Sign in with
Sign up | Sign in


DDR3 Memory Scaling On AMD's Phenom II X4

We have done many RAM performance explorations in the past, as the industry has been quick with releasing ridiculously expensive memory products that are capable of running at impressive clock speeds—but without always delivering substantial performance benefits.

Our general opinion of high-end memory is pretty straightforward: premium RAM is highly important if you’re trying to squeeze the last bit of performance out of your system. In such a case you may even require high-end RAM to be able to support hardcore overclocked processor settings. However, the performance benefit delivered just by the main memory is typically small, making premium memory a must-have primarily for the affluent enthusiast market-only.

Our results show that faster memory and quicker timings both introduce performance advantages that are mostly visible in synthetic benchmarks, while the impact on real-life performance usually is small. Games benefit just a little, and you need applications that intensively tax the memory to find serious performance benefits.

While DDR3-1600 operating under tight timings will undoubtedly give you the best performance, and DDR3 below 1066 should not be used for the sake of performance, the best compromise clearly is DDR3-1333 at reasonable timings. In most benchmarks, the difference between 1600 and 1333 speeds is negligible, but 1333 would still give you a larger benefit than the tiny gap between 1333 and 1600. In addition, DDR3-1333 is about to become the performance mainstream level, meaning acceptable prices for low-latency memory kits.
What if… Processor Speed Comes into the Equation?

We couldn’t resist doing one more benchmark run at a slightly overclocked processor clock speed, which serves as a great example in the event that you decide to spend extra money on a faster processor rather than on premium memory. Overclocking the Phenom II X4 955 from 3.2 to 3.4 GHz results in a hypothetical Phenom II X4 965 at 3.4 GHz, which we might see some day. We used the slow DDR3-800 memory setup to show that a processor speed bump can actually do much more than any upgrade in memory speed. Click here for charts with higher clocked processor.

Display all 40 comments.
This thread is closed for comments
  • 6 Hide
    cangelini , June 4, 2009 6:50 AM
    judeh101DDR3 memory scaling on AMD's Phenom II X4, why don't cha do one for Intel as well? I'd love to see the performance on the Core i7 platform

    It's coming!
  • 0 Hide
    cangelini , June 4, 2009 7:45 AM
    Much thanks judeh, I appreciate your feedback!

  • 6 Hide
    Antman56 , June 4, 2009 7:59 AM
    The real trick to seeing a greater discrepancy between different clock speed memory(s) could be done by overclocking the CPU-NB. Using SiSoftware Sandra, I get 13.2 GB/s with DDR2 @ 1066 MHz (5-5-5-15) on a 3.4 GHz Phenom II 940 X4 with a 2.4 GHz CPU-NB.

    Overclocking that Integrated Memory Controller is key!

    ...Just a suggestion :p 
  • -2 Hide
    empstar , June 4, 2009 8:34 AM
    why always missing something..... where's the DDR2 RAM ?? should include in the chart!! very disappointed.
  • 0 Hide
    EQPlayer , June 4, 2009 8:47 AM
    Yarr, good point. Overclocking the CPU/NB on these chips seems to have nearly (well that might be a bit of an exaggeration) as much of an effect as overclocking the CPU itself.
  • -1 Hide
    DjEaZy , June 4, 2009 9:25 AM
    ... it is getting better and better with TOM's... more information to compare... for the last month it's enjoyable to read...
  • 0 Hide
    Anonymous , June 4, 2009 10:16 AM
    Why you didn't have tried a 'synchronized' configuration with base clock, north-bridge, memory and cpu core running on integer or semi-integer divisors ? like 266MHz base clock, 1068 MHz memory clock, 2136 MHz north bridge & HT clock and 3199 core clock.
    This configuration is possible even with DDR2, with lower latencies memories and is much closer to the DDR3/1333 configuration than you think.
  • 2 Hide
    cinergy , June 4, 2009 10:43 AM
    Very nice bench for AMD people.
  • -7 Hide
    apache_lives , June 4, 2009 10:59 AM
    judeh101DDR3 memory scaling on AMD's Phenom II X4, why don't cha do one for Intel as well? I'd love to see the performance on the Core i7 platform

    Why does everyone vote down a valid question?
  • -1 Hide
    apache_lives , June 4, 2009 11:04 AM
    and this is the same old picture - mainstream ram has always been the best price:p erformance, no point in chancing extreme memory, and "high end" memory is only good for overclocking etc
  • -1 Hide
    chripuck , June 4, 2009 11:47 AM
    apache_livesand this is the same old picture - mainstream ram has always been the best priceerformance, no point in chancing extreme memory, and "high end" memory is only good for overclocking etc

    I wouldn't call the two top performers mainstream... with those timings there at least a step above main stream. Now to your point they're definately not the top of the line, but anyone with a clue about electronics knows the best bang for your buck is never the cheapest and never the most expensive... the sweet spot is always in the middle.
  • -3 Hide
    Kill@dor , June 4, 2009 12:39 PM
    Can't wait for Intel's review as well.
  • 1 Hide
    coopchennick , June 4, 2009 12:43 PM
    Soo... I guess Toms just decided not to do the SBM giveaway?
  • -1 Hide
    Kill@dor , June 4, 2009 1:00 PM
    I have to kind of admit this was a little unfair because DDR3 2000 was not in this line up...why i'm not sure. Intel's processors can utilize DDR3 2000 very well with timings like 9-9-9. But even so, its important to show how DDR3 2000 can perform with AMD and the benefits it has for a stable high clock on your system. If you plan on overclocking 3.0GHz to about 3.6 or 3.8 DDR3 1333-1600 is key in my opintion. Anything higher than 4.0GHz will benefit stability with DDR3 2000. I hope Toms can get a review to show this...
  • -1 Hide
    buzznut , June 4, 2009 1:31 PM
    I was disappointed that the memories weren't tested more with overclocking. You did a little bit with the 800 speed, but who is gonna run their ram at that speed?

    Well, nice article. It makes me wonder why ddr3 memory is not more effective, does not offer much beyond ddr2. It seems to make a considerable difference in video cards, ddr2 vs ddr3. Will they skip ddr4 and go to ddr5 for mainboard memory, like the upper end ATI cards?

    As it has been for a few years, latency is the most important factor with AMD processors.
  • 0 Hide
    sublifer , June 4, 2009 1:57 PM
    Only testing in 2T command rate?
  • 2 Hide
    Anonymous , June 4, 2009 2:26 PM
    Quote "Why does everyone vote down a valid question?"

    Because if that person would have read the first page of the article, they would have seen that a Intel article is coming.......
  • 2 Hide
    HVDynamo , June 4, 2009 2:30 PM
    apache_livesWhy does everyone vote down a valid question?

    Because it says there is an intel article to follow at the bottom of the first page of the article. He posted before reading.
Display more comments