backtop

Print E-mail del.icio.us 37 comment(s) - last by ADDAvenger.. on Jul 8 at 8:44 PM

Super Talent DDR3 memory kit
New 1600 MHz 7-7-7-18 DDR3 modules set record for lowest timings

Super Talent Technology this week released three new DDR3 memory kits – two dual-channel kits and one single module. The dual-channel DDR3 kits arrive clocked at 1600 MHZ with differing latencies. The Super Talent W1600UX2G7 dual-channel features 7-7-7-18 timings while requiring 1.8-volts of power. The kit features two 1GB modules and costs around $648.

Super Talent has a slightly cheaper 2x1GB kit with the W1600UX2G9. The W1600UX2G9 features slightly higher 9-9-9-21 timings and requires 1.8-volts of power. Super Talent prices the W1600UX2G9 modules slightly cheaper at around $598.

Lastly is the W1333UB2G8 single module. The W1333UB2G8 has a slightly lower speed rating at 1333 MHz. However, the W1333UB2G8 is a single 2GB module. Memory timings for the W1333UB2G8 slot in between the two dual-channel kits with 8-8-8-18 timings. The W1333UB2G8 modules still require 1.8-volts of power. Super Talent prices the 2GB W1333UB2G8 modules around $417.

All three-memory kits announced feature black heat spreaders to help dissipate heat. Expect the kits to show up at retailers in the coming weeks.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
7-7-7-7-18 Doesn't look that low
By InternetGeek on 7/5/2007 9:28:30 PM , Rating: 2
Those numbers don't look that low, though I must recognize I'm outdated in terms of ram. Anyone care to share some knowledge?




RE: 7-7-7-7-18 Doesn't look that low
By retrospooty on 7/5/2007 9:31:20 PM , Rating: 5
"Anyone care to share some knowledge"

Its not your imagination. Each new gen of ram increases speed and latency. The end result is that todays latest greatest RAM is only a little faster then good old PC3200@2-2-2 latency.


RE: 7-7-7-7-18 Doesn't look that low
By kenji4life on 7/5/2007 9:34:18 PM , Rating: 2
The main increases are generally in efficiency and bandwidth.


RE: 7-7-7-7-18 Doesn't look that low
By retrospooty on 7/5/2007 9:51:17 PM , Rating: 3
right... but the increased latency dramatically cripples those gains, and even though bandwidth tests are improving, real world performance is not a whole lot better.


RE: 7-7-7-7-18 Doesn't look that low
By InternetGeek on 7/5/2007 11:12:37 PM , Rating: 4
I can understand efficiency is improved in terms of energy usage, heat dissipation, etc.

But it's not that easy to see how bandwidth is improved if latency keeps growing. It's like being able to run faster for shorter terms of time. Yes, you reach your destination but you might take longer when you add all your resting times (latency).

It seems to me offferings are not that great now under these terms because when memory finally reaches a latency worth its price (bear with me on this one: 2-2-2-6-1) they just release a new standard.


RE: 7-7-7-7-18 Doesn't look that low
By HaZaRd2K6 on 7/5/2007 11:43:04 PM , Rating: 5
The latencies represent the number of clock cycles it takes for the memory to achieve a certain task (these numbers representing tCAS (number of cycles to access a "column" of data), tRCD (RAS to CAS delay, time between defining row and column and actually writing to them), tRP (clock cycles to terminate access to a row), and tRAS (clock cycles to access a row of data)).

With that in mind, 400MHz (PC3200) DDR RAM running at 2.5-3-3-5 would take 0.00000000625 seconds, whereas DDR3 running at 1333MHz with timings of 7-7-7-18 would take 5.25e^-9 seconds (much less time).

Basically, the latencies are directly related to the clock cycles and not to the time that has elapsed. Therefore, DDR3-1333 (PC3-10666) is faster than DDR400 (PC3200) in every way. Hope that helps. If not, check out the Wikipedia page on SDRAM Latency and this AT page: http://www.anandtech.com/memory/showdoc.aspx?i=299...


RE: 7-7-7-7-18 Doesn't look that low
By Comdrpopnfresh on 7/6/2007 12:13:48 AM , Rating: 2
technically speaking yes. but I think everyone above was referring to "timings" as latency. In which case timing, or latency by their terms, does make a difference


RE: 7-7-7-7-18 Doesn't look that low
By LogicallyGenius on 7/6/07, Rating: -1
RE: 7-7-7-7-18 Doesn't look that low
By LogicallyGenius on 7/6/07, Rating: -1
RE: 7-7-7-7-18 Doesn't look that low
By LogicallyGenius on 7/7/07, Rating: -1
By LogicallyGenius on 7/8/2007 12:43:09 AM , Rating: 4
all riet u win i luse


RE: 7-7-7-7-18 Doesn't look that low
By kamel5547 on 7/6/2007 12:21:17 AM , Rating: 3
Actually the anandtech page doesn't show much improvement at all for actual applications. Your are looking at a difference of LESS than 1% between the top DDR2 memory tested at 1066 and the Kingston DDR3 at 1500. Basically a huge increase in speed nets you the same performance due to the latency ramp up.

Not surprising since real performance of DDR2 was SLOWER than DDR for quite some time due to the latency issue.


RE: 7-7-7-7-18 Doesn't look that low
By retrospooty on 7/6/2007 10:39:02 AM , Rating: 2
"Your are looking at a difference of LESS than 1% between the top DDR2 memory tested at 1066 and the Kingston DDR3 at 1500."

Exactly. Its faster, just by a very small margin, as was DDR2 over DDR.


RE: 7-7-7-7-18 Doesn't look that low
By omnicronx on 7/6/2007 10:55:29 AM , Rating: 3
quote:
"Your are looking at a difference of LESS than 1% between the top DDR2 memory tested at 1066 and the Kingston DDR3 at 1500."


this is quite a generic statement, for both of you.
its a well known fact AMD processors received small or no improvement from ddr2 over ddr1. The conroe c2d on the otherhand makes very good use of the DDR2 bandwidth, and with the new p35 chipset adds on to this, increasing that margin by 4-8% over 9** boards in real world performance.

Also it must be noted that for a fair comparison, you would have to compare launch date minimum speed ddr2 memory vs launch ddr3 memory at its lowest spec. ddr2 at launch was crap compared to ddr especially with pc4200 and 5300 modules.

give ddr3 some time, with higher frequencies, lower timings and 1.5voltage, things should get much better for ddr3 (overclocking included ;) You didnt rush out to go buy ddr2 memory.. why rush out to buy ddr3 ;)


By retrospooty on 7/6/2007 2:39:41 PM , Rating: 3
"with the new p35 chipset adds on to this, increasing that margin by 4-8% over 9** boards in real world performance."

True, but most of that increase is the P35 chipset, not the RAM. I do realize that DDR3 ram overall is faster, my point is that in real world performance, each new gen is only slightly faster (and that is AFTER a few years when better latency sticks come out). In other words, evolution, not revolution, and extremely small baby steps at that!

from an article here at AT.... "P35 supports either DDR2 or DDR3, and we found no real difference in current performance of DDR2 and DDR3 on the P35 platform. Both were equally faster than the same DDR2 on the P965 chipset."

http://www.anandtech.com/memory/showdoc.aspx?i=298...


RE: 7-7-7-7-18 Doesn't look that low
By henrikfm on 7/6/2007 6:48:28 AM , Rating: 2
So, DDR400@2-3-2-5 is faster than DDR1333@7-7-7-18?

Because 0.005*10^-6 < 0.00525*10^-6


RE: 7-7-7-7-18 Doesn't look that low
By oTAL (blog) on 7/6/2007 9:00:24 AM , Rating: 5
DDR400@2-3-2-5 would have faster latencies but lower overall bandwidth than DDR1333@7-7-7-18.

There's a trade off, yes. But is not nearly as big as some people believe it to be.
When you double the frequency and keep the same latency the "latency numbers" will double. This leads people to believe that the new memory has double the latency, although it has the same.
When DDR2 showed up it had a little higher latency than the previous generation, which would overshadow any gains from the bandwidth. The same will be truth for DDR3 since latency is, in most applications, more important than bandwidth. There's an equilibrium here - both are important and it is easier for improved memory controllers to compensate for high latency, than for low bandwidth (prefetching, etc). Still, when the technology evolves, the new memory ends up better than the previous one, even on latency.

The problem is that, while bandwidth has been doubling every new generation, the latency only gets very slight improvements. That is a physical limitation. If you make the transistors smaller you get shorter timings, but if you lower the voltage applied you get longer timings. Since every new memory iteration drops the voltage we end up getting a higher latency, which is then compensated by the move to smaller processes.
On the other, the lower voltage allows the memory to consume less power and dissipate less heat, which allows the manufacturer to crank up the bandwidth, consuming more power and dissipating more heat.

Engineering is all about trade offs and usually you need to find the best equilibrium as you move forward. Although not in a smooth way we are getting better memory. But there are ups and downs in this process. Kind of like when a stock is climbing in the stock market. There are ups and downs, good days and bad days, but in the end of the year your stock is worth more than it was before. Next year's memory will be better than this year's memory all around: less heat, less power, more bandwidth and lower latencies.


RE: 7-7-7-7-18 Doesn't look that low
By oDii on 7/7/2007 8:19:25 AM , Rating: 2
Great post. If the Dailytech mod system liked me and had actually given me points to vote with in the past year, I'd mod you up (it hasn't for ages so I just complain sarcastically...).


RE: 7-7-7-7-18 Doesn't look that low
By geddarkstorm on 7/6/2007 12:46:43 PM , Rating: 2
0.00000000625 = 6.25e-9. So the difference between PC3200 at 2.5-3-3-5 and DDR3 1333MHz with 7-7-7-18 is only 1e-9 seconds (which is NOT much less time). And when you look at the price difference... Wow, the performance to price ratio of DDR3 is horrendous compared to PC3200.

Thanks for this knowledge and very good basic data. Now I know I need to be weary of new ram... And it seems the original poster is nearly on the ball--this new ram isn't worth it for the price at all it seems from this.


By Surak on 7/8/2007 3:52:49 PM , Rating: 2
Actually ...

It IS much less time ... 16% less time.

Anywhere else in the PC industry, an upgrade that let you do something in 16% less time would be not to shabby at all.

Don't like the price or don't think the performace is good enough yet .... then don't be an early adopter.

... but you better be grateful to the early adopters, who are the ones that really pay for the advances you will be enjoying later.


RE: 7-7-7-7-18 Doesn't look that low
By oralpain on 7/5/2007 11:31:41 PM , Rating: 1
7-7-7-18 is quite impressive for DDR3 at that speed.

Remember that latency is a combination of clock rate, and timings. DDR3 1600 with 7-7-7-18 timings is far lower latency than 2-2-2-x DDR1.


By geddarkstorm on 7/6/2007 1:11:20 PM , Rating: 2
I'm going to try to calculate this another one, correct me if I'm wrong. One Hertz is one clock cycle. If it takes 7 cycles to perform one operation, then the effective hertz, or clock rate, is 1/7th the actual.

Therefore, DDR1 at a clock speed 400MHz and 2-2-2 timing for most things would have an effective clock of 200MHz. 1600MHz DDR3 with 7-7-7 timing for most things would have an effective clock of 229Mhz. That's only a 29MHz difference for over 500 bucks more. Not appealing to me. I guess the main advantage of DDR3 is the increased bandwidth which would throw the effective data transfer rate above DDR1. But I don't know the actual numbers for either, so I can't calculate theoretical data movement speeds.


By iVTec on 7/6/2007 4:49:28 AM , Rating: 2
And also don't forget that latencies of DDR3 RAM can't go lower than 5-5-5...So 7-7-7 is pretty much like DDR2's 4-4-5 today...Anyway,perhaps these modules are equipped with the new DDR3 IC from Micron?


By ADDAvenger on 7/8/2007 8:44:25 PM , Rating: 2
The response time for that memory is equivalent to DDR2 800mhz at 3.5. The response time (latency) is measured in clock cycles, not ns; you have to remember that the DDR3 is clocking two times for every one clock of DDR2, so even if latency is 1.5 times that of DDR2, the actual response time in seconds is 3/4 of the same DDR2.


Well...
By DeepBlue1975 on 7/5/2007 9:49:35 PM , Rating: 2
I'd rather have mem makers announce CHEAP ddr3 instead of "low latency".

I've ended up learning the hard way that in normal use, low latency is not as good as benchmarks could suggest.
And even more, newest CPUs aren't as crippled by lower ram latencies than they were some years ago, they don't even seem to care that much when you overclock your CPU in sync with RAM speed or you just use a divider...
Of course synthetic benchmarks can tell the difference, but "real world usage" and "subjective perception" most of the time do not.




RE: Well...
By kenji4life on 7/5/2007 10:31:17 PM , Rating: 2
At least I still have the Nissan...


RE: Well...
By luhar49 on 7/6/2007 12:16:14 AM , Rating: 3
DDR2 prices have fallen so low recently that price/performance ratio of these new DDR3 sticks even with better latency is dismal. If money is no object and you want the absolute best, sure pick these DDR3 RAMS up. If you are on a budget though, you would be much better off spending the extra dollars on components that will show you much more pronounced real world performance improvement. Stuff like Graphics cards/ CPUs / Cooling(if you are OCing) / HDD.

DDR3 with these prices is still a couple of years away from going mainstream.


RE: Well...
By HaZaRd2K6 on 7/6/2007 11:50:33 AM , Rating: 2
quote:
If money is no object


Ain't that the truth...

We just got some new Corsair XMS3 and XMS3 DHX sticks in at work. Both run at DDR3-1333 and have 9-9-9-27 timings. The XMS3s are running for $529.99 and the DHXs will burn a hole in your wallet about $620 wide (plus taxes). I'm speaking in Canadian currency, but it's still VERY expensive. You can get a Mushkin 2GB DDR2-1000 kit with timings of 4-5-4-11 for $384.

Personally, I'm not an early adopter (I don't have the money for it) so I think these new sticks are (at least right now) a waste of money. Sure, in a few months, the landscape will be different (the same as it was with the switch from DDR to DDR2), but right now, the entry-level price points are far too high for (I suspect) most people--even those reading this.


RE: Well...
By tehfire on 7/6/2007 12:01:44 AM , Rating: 2
I too don't buy into the low-latency argument, but on my last build I used some Corsair XMS memory, and memory bandwidth was much higher than it was with generic memory of the same speed. I doubt it was because of the timings; It probably was because of the higher-quality memory that I was using.

At the end of the day, low latency memory is far superior to generic memory, but not necessarily because of the latency - more so because of the higher-quality parts.


RE: Well...
By oralpain on 7/6/2007 9:49:11 AM , Rating: 3
The same kind of memory, in the same system, running at precisely the same timings and sub-timings will perform nearly identically, regardless of quality or brand.

This is quite easy to test, just use a program that can get at all of your memory timings and change them to the same settings as your inferior ram. I would be amazed if performance varied by more than 1%.

The fastest DDR I had was a 70 dollar 1GB kit of twinmos with revised winbond utt CH-5 ICs. It beat the shit out of my corsair, mushkin, and crucial for one reason. It could clock to the same or higher speeds, with vastly lower latencies, if I put enough voltage through it.

It's always good to buy quality memory, but quality itself does not imply speed, it only implies reliability and consistencey.


RE: Well...
By SmokeRngs on 7/6/2007 10:46:43 AM , Rating: 2
quote:
And even more, newest CPUs aren't as crippled by lower ram latencies than they were some years ago


It depends somewhat on the CPU architecture being used and the memory controller actually.

For example, the P4 required a lot of memory bandwidth and high latencies did not have much of a penalty. Using low latency RAM usually did not make much of a difference compared to higher latency RAM. However, pairing a P4 with RAM which was low bandwidth, you would starve the processor which would affect real world performance.

The AthlonXP was different. It didn't need high bandwidth to shine but it did need RAM synchronized with the processor's FSB along with low latency. Anything else would show a performance hit.

The Athlon64 was also different. While you don't want to starve it for memory bandwidth, high bandwidth RAM was not required. Low latency RAM also wasn't required. This was all due to the memory controller being on the CPU die instead of in the northbridge. Many of the problems with RAM latencies/bandwidth were bypassed because of the efficiency of the design.

quote:
Of course synthetic benchmarks can tell the difference, but "real world usage" and "subjective perception" most of the time do not.


I will have to disagree with you a little on this. For the average person running average software, your statement is correct. They are unlikely to notice a difference in performance. However, there is software out there that will show a real world difference depending on the bandwidth or latencies of RAM.

The example I give is a bit outdated but I think it's still applicable. Quake3 was always a game which loved memory bandwidth over low latencies. Most games designed on this engine ran better the more memory bandwidth you would give it. It's the main reason it was one of the shining points for the P4 over the AthlonXP. It would also score much higher on the P4 due to the increased memory bandwidth.

I don't have an example in mind right now; but there is software out which would show the exact opposite. Where lower latencies will show a large performance improvement improvement over increased bandwidth.


Wonderful..
By kenji4life on 7/5/2007 9:26:16 PM , Rating: 2
Great news. So When I can afford to drop six and a half bills on 2 sticks of ram, I can also get my Maserati out of the impound.




RE: Wonderful..
By roastmules on 7/6/2007 1:16:06 AM , Rating: 2
Agreed.
I just picked up 4GB of DDR2-800/5-6-6-19 (according to CPU-Z; 4-4-4-12 according to mfg), for $170 delivered -- about $45/GB. I'll take the 4GB of this "slow" ram over 1GB or less of fast DDR3 (at about the same price).
I'm happy with what I got.


wow
By Comdrpopnfresh on 7/5/2007 10:46:26 PM , Rating: 2
isn't 1600mhz the top convention supported speed? No one has the boards and speed is maxed? I understand latency can still go down, and the speeds can go beyond convention... but still...




RE: wow
By Parhel on 7/6/2007 1:00:29 PM , Rating: 2
I also find this surprising. I know that Intel predicted much higher official speeds for DDR3. It took much longer for DDR to reach 400Mhz speeds and for DDR2 to reach 800Mhz speeds. I'm sure that over time they will release DDR3-1600 with increased densities, lower stock voltages, lower latencies, reduced costs, etc.. But it still seems like poor planning by the JEDEC to predict that DDR3 will top out at 1600Mhz, and poor planning by Intel to only support up to 1333Mhz on the most recent chipsets.


voltage
By omnicronx on 7/6/2007 9:34:57 AM , Rating: 2
am i the only one to notice that both the single and the dual channel ram kits are 1.8v? i thought the ddr3 spec was 1.5v and that although ddr2 and ddr3 chips are the same size, the pin locations are different. I just don't see why anyone would want this for a while when ddr2 chips have much lower latencies and comparable bandwidth and the same voltage rating. does ddr3 have more than one standard voltage?? if so doesn't that see weird to anyone.




RE: voltage
By Redofrac on 7/7/2007 11:10:29 PM , Rating: 2
ddr3 spec certainly is 1.5v. Rather, it's the case that some of the manufacturers don't feel a need to adhere to it, especially with some higher-performing desktop ram. You can see the same thing with some of the DDR2 ram, going over 1.8v.


"There's no chance that the iPhone is going to get any significant market share. No chance." -- Microsoft CEO Steve Ballmer











botimage
Copyright 2009 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki