Bandwidth and Memory Scaling
One of the surprises in comparing DDR2 performance on AM2 and Core 2 Duo was the much better memory bandwidth found on the AM2 platform courtesy of the on-chip memory controller. Unfortunately, this did not translate into significant performance improvements compared to a similar AMD processor running DDR. At that point we concluded that Core 2 Duo was not particularly bandwidth sensitive, since it made very good use of the memory bandwidth available.
In our earlier review we were really comparing the DDR2 memory controller on AM2 to the 975X chipset memory controller, since Intel continues to place the memory controller in the chipset. We have speculated since then whether an improved memory controller in a socket 775 chipset would bring with it improved performance.
P965 brought very minor changes, mainly in the straps and overclocking ability of the memory. The NVIDIA 680i/670/650 actually shows decreased buffered bandwidth, but unbuffered bandwidth is about the same as P965. This reinforced the notion that memory bandwidth didn't matter much with Core 2.
To begin our investigation into DDR3 performance, we compared Standard or Buffered bandwidth on the P965 running DDR2, the new P35 running DDR2, and the new P35 running DDR3. As you can see the results are very interesting.
While the purpose of this review was to compare DDR3 and DDR2 performance, something completely different emerged from the memory bandwidth tests. Namely, the memory controller on the P35 is definitely an improvement over the P965 memory controller. This is evident whether the P35 is running DDR2 or DDR3 memory.
In those cases where we can run timings the same or close to the same, as in 800 memory speed performance, DDR2 and DDR3 results are virtually identical. By 1067 the current slow DDR2-1067 timings of 7-7-7-20 are performing just as well as DDR2 running at 6-6-6-15. The superior timings of DDR2-1067 at 4-4-3 still provides the best bandwidth at that speed. Of course, DDR3 is currently alone at the 1333 memory speed, but even with the current slow 9-9-9-25 timings it performs nearly as well as DDR2-1067 at 4-4-3 timings.
We normally also test memory with buffering schemes like MMX, SSE, SSE2, SSE3, etc. turned off. While these features do provide apparent improved bandwidth, we have found the unbuffered bandwidth to correlate better with real-world application performance. Unbuffered performance does not always follow the patterns of buffered memory performance.
Unbuffered results show the same basic pattern as buffered results in this case. Here DDR3 is clearly the best performer at the same slow timings at DDR2-800, with DDR2 on the P35 behind about 3% and DDR2 on P965 about 12% lower. DDR2 is still faster at the better timings available with current DDR2 memory.
In Standard/Buffered memory bandwidth, the P35 (Bearlake) chipset is providing a 16% to 18% improvement in memory bandwidth compared to the P965. This is a significant improvement. The Unbuffered improvement is smaller, in the range of 4% to 8%. These bandwidth improvements may or may not translate into improved system performance. We will examine that in the SuperPi and Gaming benchmarks.
One of the surprises in comparing DDR2 performance on AM2 and Core 2 Duo was the much better memory bandwidth found on the AM2 platform courtesy of the on-chip memory controller. Unfortunately, this did not translate into significant performance improvements compared to a similar AMD processor running DDR. At that point we concluded that Core 2 Duo was not particularly bandwidth sensitive, since it made very good use of the memory bandwidth available.
In our earlier review we were really comparing the DDR2 memory controller on AM2 to the 975X chipset memory controller, since Intel continues to place the memory controller in the chipset. We have speculated since then whether an improved memory controller in a socket 775 chipset would bring with it improved performance.
P965 brought very minor changes, mainly in the straps and overclocking ability of the memory. The NVIDIA 680i/670/650 actually shows decreased buffered bandwidth, but unbuffered bandwidth is about the same as P965. This reinforced the notion that memory bandwidth didn't matter much with Core 2.
To begin our investigation into DDR3 performance, we compared Standard or Buffered bandwidth on the P965 running DDR2, the new P35 running DDR2, and the new P35 running DDR3. As you can see the results are very interesting.
| Standard (Buffered) Sandra XI.SP2 Memory Bandwidth - 2.66GHz | |||
| Memory Speed | P965 ASUS P5B Dlx |
P35 DDR2 ASUS P5K Dlx |
P35 DDR3 ASUS P5K3 Dlx |
| DDR2-800 3-3-3-9 | 5531 | 6456 | - |
| DDR2-800 5/6-6-6-15 DDR3-800 6-6-6-15 |
5207 | 6143 | 6156 |
| DDR2-1067 4-4-3-11 | 5782 | 6811 | - |
| DDR2-1067 5/6-6-6-15 | 5712 | 6621 | - |
| DDR3-1067 7-7-7-20 | - | - | 6613 |
| DDR3-1333 9-9-9-25 | - | - | 6757 |
While the purpose of this review was to compare DDR3 and DDR2 performance, something completely different emerged from the memory bandwidth tests. Namely, the memory controller on the P35 is definitely an improvement over the P965 memory controller. This is evident whether the P35 is running DDR2 or DDR3 memory.
In those cases where we can run timings the same or close to the same, as in 800 memory speed performance, DDR2 and DDR3 results are virtually identical. By 1067 the current slow DDR2-1067 timings of 7-7-7-20 are performing just as well as DDR2 running at 6-6-6-15. The superior timings of DDR2-1067 at 4-4-3 still provides the best bandwidth at that speed. Of course, DDR3 is currently alone at the 1333 memory speed, but even with the current slow 9-9-9-25 timings it performs nearly as well as DDR2-1067 at 4-4-3 timings.
We normally also test memory with buffering schemes like MMX, SSE, SSE2, SSE3, etc. turned off. While these features do provide apparent improved bandwidth, we have found the unbuffered bandwidth to correlate better with real-world application performance. Unbuffered performance does not always follow the patterns of buffered memory performance.
| Unbuffered Sandra XI.SP2 Memory Bandwidth - 2.66GHz | |||
| Memory Speed | P965 ASUS P5B Dlx |
P35 DDR2 ASUS P5K Dlx |
P35 DDR3 ASUS P5K3 Dlx |
| DDR2-800 3-3-3-9 | 4226 | 4536 | - |
| DDR2-800 5/6-6-6-15 DDR3-800 6-6-6-15 |
3668 | 3975 | 4098 |
| DDR2-1067 4-4-3-11 | 4608 | 4926 | - |
| DDR2-1067 5/6-6-6-15 | 4389 | 4557 | - |
| DDR3-1067 7-7-7-20 | - | - | 4547 |
| DDR3-1333 9-9-9-25 | - | - | 4702 |
Unbuffered results show the same basic pattern as buffered results in this case. Here DDR3 is clearly the best performer at the same slow timings at DDR2-800, with DDR2 on the P35 behind about 3% and DDR2 on P965 about 12% lower. DDR2 is still faster at the better timings available with current DDR2 memory.
In Standard/Buffered memory bandwidth, the P35 (Bearlake) chipset is providing a 16% to 18% improvement in memory bandwidth compared to the P965. This is a significant improvement. The Unbuffered improvement is smaller, in the range of 4% to 8%. These bandwidth improvements may or may not translate into improved system performance. We will examine that in the SuperPi and Gaming benchmarks.
Facebook
Twitter
RSS
45 Comments
View All Comments
RichyDitch - Tuesday, February 17, 2009 - link
Hey there, I'm still not sure what would be better. I have an AsRock P35 motherboard, and am totally confused as to what would be better when it comes to DDR2 or DDR3.The motherboard it's self can hold up to 8gb of PC2 8500 1066mhz duel channel DDR2 memory, and up to 4gb of PC3 10600 1333mhz duel channel DDR3 memory.
What would end up being better, if I get the max amount of memory for the motherboard. The DRR2 specifications I mentioned or the DDR3 specifications mentioned?
mikegonzalezrubio - Saturday, January 24, 2009 - link
I WANNA BUY A NEW LAPTOP BUT I DONT KNOW WHAT IS THE BEST CHOISE IS BUY A DDR2 LAPTOP OR DDR3 LAPTOP MY OPTIONS ARE:HP - Pavilion Laptop with Intel® Centrino® 2 Processor Technology - Bronze/Chrome
Model: dv7-1285dx
WITH
Intel® Centrino® 2 processor technology with interrelated Intel® Core™2 Duo processor P8600
Intel® Wi-Fi Link 5100AGN (802.11a/b/g/n) network connection and extended battery life capability.
6GB DDR2 memory
For multitasking power, expandable to 8GB. 1066MHz frontside bus, 3MB L2 cache and 2.4GHz processor speed.
Multiformat DVD±RW/CD-RW drive with double-layer support
Records up to 8.5GB 17" WXGA high-definition widescreen display
With BrightView technology and 1440 x 900 resolution
500GB Serial ATA hard drive (5400 rpm)
NVIDIA GeForce 9600M GT graphics
*LINK OF THE PRODUCT http://www.bestbuy.com/site/olspage.jsp?skuId=9166...">http://www.bestbuy.com/site/olspage.jsp...&typ...
OR TO BUY A DELL Studio XPS 16
WITH
Intel® Core™ 2 Duo P8400 (3MB cache/2.26GHz/1066Mhz FSB)
LCD PanelEdge-to-Edge HD Widescreen 16.0 inch WLED LCD (1366x768) W/2.0 MP
8X DVD+/- RW(DVD/CD read/write) Slot Load Drive
Specifications - DVD+/-RW or Bluray Drive
4GB2 Dual Channel DDR3 SDRAM at 1067MHz (2 Dimms)
320GB3 7200 RPM SATA Hard Drive
ATI Mobility RADEON® HD 3670 - 512MB4
Intel® 5100 WLAN Wireless-N (1x2) Half Mini Card
LINK OF THE PRODUCT IS http://configure.us.dell.com/dellstore/config.aspx...">http://configure.us.dell.com/dellstore/...kc=produ...
WHO CAN TELL ME WHAT IS THE BEST CHOICE HERE BETWEEN THOSE LAPTOP...?
thanks for your help i wait an aswer
sincerely
MIGUEL ANGEL
Shadowmaster625 - Thursday, May 17, 2007 - link
No there are not just two conditions. There is a third condition that I wish you'd start considering: How it screws over everyone who wants to upgrade. This article has proven that Intel can squeeze more performance out of their memory controller. What this ultimately shows is that improvements in memory controller design actually outweigh the change to a new memory standard! Is DDR2 18% faster than DDR? (Maybe now, just barely. After 3 years!) Yet intel just pulls a new memory controller out of their wazoo that does 18% better? I smell a big rat and anadtech aint talking about it.
TA152H - Thursday, May 17, 2007 - link
You're not the first person to be frustrated by how poorly memory speed increases, but there are a few things to consider.For one, the new controller is NOT 18% faster than the old, that is only in bandwidth, not latency. Latency is VERY important, and discounting is a huge fallacy. And yes, bandwidth has increased a lot more than 18% from DDR.
Secondly, this isn't by any means the end of DDR2, it's a forward looking move that paves the way for DDR3. DDR2 will continue to dominate the market, and by the time it's obsolete you'll have DDR3 ready. What would happen if Intel waited for DDR2 to already be obsolete before they introduced DDR3? You'd have that long wait while memory companies worked on the new technology, and prices were very high and performance wasn't what it should be. After a couple of years, DDR3 would finally be mature in performance and cost, but nearly obsolesence itself. So, Intel is just getting things started so when DDR3 does become necessary, it is ready. It's like a pipelined processor, Intel is starting DDR3 in stage one while DDR2 is in stage 2.
Also, keep in mind that DDR2 is not made to go much faster than it is, and will run into a wall. That's where DDR3 comes in. Again, it's forward looking by Intel. DDR2 can still increase a bit, but by the time it runs into a wall, DDR3 will be right there to go to higher speeds.
Also keep in mind the number of people that upgrade is very, very small. It's not their intention to screw people over, but it's something they are willing to do because of the importance of evolving technology, and the miniscule percentage of people that upgrade processors without motherboards. It's a necessary evil.
sprockkets - Wednesday, May 16, 2007 - link
Funny how we can let you benchmark hardware not coming out till this Autumn but we can't let out any info or benchmarks on a new chipset coming out in 5 days.
cornfedone - Wednesday, May 16, 2007 - link
Anyone who thinks 2-5% increase in system performance is a big deal needs to get their head examined as it don't mean nothing real world. Most people couldn't even see a 2-5% system performance increase on their best day. Bearlake is more hype with no tangible performance increase.TA152H - Wednesday, May 16, 2007 - link
Wow, that's a really uninformed remark.You might not notice X-Rays, but they can kill you. So don't talk about this babble with tangible (it's a poor choice of words, since it literally means "touchable" and it's bad enough Anand overuses it) and what's noticeable. You're getting an amazing increase from a chipset, and this is the mainstream chipset. On top of that, it supports DDR2 and DDR3 so it pushes the technology envelope forward. Getting any increases in memory performance is simply amazing at this point, since chipsets are so mature it's not like there is a low hanging fruit, and Intel processors have such a large cache it makes memory performance less important than would otherwise be. It's a fantastic chipset, arguably the next 440BX. I knew it was good, but even I'm shocked at just how good it is. Kind of kills the argument for the on-die memory controller, which I never was completely sold on. It's scary though, it makes you wonder just how well Intel will do with that if they can get this type of performance with the memory controller on the chipset.
It's an amazing chipset, it is shocking in terms of performance, so much so I doubt anyone thought it was possible. Give them their kudos when they deserve it, because they do with this bad boy. I wish I knew how they did it, but my guess is by finally moving their chipsets to modern lithography, they were able to include a lot more buffers and run it faster without using too much power. I am really clueless though, that's pure speculation. I don't know how they made this so much better than everything else. It's shocking.
bldckstark - Wednesday, May 16, 2007 - link
Shockingly enough, 2% is within the error of the tests performed, and therefore is not, statistically speaking, significant. Not only that, but the cost of progress is passed on to the consumer, who usually rates the speed of their computer based upon how fast their favorite website loads, not on how fast it performs computations. Shocking, I would say.TA152H - Wednesday, May 16, 2007 - link
So, you're saying that on every single test by weird coincidence, the Bearlake is higher just be statistical scatter? You're saying the huge increase in bandwidth is somehow also weird coincidence?How do you know what the margin of error is anyway? When you get repeatable results where one is always higher, you can conclude pretty easily that they are real and not statistical scatter.
And you know the cost of Bearlake how? You are sure it's more expensive? It probably uses less power, since it's made on finer lithography. So, you're talking about something you probably don't know anything about. I do agree that most people don't need the latest and greatest, so I fundamentally agree with that part, but who is going to make them buy this chipset? If they want cheap, they can still buy something cheap, if this isn't, which I don't know. But, some people need performance, and this is a great item for that, and it's a real accomplishment from Intel.
By your perverse logic, all improvements that don't increase web speed are immaterial. That's clearly wrong. That only applies to some people.
You're completely illogical.
OrSin - Wednesday, May 16, 2007 - link
Thats total system performance, from 1 part. Memory performance is 16% and that huge from a low cost part. Unless your adding a $400 video card over $200 card you will not notice the diffenecen either. One part theats 15% more in price then a similar part will rarely give you 5% improvement.Remember this is a systems and it mean each part gives some improvement to make a better system.
Yeah you will not notice 5%. Bet then why get a part that works 5% slower when the cost is similar. Some people are never happy.