Who are you talking to? Do you know their name? Can you ask their name?
2) A person has a name. To socialize with a person, you should be able to know their name
3) A login id such as "coolgopher" is not a name. Fakefuck is not a name. In fact, coolgopher and fakefuck might be a perl script talking to itself, or it might be a chinese karma bot talking to itself
4) If you think you are socializing on soylent, and social skills are appropriate here, you need a rebalance of what is the real world, and what is a webpage where people get their kicks while taking a long shit. I suggest chilling out, getting a mate, sitting on a lawn, enjoying the weather, and taking a break from the screen.
What's clear here, is you spend so much screen time, you have merged the real world, and a random website. You fucking loser.
(Score: 0) by Anonymous Coward on Tuesday May 25 2021, @03:57AM
yo astraluan roo peni so big me rebalance yo long time mate. hw my cocial skill mate? want mate mate? when yo g watch race track mate you complane drivers bad drivers because drive unsafe for school zone? when go 8gag complain inappropriate wordage - unprofessional, never get hire for job speak like that? when go soy lent see people rip each new ass in hole complain not social skill good? when g china complane they english no good? when go texas barbie q restaurant cimplane bad vegan eats?
you're in abuse. arguments are 3 doors down.
(Score: 0) by Anonymous Coward on Monday May 24 2021, @12:37AM (2 children)
Please let us know where you "engineer" at so we can avoid you and any products you might have worked on. Thx!
(Score: 1, Offtopic) by fakefuck39 on Monday May 24 2021, @05:39AM
When you speak, there is no "us." You're not speaking on anyone's behalf here, there are no people "with you" - just several accounts you have, and you talking to yourself as AC. You've never had a real job, never dealt with enterprise environments, and are not socially apt enough to actually have a profession in an office environment. "Where" I "engineer" is about 40 countries, and 49 states, for over 20 years. My products include EMC storage and HCI, NetApp, Pure, Rubrik, Veritas, Commvault, and a few others. You are already avoiding product I work on, because you buy desktops and laptops, and have no idea what companies do, because none would hire a weird unshowered creepy weirdo.
(Score: 0) by Anonymous Coward on Monday May 24 2021, @06:13AM
TLDR
Creepy PC Support guy at the office who always returns your laptop greasy, smells like balls and Axe, and stares at women, goes online and pretends to be a C-level making enterprise infrastructure purchasing decisions.
So, Mr. VP, this us you speak of who spend millions each quarter on hardware - do y'all have private jets with the best bitches? How can I join your corporate VIP VP Club? I got these five dollars here, and some racy gifs of Cindy Crawford.
(Score: 0) by Anonymous Coward on Monday May 24 2021, @02:36PM (1 child)
No, you are just a persistent idiot who pollutes threads with bullshit. Trying to ram everything into your little niche knowledge. RTFA for once. They propose this as a replacement for the standard DRAM chips on the market today, across all segments. Not as an Optane competitor in your little Enterprise segment. Yes little. The bigger DRAM market dwarfs any potential Enterprise sales.
So how the heck are the planning to do that?
Well I'm no chip designer (and neither are you) but I think I know. This isn't Flash as we know it. Was deep reading on MRAM a while back, curious why it showed such promise and basically vanished, and hit the same concept. Apparently MRAM is similar, depending on how they optimize the design they can hit really impressive numbers, just not all at once. As they ramp capacity or speed to the sky it stops being RAM like, it starts showing a write limit. They have to pick a spot depending where they want to sell the stuff. But since at least one parameter sucks they aren't finding many spots to sell into.
Reading the article this stuff has to be refreshed like DRAM. If it has a practical endurance limit it would hit it pretty fast just refreshing. I'd guess they reworked the design parameters wildly from traditional Flash. Flash has to be hit hard enough to stick a bit for years and since write speed is the selling parameter now, they hit it hard and fast, and they need it to store a basically analog value reliably. And density of flash is insane now. All those things lead to short endurance. Flash EEPROMs can be written hundreds of thousand or millions of times, but they are a few hundred kbits at most and speed isn't an issue. Long term stability is. So for this application lets look how they tuned it. Long term storage? No. Insane density? No. Single bit cell? Yes. Write speed is still going to be pretty high.
(Score: 2) by fakefuck39 on Monday May 24 2021, @08:40PM
>replacement for the standard DRAM chips on the market today
which is exactly what Optane is for. DRAM chips can be used to execute programs, or they can be used for Cache for disk IO. This replaces the DRAM used as cache in storage arrays and HCI, with Flash. Just like Optane currently does.
>So how the heck are the planning to do that
They're not. They're planning to sell a competitor to Optane, which for the use case of IO cache, is a cheap competitor to DRAM.
But I stated all this in my original comment. Maybe in addition to reading and completely not understanding the article, you should try to read the comments before you reply to them. Else you might look like the fool that you are.
(Score: 0) by Anonymous Coward on Sunday May 23 2021, @01:17AM (1 child)
You're talking to facefuck39. That isn't Koolaid powder that he's snorting. At least, it isn't just Koolaid. It's probably mixed with Pixie Stix.
(Score: -1, Flamebait) by Anonymous Coward on Monday May 24 2021, @07:40AM
And you Good Sir, seem to be talking with yourself. Please read the prescription on the bottle of pills. You have to take them every day, or after a day without your meds, the voices come back and your right hand starts replying to your left. After two days, they'll get into a knife fight over your godly cock.
(Score: 1, Interesting) by Anonymous Coward on Sunday May 23 2021, @09:16AM
You could use it as a dedicated cache. Lots of SSDs have a DRAM cache on board, and this could replace that if the power, price, capacity, size tradeoff lines up. Even if it burns out, it only has to last longer than the flash drive it's attached to. Assuming the plague of crypto "wish it was currency" doesn't die, you could also use it for that, since algorithms like ethereum are read-heavy.
Forget about using it as actual RAM, though. Not only does RAM need unlimited writes, it also needs to be addressable at the byte level. Even if you had unlimited endurance, flash needs to be rewritten in blocks.