Yeah, finding them "very easily" is a bit of an exaggeration. They exist, but only in the 4GB variant, and will cost you about 18€/GB in Europe, while you can get CL10 for about 8€/GB.
That aside, the initial timings on DD4 seem to be in line with the upper end of DDR latencies. The photo above shows all CLs to be between 6.25 ns and 7.5 ns. The oldest and cheapest offers for DDR3 are still 1333 / CL 9 (6.75 ns) and 1600 / CL 11 (6.875 ns).
It is mainly through careful "enthusiast" selection that we now can buy latencies below 4ns, and it comes at a price premium.
I mean, it should bin better over time. [anecdote] In December 2013 (6 years after DDR3 released) I bought for $90 some Corsair Vengeance 8GB DDR3-1866 @ 9-10-9-27 w/ 1.5V (4.823ns). Overclocked the sticks to DDR3-2200 @ 9-12-11-32 w/ 1.7V (4.091ns). [/anecdote]
I just borrowed your latency calculation, hehe. How accurate is to compare between generations? Does DDR4 have some underlying tweaking that makes this calculation too inaccurate?
The calculation of latency times is perfectly accurate. The technological tweaking that happened is the reason why we now get the same latencies at 1.2V which we used to get from 1.5V.
As you noticed for your own set, increasing the voltage helps. Using 1.7V instead of 1.5V you have increased the final charge of all capacitors involved, so if the capacitor will nominally switch from "0" to "1" at 50% charge, the over-volted capacitor now only has to wait for 50%*1.5V/1.7V = 44% of full charge.
So, I'm sure we will pretty quickly find some out-of-spec 1.35V sticks with better latencies. Maybe even 1.5V sticks, if Haswell-E is capable of handling this much increase in voltage.
reducing the voltage is mostly for laptops (and phones) in order to reduce power use and heat generation. However, on the desktop side of things it will be important down the line because smaller and smaller manufacturing processes are more and more allergic to high voltages, so when we start seeing single sticks of ram with 16+GB (Hynix has shown off a 128GB module) on them it is going to be possible without melting because of the lower voltage standard.
Overall I am glad to see DDR4 not increasing latency much compared to DDR3. when DDR3 came out the latency was a step backwards compared to the mature DDR2 on the market, and it took a while to catch up. DDR4 is coming out the gate with a minor latency penalty while offering massive throughput improvements, and I am sure it will get better as it matures. http://www.micron.com/products/dram/ddr3-to-ddr4 http://en.wikipedia.org/wiki/DDR4_SDRAM
power consumption, heat, and power usage. Also the longevity of the semiconductor, since at higher voltage and higher temp, you're really reducing the lifetime of the product (http://en.wikipedia.org/wiki/High-temperature_oper...
I am patiently awaiting that day! Things like DDR4 and new HDD connectivity standards are just about the only reason to upgrade a core system anymore. CPUs are continuing to improve... but the practicality of their improvements outside of competitive overclocking and extreme production environments is questionable at best. SSDs continue to get faster, but without new connection standards there are diminishing returns. PCIe v4 will be overkill by the time it is released, but it will allow PCIex8 graphics slots to be used without bottleneck to allow PCIe lanes to be freed up for other uses (thunderbolt, HDDs, etc.). It is weird to not have any real want to upgrade my PC after 3 years. Hopefully Skylake will offer something worth upgrading to.
im still rocking a i7 1366, 12gb ddr3 ....be it with a real hardware raid controller and 4ssds (ocz vertex 3) + 3hdds and semi recent graphics card (amd 7980)...
the mobo and cpu are 5 years old now and i still dont see much of a reason to update. ill probably end up going through one more gfx card in before i do (i update about every 2 generations).
sure its not the fastest rig around anymore but 5 years later it still outdoes some mid level offerings and feels fast day to day.
even when I do retire it, it will probably make a nice hypervisor system for a few linux servers.
I would love to be able to buy something worth buying... I could use 4 more real cores, I need more SATA2 connections, I need cheaper PCIE SSDs too tho, but seems nothing worth buying is coming soon. If Haswell-E will offer 8/16 @4GHz with turbo @4,4 (like refresh), native SATA2 of at least 6 and DDR4 faster than 2,4GHz... it may be worth buying.
If you compensate for clockspeed increases, what you'll find is that latency on an absolute (temporal) basis isn't much changed from before. It still takes so many nanoseconds to fetch data. We're now just able to fetch more at once.
Also, if you keep in mind the voltage required, RAM has become more efficient over time, trading latency for reduced voltage. Obviously if you pump up the voltage on a IC, latency can be reduced. ShieTar explains it better and with formulas, read his post.
CL pretty much stands for latency per cycle, not the absolute latency. Hence why higher clocked modules generally come with a higher CL, yet their real-world latency can actually remain stagnant or be improved.
Still, I am coming away un-impressed with DDR4 at this early stage, 2133mhz isn't exactly what I would call "Ground breaking" when you could pick up a Samsung Green kit and overclock those babies to 2500mhz - 3ghz with relative ease. I'm on socket 2011 and have had my DDR3 sitting at 3ghz for the last couple of years without any drama and only at 1.55v!
Can you ask someone at the show what the story is regarding DDR4 and point-to-point topology. We were told that DDR4 would allow only one DIMM per channel, and the Haswell-E slides reinforced this. But we have seen multiple Haswell-E (whose memory controller has 4 channels) motherboards sporting 8 DIMM slots. What is going on here?
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
25 Comments
Back to Article
UltraWide - Friday, June 13, 2014 - link
Where can we find DDR3 running at 2133MHz with CAS 8???morganf - Friday, June 13, 2014 - link
Corsair sells some. Look at Vengeance Pro or Dominator Platinum.ShieTar - Friday, June 13, 2014 - link
Yeah, finding them "very easily" is a bit of an exaggeration. They exist, but only in the 4GB variant, and will cost you about 18€/GB in Europe, while you can get CL10 for about 8€/GB.That aside, the initial timings on DD4 seem to be in line with the upper end of DDR latencies. The photo above shows all CLs to be between 6.25 ns and 7.5 ns. The oldest and cheapest offers for DDR3 are still 1333 / CL 9 (6.75 ns) and 1600 / CL 11 (6.875 ns).
It is mainly through careful "enthusiast" selection that we now can buy latencies below 4ns, and it comes at a price premium.
ikjadoon - Friday, June 13, 2014 - link
Agreed. Existing, but pricey.I mean, it should bin better over time. [anecdote] In December 2013 (6 years after DDR3 released) I bought for $90 some Corsair Vengeance 8GB DDR3-1866 @ 9-10-9-27 w/ 1.5V (4.823ns). Overclocked the sticks to DDR3-2200 @ 9-12-11-32 w/ 1.7V (4.091ns). [/anecdote]
I just borrowed your latency calculation, hehe. How accurate is to compare between generations? Does DDR4 have some underlying tweaking that makes this calculation too inaccurate?
ShieTar - Friday, June 13, 2014 - link
The calculation of latency times is perfectly accurate. The technological tweaking that happened is the reason why we now get the same latencies at 1.2V which we used to get from 1.5V.As you noticed for your own set, increasing the voltage helps. Using 1.7V instead of 1.5V you have increased the final charge of all capacitors involved, so if the capacitor will nominally switch from "0" to "1" at 50% charge, the over-volted capacitor now only has to wait for 50%*1.5V/1.7V = 44% of full charge.
So, I'm sure we will pretty quickly find some out-of-spec 1.35V sticks with better latencies. Maybe even 1.5V sticks, if Haswell-E is capable of handling this much increase in voltage.
Revolution11 - Friday, June 13, 2014 - link
But what is the advantage of reducing voltage? Power consumption? Or is there another benefit like component longevity?extide - Friday, June 13, 2014 - link
Power consumption is really it.CaedenV - Friday, June 13, 2014 - link
reducing the voltage is mostly for laptops (and phones) in order to reduce power use and heat generation. However, on the desktop side of things it will be important down the line because smaller and smaller manufacturing processes are more and more allergic to high voltages, so when we start seeing single sticks of ram with 16+GB (Hynix has shown off a 128GB module) on them it is going to be possible without melting because of the lower voltage standard.Overall I am glad to see DDR4 not increasing latency much compared to DDR3. when DDR3 came out the latency was a step backwards compared to the mature DDR2 on the market, and it took a while to catch up. DDR4 is coming out the gate with a minor latency penalty while offering massive throughput improvements, and I am sure it will get better as it matures.
http://www.micron.com/products/dram/ddr3-to-ddr4
http://en.wikipedia.org/wiki/DDR4_SDRAM
menting - Friday, June 13, 2014 - link
power consumption, heat, and power usage. Also the longevity of the semiconductor, since at higher voltage and higher temp, you're really reducing the lifetime of the product (http://en.wikipedia.org/wiki/High-temperature_oper...ikjadoon - Friday, June 13, 2014 - link
"Voltage moves down from 1.5 V to 1.2 V, and processors are expected to support DDR3-2133 by default."Should that be DDR4-2133?
Ian Cutress - Friday, June 13, 2014 - link
Correct. Five years of writing DDR3-2133 and it gets stuck. Updated :)ikjadoon - Sunday, June 22, 2014 - link
Haha, I understand...actually, when I wrote my comment, I wrote accidentally wrote "DDR3-2133", too, lol.toncij - Friday, June 13, 2014 - link
So this means we will get DDR4 desktop platforms by end of Q3?extide - Friday, June 13, 2014 - link
Yeah, if you are going to buy a Haswell-E system. For the "normal" desktop stuff, it won't be until 2016 with Skylake.CaedenV - Friday, June 13, 2014 - link
I am patiently awaiting that day! Things like DDR4 and new HDD connectivity standards are just about the only reason to upgrade a core system anymore. CPUs are continuing to improve... but the practicality of their improvements outside of competitive overclocking and extreme production environments is questionable at best. SSDs continue to get faster, but without new connection standards there are diminishing returns. PCIe v4 will be overkill by the time it is released, but it will allow PCIex8 graphics slots to be used without bottleneck to allow PCIe lanes to be freed up for other uses (thunderbolt, HDDs, etc.).It is weird to not have any real want to upgrade my PC after 3 years. Hopefully Skylake will offer something worth upgrading to.
jdub_06 - Monday, November 3, 2014 - link
im still rocking a i7 1366, 12gb ddr3 ....be it with a real hardware raid controller and 4ssds (ocz vertex 3) + 3hdds and semi recent graphics card (amd 7980)...the mobo and cpu are 5 years old now and i still dont see much of a reason to update. ill probably end up going through one more gfx card in before i do (i update about every 2 generations).
sure its not the fastest rig around anymore but 5 years later it still outdoes some mid level offerings and feels fast day to day.
even when I do retire it, it will probably make a nice hypervisor system for a few linux servers.
toncij - Friday, June 13, 2014 - link
I would love to be able to buy something worth buying... I could use 4 more real cores, I need more SATA2 connections, I need cheaper PCIE SSDs too tho, but seems nothing worth buying is coming soon. If Haswell-E will offer 8/16 @4GHz with turbo @4,4 (like refresh), native SATA2 of at least 6 and DDR4 faster than 2,4GHz... it may be worth buying.celestialgrave - Friday, June 13, 2014 - link
Look at the history of ddr memory, every advance lowers the voltage and raises the CL. I still have a Barton system running Hyper X DDR2 at CL 4.ViRGE - Friday, June 13, 2014 - link
If you compensate for clockspeed increases, what you'll find is that latency on an absolute (temporal) basis isn't much changed from before. It still takes so many nanoseconds to fetch data. We're now just able to fetch more at once.Revolution11 - Friday, June 13, 2014 - link
Also, if you keep in mind the voltage required, RAM has become more efficient over time, trading latency for reduced voltage. Obviously if you pump up the voltage on a IC, latency can be reduced. ShieTar explains it better and with formulas, read his post.StevoLincolnite - Friday, June 13, 2014 - link
CL pretty much stands for latency per cycle, not the absolute latency.Hence why higher clocked modules generally come with a higher CL, yet their real-world latency can actually remain stagnant or be improved.
Still, I am coming away un-impressed with DDR4 at this early stage, 2133mhz isn't exactly what I would call "Ground breaking" when you could pick up a Samsung Green kit and overclock those babies to 2500mhz - 3ghz with relative ease.
I'm on socket 2011 and have had my DDR3 sitting at 3ghz for the last couple of years without any drama and only at 1.55v!
name99 - Friday, June 13, 2014 - link
Can you ask someone at the show what the story is regarding DDR4 and point-to-point topology.We were told that DDR4 would allow only one DIMM per channel, and the Haswell-E slides reinforced this. But we have seen multiple Haswell-E (whose memory controller has 4 channels) motherboards sporting 8 DIMM slots. What is going on here?
toncij - Friday, June 13, 2014 - link
This is interesting. There would be no sense in sharing bandwidth per channel?Marucins - Monday, June 16, 2014 - link
MHz and CL compare - DDRBF4
http://s24.postimg.org/ajwe4y1g5/DDR_FREQ_CL_bf4.j...
WinRAR
http://s24.postimg.org/e4s9o65zp/DDR_FREQ_CL_winra...
Antronman - Thursday, June 19, 2014 - link
If you're the kind of person who needs 8 cores, why are you waiting for HW-E?Xeons with IB-E have support for up to 12 cores.