Lights out Performance! Crucial Tracer Memory

Memory by JNav89GT @ 2005-03-17

Do you want your cake and to eat it too? Do you have a need for speed and bling? Well step inside as we?ll skewer a few sticks of Crucial Ballistix Tracer DDR and DDR2 over our crustacean Barbie and see if these memory sticks are true speed demons, or posers with Type R badging.

  • prev
  • next

Max Frequency&Synthetic Benchm

Maximum Stable Frequencies

Madshrimps (c)


The Ballistix Tracer PC4000 shows excellent characteristics with the ability to maintain very tight timings up to 240MHz@2.85v in my testing. Loosing the timings to CL3 3:3:8 allowed a good jump in frequency, however when further timing adjustments were made, smaller gains were realized. With only a 7MHz gain from CL3 3:3:8 to CL3 4:4:8 the small jump in CPU and HTT frequency couldn't overcome the penalty from the more relax timings so I didn't continue testing as there would be no benefit for the A64 user.

Madshrimps (c)


Using the 14x Multiplier and the 3:4 ram multiplier, I was able to find the max Memtest86+ passing frequencies for each of these timing levels. I was able to raise the front side bus to a respectable 225MHz which produced a 300MHz ram speed when using the Cas3 3:3:8 timings. The Ballistix Tracer DDR2 really seemed to prefer the Cas4 4:4:14 timing level though as I was able to then scale the memory frequency of 354, or DDR2 708 if you prefer.


Synthetic BenchMarks

Madshrimps (c)


I was really expecting to see a difference between the tight and moderate latency levels. In previous testing I have noted that tighter timings can often mean a 1+ second difference. However, that was not seen here, and just shows me once again that the proof is in the pudding. That said, I suppose a possible explanation could be made that give a slightly higher or lower CPU frequency, the impact of the ram timings may have been enough to push the scores in a different direction, and that the given test setup was a "tweener" meaning in between areas where ram timing could make a difference. I'll leave that to others to discuss as I'm getting a headache just thinking about it.

Madshrimps (c)


SiSoft Sandra Memory Bandwidth Buffer'd shows gains as front side bus or HTT frequencies scale. Really no surprise here other than in the P4 setup where the ram frequency is actually less at the higher bus speed due to the 1:1 ram setting versus the 3:4 multiplier at the lower bus. However, this is easily explained by the relative inefficiency of the P4 bus and the inability of the ram divider to seemingly improve the bandwidth.

Madshrimps (c)


The bandwidth trend continues with Everest as was seen with Sandra although the gains do seem much smaller.

Madshrimps (c)


I really like the Everest Latency benchmark. What strikes me about this benchmark, is the relation you can apply in A64 system to a memory speed when adjusting the timings. For instance, will higher speed and looser timings improve performance? A quick and dirty answer is partly found with the latency test as lower is better, and if your latency is worse from the higher frequency and moderate timings, then the performance could trend that way too. This is just a casual observation I have made, and in no way meant to define absolute performance, as there are just too many variables to consider. However, if you play around with different frequencies and timings, you may notice the same trend in your own testing.

That said, the P4 latency is absolutely horrible compared to the A64. Handicapped by the net burst technology, and not aided by an on-die memory controller and there's your explanation. Yet the P4 still performs well so who am I to throw stones.

Madshrimps (c)


PCMark2004 seems to favor the lower latency in the A64 setup, and the higher frequency memory and CPU speed on the P4 platform. In hindsight, I really should have tested the P4 at same frequency of 3990MHz, but I really wanted to see if the added front side bus frequency and tighter memory timings could offset a small CPU frequency lead.
  • prev
  • next