The folks over at Xbit Labs managed to spot Intel's "DDR4 Reference Test Platform" at this year's Intel Developer Forum, which as you might expect, is the world's first system to be powered by Intel's upcoming Haswell-E processor.
Though they weren't able to provide any details except for noting that the system was "running fine," they did note that the system was likely to have been powered by the X99 "Wellsburg" chipset and that the Haswell-E processor will continue to feature a quad-channel memory control that will provide "enough memory bandwidth for [an] eight-core central processing unit."
As is the norm for upcoming products, the report should be taken with a grain of salt. We expect to have more information closer to the release, which according to unconfirmed reports, will be in Q3 / Q4 2014.
Follow us @tomshardware, on Facebook and on Google+.
1) In mobile it is going to allow for super high densities with extremely low power, making 4GB of ram in a cell phone and 8GB in a tablet very possible. This will fix a lot of the multi-tasking and productivity issues that these devices suffer in. The performance of DDR4lp will probably be no faster than what we have today in low end desktops, but the power savings will allow for much better active-idle power use to vastly extend up-time on a charge.
2) DDR3 in the desktop space has seen horrible scaling from 800 to 1866, mostly due to Intel's memory controller architecture. Hopefully DDR4 will force them to rethink their architecture in order to gain the performance necessary to entice people to upgrade. Hardly anyone upgrades for added CPU horsepower anymore, these days upgrades are all about fixing system bottlenecks, be they in memory, SATA, PCIe, or USB. DDR3 has been a known bottleneck for quite some time, so hopefully DDR4 breaks that barrier.
3) DDR4 may (and I understand that this is wishful thinking) be fast enough to use as an extension for graphics memory. Next gen games and 4K monitors are going to dictate that high end GPUs have 8+GB of graphics memory... which would make these devices prohibitively expensive! The hope is that system memory would finally be fast enough, and that next gen controllers would be intelligent enough, to allow the GPU to have direct access to system memory as an extension to their own internal GDDR, which would allow next gen GPUs to continue to use 2-4GB of GDDR, while having fast access to the 8+GB that they really need. I only replace system memory with each core upgrade every 5-6 years, while I replace my GPU every 2-3 years. It is much more justifiable to buy a metric crap ton of system memory every 6 years than it would be to buy 8GB of very expensive GPU memory every 2.
Anywho, good to hear that DDR4 is coming down the pipe! Late 2014 it will make its way into high end desktops, 2015 will be right on track to upgrade my desktop, and then we will see it in phones within a year after that.
I'm sure i am not the only one with these thoughts.
1- lower power and higher densities are possible with DDR3 too with minimal specification bending but with DDR4 just around the corner, the industry is choosing to wait for DDR4 instead of add extra rungs to the DDR3 ladder.
2- the reason Intel CPU performance scaling does not scale much with massive bandwidth increase is simply because SB/IB/Haswell do not need anywhere near that much bandwidth to max out their processing resources - if you look at memory controller benchmarks, Intel's MC actually kicks ass on both latency and bandwidth.
3- system DDR4 on a 128bits bus and 3GT/s is still only 1/4th as fast as 2GB GDDR5 with 256bits and 6GT/s on the GPU card or 1/8th as much as high-end cards with 512bits memory bus, which is nowhere near as fast so you would expect a 75-90% penalty from transferring large chunks of data from system RAM instead of local GPU RAM. Also, accessing system RAM as an extension of the GPU's own means a ton of latency going through the PCIe bus to reach the CPU's memory controller which would further degrade performance.
The thing that will benefit most from DDR4 is IGPs across the whole spectrum from mobile SoCs to desktops CPUs. Low-end GPUs using DDR4 would also get modest boost from it but not as much since they already push DDR3 clocks close to DDR4 territory.
I did the same exact thing. I plan to do the same as you... unless I need more juice to run star citizen at 120hz 1080p maxed out with an occulus rift, at least the game wont be out for another year or two.
Right now Intel isn't pushing anything in the chip market, they'll wait for AMD to be in the game then release first. The longer they can sandbag tech the better for their bottom line.
DDR4 is a minor technology improvement and since Intel's current CPUs come nowhere close to bottlenecking on their DDR3 controller under typical workloads, the only reasons they would have to bother with DDR4 are reducing platform power and increasing bandwidth for their IGPs. Two things Intel will do sooner or later regardless of what AMD does.