Related Articles

Overclocking

I made the overclocking comparison as fair as I could - all the modules were tested in exactly the same system, with the same software.

As a matter of fact, I tested all modules at the same maximum of 2.3V as well!

It was quite interesting to see how far each module would go at a 4-4-4-12 timing; and you should know that I only reported results that gave me no difficulties whatsoever during the testing - so these settings (at least with my particular samples of memory) were quite stable; they ran Windows XP, Doom 3, Sandra, WinRAR and RightMark memory without a single hickup.

Conclusion

I was pleased to see how well the Ballistix Tracer PC2-8000 did on the 975X chipset - as did the other modules I tested for this article.

One thing is for sure: There can be quite a difference between running 800-5-5-5-15 and say 1000-4-4-4-12 :-)  Fine, that should be blindingly obvious to everyone, however the difference it makes on benchmarks such as WinRAR and Doom 3 points out that you can notice the difference in something other than a memory benchmark.

It's worth pointing out two numbers from the RightMark Read benchmark. The Ballistix Tracer running at 832-4-4-4-12 was 9.75% faster than the result with it running at 800-4-4-4-12 ... so we had a 9.75% performance increase with only a 4% memory clock speed increase - with the processor running at 2.67GHz in both cases.  The reason why is that the higher result was obtained when the FSB was run at 1333MHz instead of 1066MHz!  Again this shows how Intel based computers memory bandwidth is still crippled by being funneled through the same old FSB.

Even if we ignore all of the overhead - including cache coherency traffic - the fact is that the FSB based design imposes some very hard limits, and increases latency due to the memory controller being in the chipset.  A 1066MHz data rate (266MHz quad data rate) FSB *theoretically* has the capability of transferring 8.528GB/sec; assuming there is no bus overhead and no cache coherency traffic. A 1333MHz data rate (333MHz quad data rate) FSB theoretically has a 10.66GB/sec bandwidth - again, assuming no overhead. If we assume an optimistic mere 5% overhead, the most we could achieve would be 8.15GB/sec on a 1066MHz FSB, and 10.12GB/sec on a 1333MHz FSB.

Now modern chipsets have dual channels for memory, which is more bandwidth than can be accomodated by the processor - however some of it can be put to good use for DMA transfers to/from hard drives and GPU's, so the extra bandwidth is not totally wasted - just not available to the CPU - which arguably needs the bandwidth the most.

Ok, what all this mean to you? Simple.  You will see a nice jump in memory performance when Intel releases processors with integrated memory controllers. I can't tell you how much I am looking forward to seeing latency figures in the 20ns range, and maximum memory read bandwidth in the 16GB/sec range!

But back to the Tracer PC-8000... This memory kit performed exceedingly well.  Not only does it have good specs to begin with, when pushed it ran WAY above spec and allowed for excellent overclocks.  As a matter of fact, the Ballistix Tracer PC8000, our first memory review of 2007, turned in the best memory performance we've achieved on a 975X Intel platform to date, and as such is the best performing memory that we currently have in our lab!  That's one heck of an achievement when you consider we review much of the highest performance memory here.

1084MHz at 4-4-4-12 timing. Simply outstanding.

Strongly Recommended :)