Looks good and fast, which makes me worry about cost.

A bit late to the party
Samsung announced the exact same memory 3 days ago lol.

Wow, I hate this sentence:
"Especially, SK Hynix is to continuously maintain its technology leadership in the mobile market by developing the world's first 8 Gb LPDDR4..."

I thought Samsung was first....

nunomoreira10A bit late to the party
Samsung announced the exact same memory 3 days ago lol.

When it comes to this kind of stuff, what is the big deal with being first?

james888When it comes to this kind of stuff, what is the big deal with being first?

that's the thing
samsung and hynix should have announced they managed to produce it, not that they are the first.
the more the better.

While announcing these LP modules are nice, I wish they would announce some speed/voltage/yield numbers (which speed can be extrapolated from) for desktop ddr4. All we have is Jedec specs, and those seem woefully low-balled (even at 4266) in both requirements for future core designs and what speeds whatever process will be used is capable when it finally hits mass production for consumers. Perhaps everyone will migrate to HMC and the like (stacked chips at that frequency or lower) and other close-proximity shared buffers, but it would still be nice to know.

To give people a rough idea of bandwidth requirements and perspective to what this means for mobile, the flagship desktop Kaveri seems to be clocked for use with 3000mhz ram (over it's 128-bit bus).

At 3200mhz, granted probably on a 64-bit bus much of the time, that gives you an idea of what will be possible for mobile, which is pretty crazy. While you could more-or-less (granted because of latency differences and their effect on a cpu you may need slightly more bw) substitute any full cpu design aimed towards 1600mhz ddr3, Literally half a Kaveri, or such a design created within similar limitations for mobile/cheap pcs.

When you think about the chips these will go with; 16-14nm, or perhaps rather 20nm-finfet-named-14nm from both TSMC and the CPA (IBM, Samsung, GF), this will be a huge boon for mobile....huge. Think of what a phone/tablet/convertible/small notebook will be capable of with that much bandwidth aligned with some super space-saving transistors, even at low clocks....using very little power. Pretty amazing.

DDR4 is primarily for servers. It offers no tangible advantage for desktop use over DDR3 LV. Extensive testing has shown that DDR3 running at 1600 MHz. is not a system bottleneck even for enthusiast desktop PCs powered by a CPU. If you have an APU then up to 2133 MHz. can show a small gain in system performance. Anything else is just pissin your money away on empty promises.

JorgeDDR4 is primarily for servers. It offers no tangible advantage for desktop use over DDR3 LV. Extensive testing has shown that DDR3 running at 1600 MHz. is not a system bottleneck even for enthusiast desktop PCs powered by a CPU. If you have an APU then up to 2133 MHz. can show a small gain in system performance. Anything else is just pissin your money away on empty promises.

This is a PR about DDR4, about the memory of next gen. You can't not use it with post-haswell CPUs. So, you're pretty much complaining to a concrete wall.

JorgeDDR4 is primarily for servers. It offers no tangible advantage for desktop use over DDR3 LV. Extensive testing has shown that DDR3 running at 1600 MHz. is not a system bottleneck even for enthusiast desktop PCs powered by a CPU. If you have an APU then up to 2133 MHz. can show a small gain in system performance. Anything else is just pissin your money away on empty promises.

"640K ought to be enough for anybody"

alwaysstsTo give people a rough idea of bandwidth requirements and perspective to what this means for mobile, the flagship desktop Kaveri seems to be clocked for use with 3000mhz ram (over it's 128-bit bus).

At 3200mhz, granted probably on a 64-bit bus much of the time, that gives you an idea of what will be possible for mobile, which is pretty crazy. While you could more-or-less (granted because of latency differences and their effect on a cpu you may need slightly more bw) substitute any full cpu design aimed towards 1600mhz ddr3, Literally half a Kaveri, or such a design created within similar limitations for mobile/cheap pcs.

DDR, DDR2, and DDR3 all have a 64-bit data buses per rank, this hasn't changed for ages... Kaveri's "128-bit" bus is called dual-channel memory much how quad-channel on my skt2011 machine is 256-bits wide... Just saying...

JorgeDDR4 is primarily for servers. It offers no tangible advantage for desktop use over DDR3 LV. Extensive testing has shown that DDR3 running at 1600 MHz. is not a system bottleneck even for enthusiast desktop PCs powered by a CPU. If you have an APU then up to 2133 MHz. can show a small gain in system performance. Anything else is just pissin your money away on empty promises.

It's not about more performance. It's about shoving more memory in a smaller area that uses less power.

Every time you post, you're angry. I think you need to breathe for a minute.

JorgeDDR4 is primarily for servers. It offers no tangible advantage for desktop use over DDR3 LV. Extensive testing has shown that DDR3 running at 1600 MHz. is not a system bottleneck even for enthusiast desktop PCs powered by a CPU. If you have an APU then up to 2133 MHz. can show a small gain in system performance. Anything else is just pissin your money away on empty promises.

Yeah, no. System performance as in HSA should improve with faster RAM, otherwise, in normal CPU scenarios, stuff like WinRAR will be the only thing seeing a perceivable improvement. Other than that, gaming on iGPUs see the biggest improvement by faster RAM, Hell, from DC 1600 CL9 to DC 2133 CL11 you get a over 25% increase in some cases, on top-end Trinity and Richlands. If you overclock the iGPU, it becomes even bigger.

AquinusDDR, DDR2, and DDR3 all have a 64-bit data buses per rank, this hasn't changed for ages... Kaveri's "128-bit" bus is called dual-channel memory much how quad-channel on my skt2011 machine is 256-bits wide... Just saying...

Hmmm, that may be the case for desktop, but at least on current 32bit systems (ARMs in particular), without special extended adress spaces and workarounds, it's 32bit per channel (not sure about the 32bit Athlon XP/Pentium 4 era and before tho), as I've stated before.


Edit: 2 manufacturers saying they can make these chips? Who the Hell gives a crap who's first, having a "who's second" is much more important, it means potential buyers won't get imaginary premiums/overcharged, (unless they're both spoken and prices are rigged to begin with).

Edit2: Great news for the likes of AMD Beema/Mullins and beyond, lower power usage on RAM means more thermal power headroom for the APU itself, as well as faster speeds and bigger RAM sizes (OR simply, more battery time). Well, that is, if AMD gets with the program and starts using LPDDR at least for Mullins to begin with.

NeoXFHmmm, that may be the case for desktop, but at least on current 32bit systems (ARMs in particular), without special extended adress spaces and workarounds, it's 32bit per channel (not sure about the 32bit Athlon XP/Pentium 4 era and before tho), as I've stated before.

I'm pretty sure I've explained the difference between a 32-bit data bus and a 32-bit address bus in another thread. First of all, it's important to note that many ARM CPUs don't use a full rank of memory ICs where laptops and desktops do so the size of the data bus very well may not match the size of the address bus (IC manufactures don't tend to add more pins than are necessary). Second of all, a 32-bit CPU can pull more then 32-bits worth of data from memory at once assuming the bandwidth is there. It can not address more than 32-bits worth of data though, same with a 64-bit CPU, it can not address more than 64-bits worth of memory addresses but can can pull at much data from memory as it has bandwidth.

As I said, DDR1, 2, and 3 ranks have always been 64-bit per channel, which is the width of a single rank on a memory module, not a single IC.

AquinusI'm pretty sure I've explained the difference between a 32-bit data bus and a 32-bit address bus in another thread. First of all, it's important to note that many ARM CPUs don't use a full rank of memory ICs where laptops and desktops do so the size of the data bus very well may not match the size of the address bus (IC manufactures don't tend to add more pins than are necessary). Second of all, a 32-bit CPU can pull more then 32-bits worth of data from memory at once assuming the bandwidth is there. It can not address more than 32-bits worth of data though, same with a 64-bit CPU, it can not address more than 64-bits worth of memory addresses but can can pull at much data from memory as it has bandwidth.

As I said, DDR1, 2, and 3 ranks have always been 64-bit per channel, which is the width of a single rank on a memory module, not a single IC.

Yeah, I've got something better than words that say otherwise, actual hardware. And I'm gonna believe that for the time being.
Well, that and the fact that I really got better things to do (at the moment at least, this might change LOL) than start reading through pages and pages of articles and vague/ambiguous documentation.

Edit: typo.