Super Talent & TEAM: DDR3-1600 Is Here!
by Wesley Fink on July 20, 2007 11:30 AM EST- Posted in
- Memory
Bandwidth and Memory Scaling
As you saw in our overclocking test, the Super Talent kit reached DDR3-2000+ and the TEAM DDR3-1600 topped out at DDR3-1900. We compared standard or buffered bandwidth on the P965 running DDR2, the new P35 running DDR2, and the new P35 running Kingston DDR3-1375, Super Talent DDR3-1600, and TEAM DDR3-1600.
At DDR3-1066 the DDR2 Corsair Dominator running in the P35-DDR2 board holds a slight lead. This is a pattern you will see throughout many of these test results. If you intend to run only 1066 and nothing higher, the ASUS P5K Deluxe with fast DDR2 beats DDR3 and P965. This is likely due to the very aggressive 4-4-3 timings DDR2 can manage on the P35, but the lead is very small compared to these new Micron Z9 DDR3 memory modules.
The new Z9 DDR3 memory owns the rest of the benchmarks. From 1333 to the highest OC of 2000 nothing comes close to the bandwidth of the new Micron memory chips. The best modules in this roundup are from Super Talent. The TEAM DIMMs are close in every benchmark, but the Super Talent memory wins all the benchmarks at every speed from 1333 up.
The timings are also record-breaking. 5-4-3-9 at DDR3-1066 (1333 strap) are the best memory timings we have seen in DDR3 tests at that speed. Similarly 6-5-4-12 at 1333, 7-6-5-15 at 1600 and 1666, and 9-8-7-18 at DDR3-2000 are the tightest timings seen so far with DDR3 at any of those speeds. The Micron Z9 memory chips achieve speeds and tight timings that we really didn't expect until next year or later.
We also test memory with buffering schemes like MMX, SSE, SSE2, SSE3, etc, turned off. While these features do provide apparent improved bandwidth, the unbuffered bandwidth tends to correlate better with gaming performance. Unbuffered performance does not always follow the patterns of buffered memory performance.
Unbuffered results show the same basic pattern as buffered results. At 1066 speed the best bandwidth is with fast DDR2 on the P35 chipset, and at all other speeds the Super Talent DDR3-1600 and TEAM DDR3-1600 top the results. DIMMs based on Micron Z9 memory chips are the fastest DDR3 you can buy. There will still be variations based on the memory makers' experience and expertise in binning, PCB construction, and SPD programming, but for the time being we expect all of the fastest DDR3 memory to use Z9 chips.
Of course DDR2 could not reach the 1333 speed, and with DDR3 now running up to 2000 MHz and higher DDR3 is looking like the logical choice for high-performance computing. DDR2 is slightly better on the P35 chipset only at the 1066 speed. Now with Z9 chips, DDR3 also outperforms the P965 chipset running fast DDR3 at 1066. Anything higher than 1066 is the domain of fast low-latency DDR3.
As you saw in our overclocking test, the Super Talent kit reached DDR3-2000+ and the TEAM DDR3-1600 topped out at DDR3-1900. We compared standard or buffered bandwidth on the P965 running DDR2, the new P35 running DDR2, and the new P35 running Kingston DDR3-1375, Super Talent DDR3-1600, and TEAM DDR3-1600.
 |
| Click to enlarge |
At DDR3-1066 the DDR2 Corsair Dominator running in the P35-DDR2 board holds a slight lead. This is a pattern you will see throughout many of these test results. If you intend to run only 1066 and nothing higher, the ASUS P5K Deluxe with fast DDR2 beats DDR3 and P965. This is likely due to the very aggressive 4-4-3 timings DDR2 can manage on the P35, but the lead is very small compared to these new Micron Z9 DDR3 memory modules.
The new Z9 DDR3 memory owns the rest of the benchmarks. From 1333 to the highest OC of 2000 nothing comes close to the bandwidth of the new Micron memory chips. The best modules in this roundup are from Super Talent. The TEAM DIMMs are close in every benchmark, but the Super Talent memory wins all the benchmarks at every speed from 1333 up.
The timings are also record-breaking. 5-4-3-9 at DDR3-1066 (1333 strap) are the best memory timings we have seen in DDR3 tests at that speed. Similarly 6-5-4-12 at 1333, 7-6-5-15 at 1600 and 1666, and 9-8-7-18 at DDR3-2000 are the tightest timings seen so far with DDR3 at any of those speeds. The Micron Z9 memory chips achieve speeds and tight timings that we really didn't expect until next year or later.
We also test memory with buffering schemes like MMX, SSE, SSE2, SSE3, etc, turned off. While these features do provide apparent improved bandwidth, the unbuffered bandwidth tends to correlate better with gaming performance. Unbuffered performance does not always follow the patterns of buffered memory performance.
 |
| Click to enlarge |
Unbuffered results show the same basic pattern as buffered results. At 1066 speed the best bandwidth is with fast DDR2 on the P35 chipset, and at all other speeds the Super Talent DDR3-1600 and TEAM DDR3-1600 top the results. DIMMs based on Micron Z9 memory chips are the fastest DDR3 you can buy. There will still be variations based on the memory makers' experience and expertise in binning, PCB construction, and SPD programming, but for the time being we expect all of the fastest DDR3 memory to use Z9 chips.
Of course DDR2 could not reach the 1333 speed, and with DDR3 now running up to 2000 MHz and higher DDR3 is looking like the logical choice for high-performance computing. DDR2 is slightly better on the P35 chipset only at the 1066 speed. Now with Z9 chips, DDR3 also outperforms the P965 chipset running fast DDR3 at 1066. Anything higher than 1066 is the domain of fast low-latency DDR3.
Facebook
Twitter
RSS
31 Comments
View All Comments
Wesley Fink - Friday, July 20, 2007 - link
1333 and 1066 are both at 2.66GHz - which was the best we could do. We would definitely prefer to compare ALL memory speeds at the same CPU frequency as we have done in all memory testing in the past. However, as we point out in the article, with just a 1333 strap and a 333 multiplier it just isn't possible. With boards with 1600.166 and possibly 2000 atraps we can do fixed CPU speed and varied memory speed again.Suggestions for test speeds are welcomed.
rjm55 - Friday, July 20, 2007 - link
I can see where a 1600 strap is now almost a must on motherboards with these new 1600 kits. Does anyone know of ANY Intel P35 motherboard that has support for the 1600 or 1666 strap?LTG - Friday, July 20, 2007 - link
Please cite an example of "break though performance".
It seems any benchmark gains were largely due to CPU speed differences.
Wesley Fink - Friday, July 20, 2007 - link
I consider almost doubling memory speed in less than 2 months qualifies as breakthrough, just as a 6 GHz CPU would be a breakthrough. It is true that memory is just one component in overall performance and that the impact of doubling memory speed is definitely not the same as doubling CPU speed or doubling video speed would be. That still does not change the fact that the Z9 chips are a significant memory development.It is also true that potential gains are dampened by the current lack of straps above 1333 for DDR3. However, those will come sooner, rather than later, now that memory exists that can run at 1600/1666 and 2000.
LTG - Friday, July 20, 2007 - link
Well, you still didn't answer the question so I'll repeat:Whats one single example of "breakthrough performance" provided by this memory?
Wait, let's make it easier - shouldn't the article provide any examples of significant performance differences at the same CPU speed (aside from artificial benchmarks)?
bryanW1995 - Friday, July 20, 2007 - link
A 50% increase in memory speed is not "breakthrough"? This is enthusiast/overclocking memory, it's not designed for the wannabe. Which one are you?DigitalFreak - Friday, July 20, 2007 - link
It's not when there is only a minuscule real world performance increase. It's the same situation as the P4 clock speed crap. The P4 may have run at 3.6Ghz, but it was still bested by an A64 running a Ghz or more slower.TA152H - Saturday, July 21, 2007 - link
The fallacy with your argument is that the tests presented do not include every "real world" situation (no amount of tests could), and there will be situations where the extra memory performance will exhibit extraordinary improvements, depending on the software. I was not crazy about their choice of operating systems either, and you would expect the memory difference to be more in Vista than in XP, simply because Vista uses more memory and resources, and should have a lower cache hit percentage.It's also useful within a hardware context too, not everyone will be buying a Core 2 with 4 MB cache. Right now, yes, it will be what most people get, but when AMD goes to DDR3, and DDR3 prices drop so it becomes mainstream, it will be used on systems with a smaller cache and you'll see a better improvement in speed. So, it's informative.
Also consider that DDR3 does all this with lower voltages than DDR2, so is meaningful in a performance/watt criteria.
If all you are walking away with is a 2.5% improvement with a huge increase in cost, that's not much of interest because it's not worth it for most people. But, extrapolate from that in terms of different hardware and software, and the incredible changes in DDR3 performance lately, as well as the inevitable price drops, and you might get more value out of the review.
DDR2 is obsolete. I said it a month ago, and I'm saying it again. It's low cost, but the performance is not there. It is fading fast (even faster than I thought, to be honest). In less than a month it went from being very competitive in performance and much lower cost against a technology that was showing potential but little real world current value, to already being a low cost, low performance alternative. It is not power efficient either, so all that remains to happen for DDR3 is for the price points to drop. Obviously, the performance delta will increase, but it's already better at that. Reviews like this are useful in that they show this to be true and they will give you a way to plan your next system, or perhaps put off a system purchase until a better time.
Would you buy a DDR2 based system now? It would be like buying a DX9 based video card. Why buy obsolescence when already the next generation is showing real improvements. Time will make the differences greater.
strikeback03 - Monday, July 23, 2007 - link
Speaking of voltage, no complaints over the 2.25V they used with the Super Talent DDR3? You complained about 1.7 volts in the Kingston article after all.And the comparison with DX9 video cards is not a great once for making your point either, as you already know that most people who comment here disagree with you on that point as well.
domski - Saturday, July 21, 2007 - link
Do you have any real-world power consumption figures to back up these assertions?
Don't get me wrong -- I believe you. But I would be interested to see the magnitude of the difference in both absolute power consumption and performance per watt.