I've decided to simplify my reviews, as I'd discovered not everyone is interested in Reductionism methodology of hardware testing and the existentialist approach to technical writing. In other words, this review will be simple and to the point.
Test Platform Specifications:
I've found many of OCZ's memory kits seem to perform quite well on the Canterwood and Springdale (875/865 respectively) chipsets. Originally deriving from the E7205 Granite-Bay chipset, which housed the world's first Dual DDR MCH (Memory Controller Hub), the Canterwood is based on a 1005-BGA package, supporting Dual DDR400, AGP 8X, and CSA (Communication Streaming Architecture) theoretically capable of 6.4GB/s memory bandwidth, AGP through-put of 2.1GB/s, all from the North Bridge. Intel also integrated a feature known as PAT (Performance Acceleration Technology) into the 875 which shaves two clock cycle's off the memory bus, allowing for tighter latencies.
My experience with PAT, is that it does in fact boost performance, and by a substantial sum over its step sister, Springdale (865). PAT and its illegitimate counterpart GAT (on Abit) offers end-user's a plethora of BIOS features, many of which in the latter are simply esoteric memory adjustments.
Where BIOS adjustments are concerned, most Enthusiasts prefer Manually setting latencies in an “Advanced Chipset Features” section, and with GAT you can even have Auto-Racing connotations. If it is pure speed you're looking for, ironically I've found SPD works best with the 875/865 (and 915) chipset's. Why is this relevant?
Just about the time Winbond BH-5 began drying up, Samsung and a few other IC makers began producing high speed memory, which was SPD malleable. What does this mean? Well basically if you take Corsair's 3200XLPRO or OCZ's Platinum EL3200, both will run at 2-2-2-5 at 200MHz. Then simply leave the Advanced Chipset (memory) under the Auto or SPD (Serial Presence Detect) command and raise the FSB to 250Mhz. The memory will self adjust from 2-2-2-5 to 2-4-4-8. Early revisions of these IC's weren't able to run at both latencies. You could run PC4000 at PC3200 speed, but the best you'd get is 2.5-3-3-6, and at speed (DDR500) 3-4-4-8. So today we basically have IC's from both of the above categories. The GOLD 3700 will most likely run up to PC4000 at 2.5 4-4-8 and the Platinum 3200EL, will most likely attain PC4000 ad higher at 2-4-4-8 of course at lower speeds holding 2-2-2-5. Both are versatile products and reflective of the best IC's optimized with the best PCB's.
It's all we have at the moment, until something better comes along. Memory manufacturers have tried to downplay the importance of latency at speed, and this is in part because of the insanely high latencies of DDR2. Of the kits I have seen from different makers the numbers made my head spin: CL4-4-4-12 and even CAS-5-4-4-15 with DDR2.
Where OCZ's Platinum is concerned, when seeking the greatest speed, while maintaining tightest timings, the memory should hold 2-3-3-6 up to, and possibly beyond, 230FSB (DDR460) Raising the VDIMM will not have the same effect it did with Winbond BH-5 equipped modules, for example. There's a voltage ceiling I believe to be around 2.9V, any further and your actually degrading performance. With BH-5 now a memory (pun intended) OCZ as well as other companies, must rely upon Samsung Fab's, among a few others to find that Holy Grail of 2-2-2-5 DDR600 memory.
On to the test results:
Bandwidth measurements do not tell the whole story. The tighter latencies at which OCZ's Platinum Series run will deliver better performance especially at lower speeds. For those of you whom aren't concerned with prima facie overclocking, and understand even at higher speeds, that timings as loose as CL2-4-4-8 are relatively ineffectual for many applications. Next we'll take a look at some PiFast numbers.
Compared to the majority of PC4000 on the market which runs at CL3-4-4-8, when OCZ's Platinum 3200EL Rev.2 is overclocked to DDR500 speeds there seems to be a deficit. The memory does very well, however; when just a little bit of VDIMM is thrown at it, and timings are set Manually to 2-3-3-6 @ 230FSB. This produced the best results overall, and would be my recommended setting. In so far as the GOLD 3700 Rev.3, it loves speed, albeit at loose timings.
I've always found OCZ memory to be some of the best overclocking memory out there, and for overclockability with tight latency, their Platinum series does not disappoint. The GOLD series Rev.3 3700 definitely has been improved, running faster at lower voltages, and reaching better overclock speeds in general. My system was limited as my 3.0C wouldn't allow me to get far past 3.75Ghz, even though I have had it up to 265FSB or 3.9GHz. I believe this may have been due to my Sapphire X800 Pro acting up, which completely bugged out after the review. In Part-II you'll see Jnav89GT has some better luck with overclocking, and takes both kits a little further. Regardless of my handicap, both kits surpassed their default settings, and by a substantial margin.
Anytime you purchase a product, and that product surpasses the manufacturer's spec's, you're ahead of the game. I believe as Overclockers we are perhaps, the most demanding of PC consumers. Imagine if every CPU, memory module, and graphics card purchased would run only at its default speed, and not a MHz more? How many of us would be standing in our front lawns shaking our fists at the Silicon Gods, kicking sand on every beach. I don't know about you, but I can't go through life being angry at what fundamentally reduces to sand. But that's Reductionism, and as I stated earlier I promised to stay off that topic. OCZ Technology has given us a few more products, or rather revisions of products, which reflect their philosophy. That is; squeeze every MHz from the silicon, and give the PC-user the best you can for his/her money. That's a good thing in my ledger, and I for one am very happy with the memory tested here. I'd like to thank OCZ Technology
for the samples.
Onto the second part of our review ->