Adata's DDR4 Memory Modules Are Finally Arriving
Adata's DDR4 memory modules are coming to shelves soon.
Back at Computex 2014 we saw DDR4 memory in action, where we made an Adata employee very nervous by making it appear as if we were trying to see information they were obviously trying to hide. Now, the company has announced this memory, and we finally have the details.
The memory is DDR4-2133, which isn't much faster than high-end DDR3 memory. In fact, there is faster DDR3 memory. That's normal though, and we've seen this pattern in the past with the release of DDR3, as well. Mind you, one of the biggest goals of DDR4 is lower power consumption, so rather than running at 1.5 or 1.65 V, this memory runs at just 1.2 V. The memory modules will provide a bandwidth of about 17 GB/s. Another target of DDR4 memory is higher density, allowing for larger-capacity DIMMs, though we will probably only be seeing that a bit further down the line.
Timings for the modules are set at 15-15-15, and they are built to be used with the upcoming X99 platform.
These memory modules will be available in kits of two DIMMs, with total capacities of either 8 GB or 16 GB. There is no exact word on pricing or availability yet. We have reached out to Adata and will let you know when we hear back.
Update: Despite the image above, Adata will only be shipping single-DIMM kits. The 4 GB module will cost $69.99, while the 8 GB modules are priced at $129.99.
Follow Niels Broekhuijsen @NBroekhuijsen. Follow us @tomshardware, on Facebook and on Google+.

Well, the DDR4 transition has to start somewhere... even if the starting point seems to make little to no sense.
From what I have read, it seems like LGA2011-3 is effectively a new socket which requires x99, meaning IB-E is end-of-line for current LGA2011 motherboards.
We will have to address heat dissipation because of TSV stacking NANDs and so higher heat density. I really hope we abandon that aluminum heresy going on this days to address the problem with copper plating like Patriot Viper Xtreme 2000 9-10-9 have and they are regular height. We will probably see LP banks later and with lower capacity than the maximum 128GB unless in the meantime 20nm NANDS go down another two steps.
What do you guys think? Aim for the per core slower octa or the quicker hexa core for gaming!?
Regular DIMMs already have built-in copper heatsinks: the ground and power planes in the PCB itself. The DRAM chips' packaging is made of plastic or epoxy, which are far worse thermal conductors than aluminum so the bulk of the DRAM's heat goes in the PCB through the solder balls anyway. Extend the PCB by 5-10mm, fill the extra area with copper planes and you can ditch the silly heat-spreaders altogether.
Well, the DDR4 transition has to start somewhere... even if the starting point seems to make little to no sense.
I hope you do realize that 2133-15 DDR4 will not have a 66% performance penalty over 2133-9 DDR3.
15 cycles latency / 1066MHz clock = 14ns latency for DDR4
9 cycles / 1066MHz clock = 8.4ns latency for DDR3
That makes DDR3 40% faster on first-word latency. Bandwidth is the same since both are the same width and same data rate.
Back in topic: If you think an i7 5930k is the best pick for gaming you have been really living under a rock. They compared an Athlon 5350 with a 750 Ti with a 4960X, a 25W TDP cpu against a 130 or 150W TDP cpu. One is 4 cores 4 threads the other 6 core 12 threads and the results were pretty impressive giving the Athlon only have 4 PCI lanes and the 4960X has 10 times more. There are a few games that like to have more processing power, but almost all depends on GPU power. Now the options available are 256 bit mid-low to high level GPUs, 128 bit low level GPUs and 384 and 512bit extreme level GPUs. Well the bits that should enable us to the 4K gaming world simply are not working as expected, and the 384 bit aswell. While you cannot call yourself a gamer at that level of money spent but just a jocker who spent loads of money on things that will get old really soon it's better to address what I and some wise people call "diminishing return". You are really going to spend loads of money on a processor that will initially be starved by the ram, giving it will have a legendary good controller, you will get initially the same amount of ram as nowadays PCs and lower performance (it depends on the ram controller but the latency is really huge).
You are going to get a CPU fast and powerful like the Q9550 was, even the Q6600 do as an example, overclockable and one of the most powerful in the market just for a few fps more, if any, and get obsolete by the day that many threads will be used. Well, the wallet is yours, I cannot tell you anything else than this two words:
DIMINISHING RETURN
15 cycles latency / 1066MHz clock = 14ns latency for DDR4
9 cycles / 1066MHz clock = 8.4ns latency for DDR3
That makes DDR3 40% faster on first-word latency. Bandwidth is the same since both are the same width and same data rate.
Well, the DDR4 transition has to start somewhere... even if the starting point seems to make little to no sense.
I hope you do realize that 2133-15 DDR4 will not have a 66% performance penalty over 2133-9 DDR3.
You cannot calculate the real performance loss (bacause nowadays there is no gain) because even if it's an Haswell architecture the ram controller may behave differently. I think the performance penalty is more mathematical than real world experience, but it's there. We have not yet seen what Vpp boost does so the first word access latency could be addressed in this way. So we cannot really compare DDR3 with DDR4 because they work differently and hopefully much better with the latest having an advantage even with NAND chips apparently weaker.
Note that even with lower voltage, the DDR4 are going to consume more power than the same DDR3 memory.
We all hope that the semiconductor problem gets addressed by smart logic.
15 cycles latency / 1066MHz clock = 14ns latency for DDR4
9 cycles / 1066MHz clock = 8.4ns latency for DDR3
That makes DDR3 40% faster on first-word latency. Bandwidth is the same since both are the same width and same data rate.
Well, the DDR4 transition has to start somewhere... even if the starting point seems to make little to no sense.
I hope you do realize that 2133-15 DDR4 will not have a 66% performance penalty over 2133-9 DDR3.
You cannot calculate the real performance loss (bacause nowadays there is no gain) because even if it's an Haswell architecture the ram controller may behave differently. I think the performance penalty is more mathematical than real world experience, but it's there. We have not yet seen what Vpp boost does so the first word access latency could be addressed in this way. So we cannot really compare DDR3 with DDR4 because they work differently and hopefully much better with the latest having an advantage even with NAND chips apparently weaker.
Note that even with lower voltage, the DDR4 are going to consume more power than the same DDR3 memory.
We all hope that the semiconductor problem gets addressed by smart logic.
Differently? The basic operating principles and signaling scheme for DRAM has remained nearly unchanged since capacitor-based DRAM chip was invented. With custom PCBs and firmware, it would probably be possible to use LPDDR3 interchangeably with DDR4.
The CAS, RAS, RD, Precharge, command, etc. latencies mean exactly the same thing on any memory from the ancient Fast Page Mode asynchronous memory to DDR4; the only difference is that it is expressed in terms of clock cycles on synchronous memory.
On a fundamental basis, all mainstream SDRAMs are awfully similar.
DDR4 getting everywhere is not an if; it is merely a question of when: DDR3 production will start ramping down after Skylake comes out, ramp down even faster once AMD announces their own DDR4 chips, add maybe two more years and DDR3 will be almost history.
The transition will take about four years... just like the DDR1-DDR2 and DDR2-DDR3 transitions before it.
We will have to just wait and see how it plays out in actual real world test comparisons and benchmark results. Obviously these DDR4 timings will change a lot in quick fashion anyway and the voltages are quite a bit lower. They'll likely lower the timings and bump up the voltages a bit over time and probably boost the clock speeds as well as it matures.
The increase of latency is history repeating and as a matter of fact, in each jump the first new modules were slower than the older and they consume more. DDR4 2133 consume 4 times more a DDR 133, and only the same capacity modules will consume less, but DDR4 introduces Vpp at 2,5V and its logic seems to improve memory usage so without a benchmark to tell us how much I think DDR4 consume more than DDR3 and is going to consume a lot more with higher density memory. The latency is always a concern because we never broke the 7.5 ns wall. Data moves a lot faster on higher frequencies ram and controllers have been designed to take advantage more of frequencies than latencies so while I consider both for quality in the end only benchmarks will tell us how the new logics uses DDR4.