The First DDR5 RAM Modules Promise Faster, More Efficient PCs (engadget.com) 104
Korean chip manufacturer SK Hynix has unveiled the world's first 64GB DDR5 RAM modules, marking a big step away from DDR4 DIMMs that have dominated PC memory since 2013. From a report: The DDR5-4800 chips support speeds between 4,800 and 5,600 Mbps with faster potential data rates than DDR4, while using less power. The technology also allows for modules of up to 256GB in size. The JEDEC standard for DDR5 RAM was officially published in July this year, but SK Hynix unveiled its first chips in 2018. Apart from the memory gains, DDR5 will have two 32-bit channels instead of a single 64-bit channel, making it easier to increase peak bandwidth. The modules themselves will also regulate voltage instead of the motherboard, allowing the DDR5 RAM manufacturer to control the all-important clock speeds. All told, it could make for some very interesting enthusiast RAM options. SK Hynix has already tested modules at 6,400 Mbps and has 8,400 Mbps speeds on the roadmap.
Like how DDR4 did and DDR3? (Score:5, Funny)
Newer Technology works better than Older Technology. News at 11.
I remembers when I had upgraded my Amstrad 1512CPC (Used an 8086 processor with 512k of ram, CGA Display) for a 486DX 50 MHZ. (4 Megs of RAM, SVGA display and actual hard drive) This was a huge upgrade for me, the biggest one I can remember. This changed how I could use a computer and what I did with it. The rest of the upgrades, the Computer was a bit faster, more ram, etc... But my usages didn't change that much, as when I upgraded it, I only had a few major changes.
We hear about new CPU, RAM, Batteries... While they are much better than their previous versions, they are not that much better than we actually think, As the computer will often have other bottlenecks that havn't been resolved that we haven't been about to use yet. Software not supporting it, Slow Disks, Increased Power Usage...
Re: (Score:3)
Honestly, I'm surprised that that is the most notable upgrade you've experienced.
I would have thought HDD->SSD would be a more notable upgrade.
Re:Like how DDR4 did and DDR3? (Score:5, Informative)
I would have thought HDD->SSD would be a more notable upgrade.
From an 8086 with a floppy drive to a 486 with a hard disk? Are you kidding? Frankly I wonder if he even had a 3.5" floppy or a 5.25 one.
Mine was even bigger, I transitioned from a BBC Master to a P5 with 72M of RAM and a 900M hard disk and a Riva128 card.
Re: (Score:3)
It was a 5.25 Double Density disks. I did have duel floppies so I could put the Data on the B: and the Executable on the A: then
A> B:
B> A:\Runfile.exe
which would have my default folder on the B drive, while the executable runs on the A drive, but reads the data off the B
Re: (Score:2)
When I complained about how slow and tedious it was to work with the Hourly Analysis Program without a hard drive, my boss said we don't need to spend money on a hard drive, two floppy drives is plenty. It was a complex enough program that you had to constantly swap discs: take out the project data disc, put in the weather data disc, put the project data disc back in, swap one disc for part of the program for another, then swap again, back and forth. When the program actually ran th
Re: (Score:2)
It was a 5.25 Double Density disks. I did have duel floppies so I could put the Data on the B: and the Executable on the A: then
A> B:
B> A:\Runfile.exe
Ah dual drives... lucky bugger. I only had one on my Master, which made copying floppies astoundingly tedious. On the other hand the OS and editor were in ROM, so that wasn't so bad.
Re: (Score:2)
CGA -> SVGA is a pretty big leap.
160x100/320×200/640×200 in 16/4/2 colors. Image that. And then going to 1280x1024.in at the least 4 colors, and on top of that you can finally get 3D.
Re:Like how DDR4 did and DDR3? (Score:5, Interesting)
It's because you probably never used a 8086 running at 8MHz with low resolution graphics in 4 colours and no hard drive.
The jump to a 486DX running at 50MHz (6.25 times as fast by the clock speed alone) with 8 times more RAM, high resolution graphics in 256 colours and a hard drive is huge. We were manually swapping floppies between booting the computer and loading a game. For big games, we were also swapping floppies during gameplay (ex: King's Quest IV came on nine 5.25" floppy discs).
In comparison, the jump from HDD to SSD is a joke.
Re: (Score:2)
Re: (Score:2)
How about "having a mechanical computer do naval gun targeting calculus after a knobs setting data are in position instead of having to crunch the math manually"?
Re: (Score:2)
Oh Kings Quest IV. The only game I got that Actually supported the advanced Features on the Amstrad PC Video Card, (As it really was an improved CGA card that could do most of what EGA could do, but not compatible). Seeing a Game in 16 colors was like looking into the future than. Luckally I had learned I could swap the DRV files with other Sierra games, thus have other games in 16 colors too.
Re: (Score:2)
Sounds like Tandy 1000 graphics to me.
Re: (Score:2)
There are two big jumps that stand out for me in my experiences:
Going from a TI 99 4a to a Commodore Amiga.
Going from raycasting on a SVGA card to "real 3d" using one of the original Nvidia cards.
Spinning disks to SSD does not even register on those scales.
Re: (Score:3)
Not so much for me. I had a SATA Solid State Drive on my old Decade Old Laptop. Which was faster than a Mechanical Disk, but not super speed. I then have an NVME Solid State on my new laptop, still much faster but it was a gradual setup up.
Re: (Score:2)
I once replaced the mechanical 2.5" HDD with a compactFlash-to-IDE adapter in an old Pentium Thinkpad.
Re: (Score:2)
How long did it last?
Re: (Score:2)
Long enough that I stopped using the Thinkpad years later and it still worked. Keep in mind that I disabled the Windows 98SE swap file.
Re: (Score:2)
Wow impressive. I wonder how long it would have lasted with the swap file.
Re: (Score:2)
The transition from running everything off floppy disks to having a hard drive is a much, much bigger deal than HDD -> SSD.
Plus with a CGA display, most things you did would be text based, with very minimal graphics capability. The 486 was almost certainly GUI based.
HDD -> SSD is a great performance upgrade, especially for one that only involves replacing one component. But for most people it's not changing their workflow that much. The changes the parent was talking about completely change how you us
HDD - SSD (Score:3)
really, it was only a modest gain for those of us who went through the floppy to HDD transmission, let alone the transition *to* floppies!
My transition from toggle switches to keyboards, though, wasn't as big a deal . . . I wasn't really *using* those . . .
And in other such things, the transition to 2400 baud was bigger than the later bumps in speed--I was waiting for characters to come onto the screen at 1200, but it took a bit of effort at 2400
hawk, dating himself again
Re: (Score:2)
If you think the jump from 1200 to 2400 baud was big, imagine going from 300 to 14.4k.
Re: (Score:2)
oh, definitely.
But 300 (and less) to 1200 meant waiting for the next character, while 2400 to gps puts the characters on screen (rather, has them available) fast enough for casual reading. That's where the big difference came in . . .
Re: (Score:2)
And then you start playing door games and ANSI "graphics" make it feel like you're back to 300bps.
Re: (Score:2)
Now I'm wondering what modern nethack would be like at lower speeds . . . I fist encountered hack on a 9600 baud vt100, but now there can be so many monsters moving on each move, and the animated fireballs and such . . . at 300, that fireball would just creep down the hall . . .
Re: (Score:2)
I would have thought HDD->SSD would be a more notable upgrade.
Millennials. The only reason the HDD > SSD looks like a performance improvement is because it's the only component that didn't go through a leap.
Sit in front of a screen that only displayed 4 colours at 320x240 and then go to 16million colours at 800x600
At the same time make a switch to reading from a floppy disk into 512kb to reading from a HDD and actually having RAM and a PC fast enough to draw proper graphics on the display.
LOL SSD, incremental at best. The jumps we made in the 80s and 90s would blow
Re: Like how DDR4 did and DDR3? (Score:2)
Re: Like how DDR4 did and DDR3? (Score:2)
Re: (Score:2)
Re: (Score:2)
Newer Technology works better than Older Technology. News at 11.
For RAM this actually is news. In most cases the generational jump in RAM has resulted in initially decreased performance in terms of crippling latency until companies managed to optimise and surpass the benefits of the previous technology.
I fully expect DDR5 to not be good on launch.
Re: Like how DDR4 did and DDR3? (Score:1)
In the old days of PC's that used memory chips, if graphics memory was used on the mainboard instead of regular dram, the pc was very fast.
The introduction of the ram drive helped those old PC's performance versus the hard drive but the cost was quite steep.
These days, solid state drives helped to speed up PC's running all these huge code line software. The addition of using ddr5 (graphics memory) makes me wonder why it took so long to figure that out.
5,600 Mbps? (Score:1)
Re: (Score:2)
Nope, 5.6Gbps, not bytes.
Don't forget that the spec is that data rate per data pin.
There's quite a few data pins on the module.
Re: (Score:1)
Re: (Score:3)
It's not bits, it's transactions per lane. 5600 MT/s = 358400 Mbps (since there are 2x 32bit data lanes in a DDR5 module), or 44.8GB/s
apple mac pro priceing $2000 each min 4 to fill ch (Score:3)
apple mac pro priceing $2000 each min 4 to fill channels.
Re: (Score:2)
no worries, they'll be $100 a pop before you know it
Re: (Score:2)
One more thing - Apple will solder these to the motherboard to make sure they are not user up-gradable. Plus, there will be no free slots to add more RAM.
Overclockers pissed? (Score:2)
The modules themselves will also regulate voltage instead of the motherboard, allowing the DDR5 RAM manufacturer to control the all-important clock speeds.
Pay extra for unlocked voltages? The average user doesn't touch voltages or timings so this seems like another tool to squeeze the enthusiast market for cash.
I'd like to believe this is about stability but seriously doubt it. The summary doesn't mention that DDR5 will come with some ECC standard and I think that has been long overdue.
Overclockers... and regular users? (Score:2)
DDR5 will adjust all of that automatically to reach its "recommended" values.
As for the overclockers, I'd be surprised if the new protoc
Re: (Score:2)
Doc Brown tinkered with clocks and we all know what happened.
Re: (Score:2)
Re: (Score:3)
The average user doesn't touch voltages or timings so this seems like another tool to squeeze the enthusiast market for cash.
Pretty much no one is squeezing the enthusiast market for cash, except maybe Intel, but they have always been shits.
I'd like to believe this is about stability but seriously doubt it.
Having a dedicated voltage regulator on module is huge for improving power stability at high data rates. Come up with conspiracies all you want, electrically there's a lot to gain from this approach.
Re: (Score:2)
The summary doesn't mention that DDR5 will come with some ECC standard and I think that has been long overdue.
If that is like GDDR5, then the ECC only protects the interface and not the memory array. That is required because the timing margin of the interface is so small that it drifts out of calibration as the temperature changes and require retraining.
Return of the RAM disk. (Score:2)
I wouldn't mind if these get cheap in a few years having 256 gigs, and creating a nice RAM disk for playing 'old' games like RDR2 with basically zero load times, other than script execution on level load (lazy Unity games will still load slow).
We'll probably need much more convenient tools for managing and automating the process though - since it's a bit fiddly now. After a reboot, you'd still have to wait to have the data re-mapped to RAM, but if automated, it would just be a progress indicator you'd see
pci-e disks are faster then ram disks do to less o (Score:1)
pci-e disks are faster then ram disks do to less over head.
Re: (Score:2)
I have tried it. Any game that also connects to some server will see insignificant decrease in loading time. The bottleneck is the server it connects to for validation and other crap.
Re: (Score:3)
You already can have that with a NVMe. PCIe4 ones run at 5 GB/s, and I don't think any game is really going to benefit from anything faster, anyway, since at such speeds you easily run into bottlenecks elsewhere, like decompression and deserialization.
Re: (Score:3)
Re: (Score:2)
I just started testing NVMe and so far the only thing which is noticeably faster are benchmarks.
This is great, but . . . . (Score:5, Interesting)
Am I the only person alive who has extreme trouble figuring out which RAM is compatible with which CPU and which motherboard?
I don't have to do that very often, but when I do, I'm typically at a loss, and often have to rely on parts vendors with a vested interest in upselling anything they can.
Re: (Score:3)
Re: (Score:2)
Which only creates very perverse incentives. I mean if Corsair paid Asus to be on the list, and Crucial did not, Corsair might pay more if Asus made their motherboards perform better with their RAM instead of Crucial's even though they are the same specification. It doesn't have to be huge things -
Re: (Score:2)
If only the motherboard manufacturers tested ram at various speeds and published that info on the same page they advertise their motherboards.... maybe they could call it the "Qualified Vendor List" or something.
Yes...but the GP isn't wrong.
Even if the motherboards list out their compatibilities, RAM manufacturers are generally terrible at listing out what speeds a given module supports. Sure, it'll list out its maximum speed, but how about the others it's compatible with? That's super helpful to know, especially during the midterm upgrade cycle when extra RAM and a GPU will give you another 2-3 years out of a computer. In many cases, a computer has been compatible with the original RAM *or* the additional RAM, but
Re: (Score:2)
I usually find the qualified vendor list to be useless because all of the listed parts are unavailable.
Re:This is great, but . . . . (Score:5, Informative)
A motherboard will only be compatible with one type of DDR specification, which currently is DDR3 and DDR4. A DDR3 ram stick wont fit in a DDR4 slot and vice-versa, there is no adapter. I recommend looking at PC Part Picker [pcpartpicker.com], it has a compatibility filter.
Re: (Score:2)
Re: (Score:2)
Generally speaking you can use faster RAM than spec with no downside; it will just run slower. Some boards will support slower than spec speed. A few are picky and only support one speed, but that's mostly a server thing.
Main thing is make sure it all matches, and sometimes that means same *batch*. I have a bunch of that Ballistix DDR3, and in some PCs the made-in-USA and made-in-China sticks do NOT get along, even tho they're the same model and part number. (Other PCs don't care.)
Having started when the bi
Do you want to get a meme? (Score:2)
Because that's how you get a meme:
https://imgflip.com/i/4ho855 [imgflip.com]
Re: (Score:3)
Both AMD and Intel list the maximum supported speed on their web site, for every CPU:
https://www.amd.com/fr/product... [amd.com]
https://ark.intel.com/content/... [intel.com]
Motherboards makers sometimes advertise higher frequency, but this is overclocking and most people shouldn't care, it's usually not worth it anyways.
If you get lower frequency memory, it will still work, but you will get a performance hit. DDR4-3200 is about the same price as anything slower anyways.
Re: (Score:2)
Re: (Score:2)
Am I the only person alive who has extreme trouble figuring out which RAM is compatible with which CPU and which motherboard?
I think the answer is yes.
The only thing difficult about selecting RAM is how to find the one that best overclocks on your system. Otherwise just buy a stick of DDR4 and go at it. If you really 100% want to ensure compatibility just buy one of the many sticks listed on the QVL for the motherboard (but those are only tested sticks, not an exhaustive list of sticks which work).
Shit man I have the same Corsair sticks in all my PCs right now, intel, AMD, Zen 2+, Zen, didn't even give it a second thought.
Oh actu
Re: (Score:1)
I'll put in a plug here for PC Part Picker [pcpartpicker.com]. (I'm not affiliated with it: I've just used it and found it useful.)
Use the "System Builder" link. Add a motherboard (from a list of 3294 options). Choose to add RAM ... and it'll only show you compatible RAM. It'll even handle (though I haven't thoroughly tested this) obscure compatibility issues like whether there's room for a particular aftermarket CPU cooler within the case you've selected.
Re: (Score:2)
I love the warm feeling i get when i see 64 (Score:3)
This warm feeling when we see 64.. 64kB in the 80's, 64MB around 2000 and 64GB in 2020.
Hope to live see 64TB modules coming in 2040.
Re: (Score:2)
Unless we're willing to scale up the size of the computer I don't see that happening any time soon.
Re: (Score:2)
That's basically the same thing people have said for the last four decades.
Re: (Score:2)
Fair, but there's no denying that the demonstration of Moore's Law has been slowing and there is a point you simply cannot fit more atoms into a square millimeter. Right now the biggest breakthroughs for computers will be either nanomachines or finding a metal that has multiple tiers of conductivity that is usable for computers, much like how silicon needs 50 volts.
Of course the biggest issue in computers right now is people refusing to optimize. Websites especially assume that everyone has uncapped interne
Re: (Score:2)
Being in that field, I can tell you that one of the problem is management. They buy top-of-the-line computers with lots of RAM, etc. because they don't want their hourly employees to be slowed down by the equipment. The side effect is that those employees are detached from the reality of the computers and phones used by Joe Average.
People who work on graphics and compile software should have top of the line computers. People who work on the web front-end should not.
Re: (Score:2)
Do you adore your 64?
Did you not havr basic units in school? (Score:5, Interesting)
8400 Mbps?
So 1.05 GB/s in sane units.
Wait, why is that slower than an SSD?
Could I wire 4-8 SSDs into a DDR5 module of massive capacity without memory loss?
Re:Did you not havr basic units in school? (Score:5, Informative)
That number is per (differential) pair of traces, with 64 data pairs -- or 80 including ECC -- per DIMM.
Re: Did you not havr basic units in school? (Score:2)
So you are saying I need 512 SSDs? :D
Re: (Score:2)
Sure, if you want to match the speed of one DIMM. A lot of desktop computers require DIMMs to be installed in pairs, and some get maximum benefit from having four or more DIMMs in parallel :)
On the other hand, if you look at higher-end SATA SSDs, they can transfer 500+ MB/s rather than 250, and if you have enough NVMe channels, those SSDs are an order of magnitude faster than SATA.
Re: (Score:2)
latency.
Re: (Score:2)
Re: (Score:2)
I should note that it makes sense for the DDR specification to be in "transfers per second" because they do not know how wide the bus will be. Once the memory is packaged onto a DIMM then that DIMM will have a total bandwidth associated with it. As a result you have 2 ratings - one for the DIMM and one for the memory type.
This notation becomes somewhat confusing for end users because most people just think of the bandwidth of a DIMM and the number of channels to the CPU. But things make a little more
Re: (Score:3)
Channel rates for data interfaces are never specified in GB/s. They are also not specified in Mbps to stop people like you (and in your defense TFS) making the comparison you just. DDR speeds are measured in MT/s (millions of transactions per second) which is also the marketing name given to it.
DDR4 3200: 1600MHz modules. Double Data Rate = 3200MT/s. 2x32 transfer lanes = 204.8Gbps peak transfer rate, or 25.6GB/s since you seem to work in consumer units rather than actual bits.
DDR5 8400: 4200MHz modules. Do
Re: (Score:2)
My brain hurts... thanks for the info. Boils down to "amazingly fast" !!
From the viewpoint of my frankenputers, all that matters is "does the board support this type and speed?" Then again I'm not a gamer trying to squeeze out that last fps.
Re: (Score:2)
Board support for speed is always a bit of a curiosity. Fundamentally the spec for DDR4 has a base speed of 2166MHz and and everything else is considered an overclock. Fundamentally AMD CPUs support a max speed of 3200MHz and everything else there is considered an overclock. And this despite my motherboard having 4133MHz chips in the Qualified Vendor List.
However this is one of the areas where that "overclock" is actually a norm. DDR chips come with an XMP (eXtreme Memory Profile) which tells the BIOS the t
Re: (Score:2)
More good info, thanks -- I'll remember this when I crawl up to that next build level. Good reason to buy the faster RAM, if you have to choose; might as well run it at the speed it's really capable of.
Yeah, the QVL are useful, especially if you need stability, or are running into some mystery error.
Tho occasionally not accurate... friend has an older Lenovo that's spec'd for 8GB of N-speed, but will only boot with 6GB of slower-speed.
Re: (Score:2)
If you're planning your next build and it's AMD, it's worth remembering also that due to AMD's architecture the speed of the chip also benefits from faster RAM. The reason being that inter-chiplet and cache communication is done over the Infinity Fabric, and the Infinity Fabric is clocked at the RAM speed. So even applications which don't depend on RAM speed can stand to get a slight performance boost with faster RAM on the Ryzen platform.
This does not apply to Intel.
It's also why RAM overclocking is such a
Re: (Score:2)
Not me... was cured of AMD many moons ago. Will use one if it falls on my head, but won't pay money for it.
That's a peculiar way of doing things -- how did it evolve? It seems rather... indirect.
Re: (Score:3)
Not sure what "cured" means. I tend to buy whatever price/performance/features dictate at the time but whatever.
How did it evolve? Actually by natural progression. While Intel was sitting on their laurels focusing on boosting clock speeds alone with their R&D on expanding core count limited exclusively to their server offerings such as the Xeon Phi, AMD looked to introduce incredibly high core counts at a budget. Intel's interconnect (Ring Bus) at the time was incredibly high performance for allowing co
Re: (Score:2)
Very interesting details and very informative (someone mod parent up, please). Quite a difference in design philosophy. Glad to see they're making themselves competitive; keeps Intel from, as you say, resting on their laurels (or perhaps more often, having a too-successful batch run so they have zillions of 'em to sell and little incentive for improvement).
Number of reasons I quit buying AMD a very long time ago, and have seen no incentive to switch back (I buy last-year's model, if not older, so the relati
Re: (Score:2)
Yeah AMD has a checkered history. Intel however does have a checkered present. The industry has for a large part solved the quality issues by doing away with 3rd party chipsets. God I don't miss Via (on intel or otherwise).
Two things worth mentioning though, errata isn't ancient. They are still published for current generation products, and they have to be. An errata is a deviation from the formal specification. AMD from the 90s had a long list of minor bugs not because they were "running Intel's code" but
Re: (Score:2)
Most of my hardware is a bit outdated, so no surprise there! :)
Good to know AMD is catching up. Better to have competition that's neck-and-neck; more incentive to outdo the other guy. Still, not enough to entice me to AMD... they've kinda burned their bridges with me.
I don't need super-video, but is indeed nice to have 'em not double as a camp stove, especially as I'm weary of noisy GPU fans and have developed a taste for fanless.
Faster, More Efficient (Score:1)
What could the difference possibly be in this context between "faster" and "more efficient"?
"Faster" means "more (of something) per unit of time". "More efficient" is more general — because some other unit can be in the denominator. But, as far as RAM chips are concerned, what other units matter?
Re: (Score:3)
Power consumption
Re: (Score:1)
Does it matter? DDR3 sticks were using about 4 Watts [tomshardware.com]. Whether DDR5 uses 5 or 3, who really cares?
Re: (Score:2)
Some of the improvements to the various revisions of DDR reduce idle consumption as well as overall consumption. If you are trying to squeak out a few less milliwatts, every improvement is important (although that will likely just be wasted at the software layer).
Re: (Score:2)
That's literally 20% worse or 25% better power consumption. 4 watts becomes quite a dominating factor if you want to get a typical 40Wh battery powered device through an average workday without a recharge.
New product better than previous product shock. (Score:1)
One expects Apple and others trying to sell overpriced tat to trot out the usual lame "Faster and better than anything we've every made" BS.
But Slashdot aren't selling anything.
Are they ?
Re: (Score:2)
They sell eyeballs.
Slashdot: "I just do eyes..."
Re: (Score:2)
Slashdot is not exactly the Nexus of news sources. If only you could see what we've read with our eyes.
Meh. There's DDR6 in This FPGA Card (Score:2)
Re: (Score:1)
That's 'G' DDR6 - not the same thing.
RowHammer (Score:2)
Is the new DDR5 still susceptible to the RowHammer attacks like the previous DDR versions ?
Re: (Score:2)
Rowhammer is about the memory chips; it's not a weakness of the memory bus that connects the memory to the CPU. As such, rowhammer isn't a problem that the memory bus like DDR5 can solve.
DDR4 domination since 2013? (Score:2)
In my experience, the transition from DDR3 to 4 dragged on for years, because the newer generation didn't offer that much improvement over the previous one. For example, I bought 3 new machines in 2016, all of which used DDR3, and in 2017 I got my first DDR4 machine.
In particular, memory latencies haven't improved much since regular SDRAM. While each DDR generation has doubled the clock speed, the latency in terms of clock cycles has doubled as well. I remember numbers such as CL2.5 in DDR, CL5 in DDR2,
Comfortably Numb... (Score:1)