Given current pricing, those looking to build either a budget or high-end computer based on an AMD processor, might as well invest in the new AM3 platform. Just a few months ago we were suggesting budget builders avoid the AM3 platform as it offers no real performance advantage when compared to a similar AM2+ setup, and more importantly it forces users upon the more expensive DDR3 memory standard.

However it is now possible to purchase a decent DDR3-1333 memory kit with a 4GB capacity for roughly $65 US, while a decent DDR2-1066 kit is just $15 US cheaper. The improved DDR3 pricing has helped to make the AM3 platform a more attractive option now, and although it still fails to deliver any real performance benefits, it does have one important advantage, and that is its upgrade path.

The AM2+ platform is nearing the end of its life, and while a Phenom II X4 940 processor will keep most users smiling for some time to come yet, chances are the next upgrade will see you replace not only the processor but also the memory and motherboard, making for quite an expensive upgrade venture.

Therefore those that make the move to the AM3 platform now, rather than invest in the slightly cheaper AM2+ platform, will be ensuring that they have a feasible means of upgrading down the track. Both AM2+ and AM3 motherboards use the same chipsets, and therefore users can expect to pay roughly the same price regardless of the supporting socket.

Back when building an AMD computer based on the AM2+ platform was the smart thing to do, a decent DDR2-1066 memory kit was the way to go. The price difference between 800MHz and 1066MHz DDR2 memory was trivial, while the higher clocked memory did deliver a slight performance advantage.

However when it comes to DDR3 memory there tends to be quite a large variance in frequency and even timings. For example, a 4GB DDR3-1066 kit can be purchased for $60 US supporting CAS7 timings, while for just $75 US it is possible to purchase a 4GB kit that can work at 1600MHz with CAS8 timings. If you want to spend even more money you can get DDR3-1600 memory that can work at even tighter timings.

The question is, to get the most out of your AM3 processor, what kind of memory do you need exactly? Is low-latency 1066MHz memory the way to go, or is high-speed memory with less aggressive timings a smarter option? With such a small difference in price it could make more sense to just go with the high-speed memory. However if it turns out that lower clocked memory with tighter timings is just as fast or faster, then saving a few dollars where you can would be the smartest option.

In order to try and determine what is the best course of action to take here, we have paired AMD’s flagship AM3 processor, being the Phenom II X4 955 Black Edition, with the ASUS M4A79T Deluxe motherboard and two sticks of Crucial Ballistix Tracer 4GB DDR3 memory. This low-latency memory is designed to work at 1333MHz, though it can handle frequencies well beyond 1600MHz with respectable timings.

Phenom X4 Test System Specs - AMD Phenom X4 II 955 (AM3) - x2 Crucial Ballistix Tracer 2GB DDR3-1333 (CAS 6-6-6-20) - ASUS M4A79T Deluxe (AMD 790FX) - OCZ GameXStream (700 watt) - Seagate 500GB 7200-RPM (Serial ATA300) - ASUS GeForce GTX 285 (1GB) Software - Microsoft Windows Vista Ultimate SP1 (64-bit) - Nvidia Forceware 185.85

First up we are going to take a quick look at the impact running the memory in either the Ganged or Unganged modes has. By default most AMD motherboards should opt for the Unganged mode, as traditionally it does provide better real-world performance.

As you can see, running the memory in the Ganged mode does deliver slightly better bandwidth performance when opposed to the Unganged setting. When testing with EVEREST Ultimate Edition the write performance remained the same regardless of how the memory was configured, while the Ganged mode did deliver 4% more performance.

Although the Ganged mode does deliver better memory bandwidth performance it is not as effective in real-world applications, as we see here when testing with WinRAR. Although the margins were small, the Unganged configuration was faster in both the single and multi-thread tests.

When using Street Fighter IV to measure gaming performance, we again see that the difference between running the memory in either the Ganged or Unganged mode is minimal. Still, the Unganged configuration was slightly faster, and that is why most users prefer this setting.

First up we have used EVEREST Ultimate Edition once again to measure the memory bandwidth of each configuration. Again we find that regardless of the settings, the write performance remains very much the same. The read performance does on the other hand change, with an 8% difference between the CAS6 and CAS9 configurations.

This had a mild impact on the WinRAR performance, and again we saw an 8% performance difference between the CAS6 and CAS9 configurations, at least for the single thread test anyway. When running the multi-thread test there was just a 6% performance margin favoring the CAS6 configuration. There was almost no difference between the CAS6 and CAS7 configurations. However there was a noticeable drop between the CAS7 and CAS8 configurations, and while it was nothing extreme, the biggest single drop was seen here.

However when it came to gaming, the only configuration that was a little off the pace were the CAS9 timings. That said, at 1920x1200 where the GPU was the bottleneck, the CAS9 timings matched the CAS8 and CAS7 configuration. It would seem that those interested in gaming performance should not concern themselves with memory timings on the AM3 platform, and this is much the same case with the latest Intel Core i7 processors.

Interestingly, when testing with SiSoftware Sandra 2009 the 1333MHz and 1600MHz configurations delivered the same 13.0GB/s bandwidth. This made them both 10% faster than the DDR3-1066 memory, and a whopping 38% faster than the DDR3-800 memory configuration.

Once again the EVEREST Ultimate Edition bandwidth test reports the same write performance regardless of the memory settings. However this time we do see a difference between the DDR3-1600 and DDR3-1333 memory configurations, with the 1600MHz frequency delivering a 6% greater read result. Furthermore, the DDR3-1333 memory was 9% faster than the 1066MHz configuration.

The memory frequency appears to have by far the greatest impact on performance. When testing with WinRAR we see a fairly significant difference in performance between the various configurations. The DDR3-1333 memory is 5% slower than the 1600MHz memory, while it was 5% faster than the 1066MHz memory in this test.

The Super PI performance was not greatly affected by the memory frequency and the 1066, 1333 and 1600MHz configurations all delivered roughly the same performance.

When it came to gaming we found something most unusual that we cannot explain. Despite providing greater memory bandwidth and better application performance, the DDR3-1600 configuration continually delivered weak 1024x768 gaming performance. This had us a little confused, but after running the test several times and checking the settings each time, the results consistently came back the same.

Nevertheless, for gamers the point would once again be that memory frequency is irrelevant, as at 1920x1200 the DDR3-800 configuration delivered roughly the same result as the DDR3-1333 and 1600 configurations.

Ideally DDR3-1333 memory with CAS6-7 timings would be best, but again we found that CAS8-9 timings made very little difference, so it is hardly worth spending extra money on the lower latency memory if it is fetching a price premium. In fact, surprisingly the Crucial Ballistix Tracer 4GB DDR3 memory that we used in this article can be had for just $70 US.

Given that the Crucial Ballistix Tracer is designed to work with CAS6 timings at 1333MHz, makes it ideal AM3 memory for just about everyone. This memory can be safely overclocked well beyond its default operating specification, which also makes it suitable for overclockers.

Something we did not cover in this article was the impact memory capacity has on performance, and this is an area that we have covered in the past. However given that no one in their right mind would upgrade to the AM3 platform with anything less than a pair of 2GB memory sticks, we did not see the need to test different capacities.

Furthermore, while 8GB of memory may allow the user to do more at once, it will not influence performance in any way, unless as we said, you are doing some serious multi-tasking. Therefore we feel that for most users a 4GB memory kit is the sweet spot, and recommend that anyone building a new PC aim for at least this much memory.

Based on our findings there is a $10 US price difference between DDR3-1333 memory supporting either CAS6 or CAS9 timings, with the latter being the slightly cheaper option. The OCZ Reaper HPC DDR3-1333 memory kit is another fine example, as it supports CAS6-6-6-20 timings while it costs just $70 US for the 4GB kit.

Finally, those looking solely at gaming need not worry about timings or frequency, as neither really affected performance, and as we mentioned earlier, this is also the case with the new Intel Core i7 processors. Still, with low-latency DDR3-1333 memory being so affordable now, it really does not matter.

Reviewed By Steven Walton