Crucial Tracer PC2-5300 revisited - C19 Performance Review

Memory by KeithSuppe @ 2005-11-14

We first reviewed Crucial?s Tracer series in our ...Lights Out Performance article authored by the magnanimous JNav89GT. The DDR2 part in that test performed well on the P5AD2-E Premium at 704Mz 4-4-4-12. Today we?ll retest Tracer PC2-5300 on the Asus P5ND2-SLI Deluxe. We?ll test for higher clock speeds, tighter timings, and compatibility with the nVIDIA?s C19 chipset. At the current sale price Tracer in 1024MB-kit form may be the ergonomic sweetheart and performance value of the year.

  • next

Intro

We first reviewed Crucial's Tracer series in our ...Lights Out Performance article authored by the magnanimous JNav89GT. The DDR2 part in that test performed well on the P5AD2-E Premium at 704Mz 4-4-4-12. Today we'll retest Tracer PC2-5300 on the Asus P5ND2-SLI Deluxe. We'll test for higher clock speeds, tighter timings, and compatibility with the nVIDIA's C19 chipset. At the current sale price Tracer in 1024MB-kit form may be the ergonomic sweetheart and performance value of the year.

Madshrimps (c)


I've tested several "lighted" memory modules in the past and while Madshrimps tested Ballistix Tracer I haven't had the opportunity personally. I must say with all due respect to Corsair, Crucial has truly produced a unique product. The back-lit DIMM effect is most alluring as well as the intensity of the LEDs. The illumination isn't simply brighter; the hues are warm and deep. If you’re going to twinkle this may be preferable. Corsair's LED are designed less for aesthetics acting as bank usage meters. Seen below Ballistix Tracer compliments a "lighted" system quite well.

Madshrimps (c)


Aesthetics aside it's performance were looking to Crucial to deliver. There are just a few maker's of DDR2 IC's fortunately for Crucial they're affiliation with Micron given them an advantage. I chose not to pull apart these modules due to their specialized construction (LEDs). Crucial lists the following specifications for part number BL2KIT6464AL664.

* Part Number: BL2KIT6464AL664
* Module Size: 1GB kit (512MBx2)
* Package: Ballistix Tracer 240-pin DIMM (with LEDs)
* Feature: DDR2 PC2-5300
* Configuration: 64Meg x 64
* DIMM Type: UNBUFFERED
* Error Checking: NON-ECC
* Speed: DDR2-667
* Voltage: 1.9V
* Memory Timings: 4-4-4-10
< Specs: DDR2 PC2-5300 • 4-4-4-10 • UNBUFFERED • NON-ECC • DDR2-667 • 1.9V • 64Meg x 64 >


We'll be testing this memory on the Asus P5ND2-SI Deluxe motherboard based on the NVIDIA C19 C19 chipset. NVIDIA nForce chipsets have been widely embraced by the Enthusiast and Overclocking communities since the introduction of their original NFORCE chipset. It was NVIDIA whom gave us Twin Bank (Dual Channel) memory architecture and their innovations played an integral role in the evolution of "integrated chipsets." nForce chipsets put smiles on the faces of many PC-Enthusiasts and did pretty well for the semi-conductor world (memory) as well. Until recently only AMD devotees were privy to architectural features such as DASP (Dynamic Adaptive Speculative Preprocessor) ironically with the introductions of AMD's A64 that feature became unnecessary. If there were ever a processor that could benefit from features such as DASP it was Intel's. Then came the C19.

Madshrimps (c)


For years Intel Enthusiasts wondered what an NVIDIA chipset could do for performance, especially since the advent of DDR2 where Intel's propensity for low latency and high bus speeds were further enabled by in-house engineered chipsets. While PAT technology contributed much towards improving memory performance since it's inception the issue of CMD Rate went as far as to spawn "pseudo PAT" architecture found in the controversial Asus P4P800. Dr. Michael Schutte (whom I have learned more from then any other writer) summarized the issue in that article’s opening paragraph;

On the first day, Intel created Granite Bay. On the second day, Canterwood, featuring PAT saw the light of the day. On the third day, Springdale appeared. On the fourth day, ASUS came out with a little modification and chaos has reigned since...i

While it may seem as if I'm confusing the 875 vs. 865 chipset issue on the surface, this has had far reaching effects as it seems to be a recurring problem deriving from Intel CPU architecture in general. Something I've made reference to in numerous articles with anecdotes such as "you need a dual degree in EE and Speleology to fully explore the depth of Intel's pipelines." This leads me to my next topic which is: irregularities in the C19 chipset. The C19 offers many unique features at the top of the list is SLI, but for Overclockers in particular motherboards such as the P5ND2-SLI Dlx provides Command Rate adjustment from 2T to 1T. Something Intel DDR2 based chipsets purportedly lacked. Initially

Additionally the chipset provides more divisors then I've ever seen, problem is these don't seem to be accessible to the user. When setting memory speed at say 800MHz and CPU FSB at 200MHz (800Mhz QDR as the BIOS displays this) the results are abnormally low bandwidth. The FSB speed must be set at or above 1066FSB in order for the results to be congruent with memory speed. I was confounded with the C19 when I found utilities such as CPU-Z, SiSoftware Sandra and Lavalys Everest Home Edition were reading (or misreading) memory speeds at 1/2 the value set manually in BIOS?

Perhaps this is someway related to antipodal differences in CPU design between Intel and AMD the C19 being the first chipset ever designed foir Intel. What gets stuck in my craw is that no other hardware review site seems to have delved into this nor even mentioned it? At first I thought it was related in some way to new BIOS editions until I flashed back from Asus 0802 to 0708 and finally 0605. I've validated the anomaly repeating the result using multiple versions of C19 based motherboards, memory brands/types, BIOS versions, software versions and Intel CPUs. The photo below exemplifies one of three BIOS settings (formula) from which benchmarks were run in this review.

Madshrimps (c)


After successfully booting into Windows under the above settings, I ran Sandra. Once Sandra Memory Benchmark was completed I opened CPU-Z and WCPUCLK which I then crop together in a screenshot to verify memory speed, timings, and divisor ratio which resulted in the benchmark score. I've used this method for several years to ensure each benchmark is validated and recorded along with data from the utilities.

Madshrimps (c)


The discrepancy between the BIOS setting and CPU-Z and/or Sandra in particular can be resolved by multiplying the frequency x4 or x4 for the actual speed for CPU-Z. As you can see from this [H]ard Forum post I'm not alone. Next running Everest Home Edition memory READ benchmark once again the frequency (compared to those around it) is displayed once again at half the actual value, note I've also included CPU-Z "memory" screen to indicate pertinent data.

Madshrimps (c)


If the frequency discrepancy isn't proof enough simply looking at the scores in both benchmarks would indicate something is amiss. I struggled with this anomaly for sometime since Asus sent me their first P5ND2-SLi Deluxe, which eventually failed, most likely due to excessive overclocking and repeatedly clearing CMOS. This issue also afflicted the Gigabyte GA-8N-SLI Royal which were in the process of testing. I recently mentioned this anomaly in my last article (Corsair 5400UL review) and received an email from Franck Delattre the author of CPU-Z. He explained the anomaly as follows;

memory frequency is obtained from the FSB, and recent chipsets allow setting a ratio between both frequencies, for more flexibility. The C19 chipset allows configuring the memory frequency very precisely, with two coefficients, let's call them M and N, and the memory frequency is computed as following:

Memory = FSB x M / N

The frequency obtained with this formula is the real clock, I mean for DDR2 400 it will report 200MHz (the x2 being done with the DDR technology).
In most cases, this formula works on the C19, but in some cases, especially when the FSB is set to something above 266MHz, it does not work and returns half the real speed....


Hopefully the issue will be resolved since this is a new chipset there's bound to be "bugs." I would like to thank Franck Delattre of CPUID.com for taking he time to write me concerning this. By the way CPU-Z v1.31 is available for download from his site.
  • next