DDR4 Memory Will Be Released By Next Month

[Heinz68]

It seems that DDR4 isn?t as far away as we thought . According to Crucial Memory?s promo page it?s going to come out late 2013. There is just one month left till the years end. So that being said, we are going to have DDR4 in our PC?s hopefully by next month.

Of course DDR4 has a different architecture, meaning we are going to need a different motherboard, we can?t just put them in our old DDR3 systems. But, is it worth upgrading to DDR4? Crucial Memory also provided a comparison chart with some specifications of what DDR4 will offer, just to show you the difference.

The DDR4 will only eat up 1.2Volts as stated by Crucial , while having twice the speed of DDR3 Memory. DDR4 will run on a base memory speed of 2133MHz while having 4GB as their minimum density. According to the chart, we can see that DDR4 is 100% faster than DDR3, requires 20% less voltage and has 300% more density than that of DDR3.

If true i might postpone my system upgrade.

source

[ZX2Slow]

What can use it that will be available at launch?

[Heinz68]

What can use it that will be available at launch?

Good question, also I just search Crucial Memory site and can't find any news about it.

here is another article read more at PCWord

DDR4 will first go into servers early next year and then into clients like laptops and desktops in 2015, said Mike Howard, principal analyst at IHS iSuppli. DDR4 will succeed DDR3 SDRAM, which ships with most computers today.

Buyers will pay a 30 percent or higher premium for DDR4 memory compared to DDR3 next year, though the price differential is expected to decline to around 10 percent in 2015,

[Darakian]

Dev boards for haswell-EP. Dell, HP and the like are probably testing designs already.

[Hornet331]

No point for consumers to get it anywhere before the next 1.5-2 years.

The only thing where it might be interesting is for socs and a "potent" graphic card, so manly notebooks.

[hecktic]

for business related applications and Analytics type systems and software applications, the more RAM you have the better and the faster it is the more efficient your work flow is going to be. But I imagine the price tag is going to be through the roof for DDR4... but then also remember DDR3 prices will drop as a result so this is good for everyone!

[hecktic]

and if DDR3 frequencies can hit 3 to 4 ghz then I can only imagine how fast DDR4 will hit. This is very exciting news for memory manufacturers

[22JHP]

jedec announced new ddr4 standart in 2015 and first cpu's for ddr4 will be skylake and carizio.

[ZX2Slow]

No point for consumers to get it anywhere before the next 1.5-2 years.

Whew, Just picked up 32GB (4x8GB) of ddr3 2400 ram for a IB-E build.

[madcho]

up to 20% power save ...

OK THE MARKETING GUYS DONT KNOW ANYTHING ABOUT POWER !!!

P = Alpha * C * f * V^2

( without taking account of the f is non linear too )


DDR2, didn't decreased power, and DDR3 not really too. They add some more mem chips on board. Huge OC, to keep the TDP arround the same.



SO PLEASE FIRE THAT MAN !!!

[josh1980]

up to 20% power save ...

OK THE MARKETING GUYS DONT KNOW ANYTHING ABOUT POWER !!!

P = Alpha * C * f * V^2

( without taking account of the f is non linear too )


DDR2, didn't decreased power, and DDR3 not really too. They add some more mem chips on board. Huge OC, to keep the TDP arround the same.


SO PLEASE FIRE THAT MAN !!!

What's the variables exactly?

Been trying to find that equation and explanation for a while(engineer by hobby)...

[Nightfire]

What's the variables exactly?

Been trying to find that equation and explanation for a while(engineer by hobby)...

Alpha = Activity factor. Just a constant between 0 and 1. I guess I should add that this factor will be 1 if said component switches for each clock, but lower if it doesn't. If you have a flip-flop which do not trigger every clock this constant will be lower than 1, just to scale the power consumption accordingly.
C = Capacitance
f = The frequency
V = voltage.

This is why you see this huge step in power consumption and heat when overclocking. The power consumption will be (roughly) linear to the frequency. E.g. a doubling in frequency will equal a doubling in power consumption. On the other hand if you double your voltage that will equate to a quadrupling in power consumption. This is not a perfect model, but it will give you some pointers about what is going on.

[Flanman]

Alpha = Activity factor. Just a constant between 0 and 1
C = Capacitance
f = The frequency
V = voltage.

This is why you see this huge step in power consumption and heat when overclocking. The power consumption will be (roughly) linear to the frequency. E.g. a doubling in frequency will equal a doubling in power consumption. On the other hand if you double your voltage that will equate to a quadrupling in power consumption. This is not a perfect model, but it will give you some pointers about what is going on.

thank you much for the explanation.

[AliG]

Can someone explain to me the difference between DDR and GDDR? If I recall correctly, GDDR4 was basically useless and appropriately effectively bypassed for GDDR5. I'm just wondering what the differences are, and if DDR4 will actually be worth the price premium

[Nightfire]

There are many similarities between DDR and GDDR type memory. DDR has a much lower latency than GDDR, but GDDR have the advantage when it comes to bandwidth. I guess you can say that DDR is better at smaller data and GDDR is better at larger data.

DDR4 will have a smaller power consumption than DDR3 because of lower voltages, which is nice for laptops at least. I've heard somewhere than the current will remain the same, but I can't remember where. The relation between current, voltage and power are: P = V*I. So a lower power consumption of about 20% seems fair. DDR4 will also be significantly faster in terms of frequency than DDR3, I don't know about latency, but this will propably go up. DDR4 now also support 3.2 GT/s (giga transfers per second), which is twice that of DDR3 (1.6 GT/s). So if it is worth the premium we don't know, we'll just have to wait and see.

[AliG]

There are many similarities between DDR and GDDR type memory. DDR has a much lower latency than GDDR, but GDDR have the advantage when it comes to bandwidth. I guess you can say that DDR is better at smaller data and GDDR is better at larger data.

DDR4 will have a smaller power consumption than DDR3 because of lower voltages, which is nice for laptops at least. I've heard somewhere than the current will remain the same, but I can't remember where. The relation between current, voltage and power are: P = V*I. So a lower power consumption of about 20% seems fair. DDR4 will also be significantly faster in terms of frequency than DDR3, I don't know about latency, but this will propably go up. DDR4 now also support 3.2 GT/s (giga transfers per second), which is twice that of DDR3 (1.6 GT/s). So if it is worth the premium we don't know, we'll just have to wait and see.

Thanks for the explanation, I appreciate it.

I don't like that the current will remain the same. Recall that power is also I^2 * R; in the end current is everything when you're talking power efficiency. I would imagine that the latency would go up, just because it has gone up each new ddr. As an example, right now cas 9 is considered pretty solid, but with ddr you could get down to cas 1.

[Nightfire]

Thanks for the explanation, I appreciate it.

I don't like that the current will remain the same. Recall that power is also I^2 * R; in the end current is everything when you're talking power efficiency. I would imagine that the latency would go up, just because it has gone up each new ddr. As an example, right now cas 9 is considered pretty solid, but with ddr you could get down to cas 1.

When working on this level things are a bit more complicated than just ohms law. This is because of capacitances and inductance at higher frequencies. There are parasitic capacitances at transistor level between gate, drain, bulk and source. Power efficiency is not really only depicted by the current, but as a product of the current and voltage. You mentioned that P = I^2 * R, but P = V^2/R also. So if the current stays the same and the voltage drops, the power consumption will drop. The input impedance of a circuit will not remain the same for all frequencys. This can easily be explained by looking at a capacitor. A capacitor will block all DC current, but will act more and more as a short when frequency increases. A capacitor will actually behave like an inductor when the frequency gets really high, because of the leads to the cap. This will change the input impedance of a circuit, so R will not be given as a constant, but will vary with frequency.

[AliG]

Lol I know all about basic circuitry, I have a degree in EE All I'm saying is that people normally design their transistors off current requirements; a high current BJT will always run hotter than a BJT with same potential but very high impedance (thus lower current)

[Nightfire]

You don't use BJT for DDR memory though. BJTs are better where high speed and a high breakdown voltage is needed. For example in high voltage radio-frequency circuits. You are using CMOS technology.
That people design their transistors off current requirements are indeed correct, as the speed is dictated by the capacitance and the current.

I can't remember, but doesn't the Rds in CMOS decrease when the current Ids increases? At least in the active region.

For the interested: The reason that CMOS technology is not as good when it comes to high-frequency breakdown is that the gate dielectric is very thin on CMOS in order to get to high frequencies. Therefore the breakdown voltage of modern CMOS are around 1V.

I think we are slowly drifting off topic here

[BrowncoatGR]

Can someone explain to me the difference between DDR and GDDR? If I recall correctly, GDDR4 was basically useless and appropriately effectively bypassed for GDDR5. I'm just wondering what the differences are, and if DDR4 will actually be worth the price premium

GDDR4 and GDDR5 are both based on DDR3, so there is no relation between GDDR4 and DDR4

[madcho]

You don't use BJT for DDR memory though. BJTs are better where high speed and a high breakdown voltage is needed. For example in high voltage radio-frequency circuits. You are using CMOS technology.
That people design their transistors off current requirements are indeed correct, as the speed is dictated by the capacitance and the current.

I can't remember, but doesn't the Rds in CMOS decrease when the current Ids increases? At least in the active region.

For the interested: The reason that CMOS technology is not as good when it comes to high-frequency breakdown is that the gate dielectric is very thin on CMOS in order to get to high frequencies. Therefore the breakdown voltage of modern CMOS are around 1V.

I think we are slowly drifting off topic here

True

BTJ are more powerfull than CMOS, but MORE power hungry, and have input current, and not CMOS. BJT are less noisy in flinker noise and the most effective effect in BJT is diffusion current, CMOS use mainly the drift current. ( the diffusion is main in weak inversion ).

So in very high speed area some designer in digital can choose to use bjt, but in trade off power, to increase speed. More and more people want to work in weak inversion for more performance / watt. it's slower, and cause a lot of chance of defect chips. So It's not used until now in digital. In analog you can do it but your trade off will be size. HUGE size.

So people intend to work in moderate inversion. Maybe the best area to work in digital for the next years.

Not interesting in high performance part of a ship, but there is more and more black silicon, and accerlerators. So Maybe the best mode to work for this CMOS Ts.

[Olivon]

jedec announced new ddr4 standart in 2015 and first cpu's for ddr4 will be skylake and carizio.

Haswell-E will bring DDR4 sooner than Skylake.

[Hornet331]

To most interesting part probably will be for IGPs, with a rated speed of 3200 MT that would be 52Gb/s which is in the range of a AMD 260 or Geforce 650, but then probably to expensive again to be of any use in the near feature before DDR5 is around the corner.

[zir_blazer]

As far that I remember, the very first platform to use DDR-4 was going to be Haswell-E, which is still quite far away.

[Tenknics]

As far that I remember, the very first platform to use DDR-4 was going to be Haswell-E, which is still quite far away.

sadly...


But it looks like the time between Ivy-E and Haswell-E will be a year or so, rather than 2 years between Sandy-E and Ivy-E...

I hope Haswell-E hits beginning of summer rahter than late/fall