ASUS Striker II Formula meets the QX9650

ASUS recently sent us their ASUS Striker II Formula based on NVIDIA's 780i chipset. Raja will have a detailed look at the board along with his usual extreme overclocking results and our first NVIDIA BIOS guide in the near future. In the meantime, I have been using the board for upcoming articles on SLI/PCI Express 2.0 video performance and memory tuning for Vista 64.

One of the questions we have been asked repeatedly is what kind of FSB clocking rates can the new 780i chipset hit with the Yorkfield and Wolfdale processors. We will have Wolfdale results this weekend as our retail processors finally arrive tomorrow but for now we have a couple of interesting QX9650 results. We hooked up our Boreas cooler from CoolIT Systems, installed 8GB of OCZ's new PC2-6400 Reaper CL4, inserted two MSI 8800GTS-512 cards in SLI, and then decided to find the sweet spot when clocking this board under Vista 64 Ultimate. Although our QX9650 was happy running at a final (extremely stable) 468FSB rate, the performance was not up to par with ASUS's excellent X48 based Rampage Formula. The main reason being we had to really loosen up memory timings (5-5-4-12) with the 0901 BIOS to run in Sync mode at DDR2-936 plus we hit a strap setting that was hindering performance, basically a double whammy to system performance.

Not being ones that are really impressed with high FSB rates as they typically induce high memory controller latencies with reduced throughput, we decided to figure out the best performance on this board with our component selections. We did not have to go to far down the FSB ladder as 450FSB (1800 QDR) turned out to be the sweet spot with the QX9650 and our 8GB memory selection.

ASUS Striker II Formula - QX9650 at 10x450

ASUS Striker II Formula - QX9650 at 9x450

In our first screenshot, we able to hit 4.5GHz at a 10x450FSB setting that maxed out our CPU ceiling without going nuts with VCore. Our QX9650 is the runt of the litter and requires anywhere from 1.4250V to 1.4500V real voltage to hit 4.5GHz and remain stable. The problem is that running these voltages on the 45nm processors usually result in throttling conditions unless the CPU is kept chilled, even then, we have still noticed core 3 throttling or dropping out completely under heavy loads around 4.55GHz with our sample.

We had to set VCore to 1.525V in the BIOS to hit our 1.44V target range for stability on this board. Temperatures were not a problem with the Boreas cooler as idle temps were around 10C with load temps hitting 19C~23C. Our memory was run in Sync mode at DDR2-900 with 4-4-3-12 settings and tRD set to 6, resulting in very good memory throughput results and latency at 45.6ns, which is just slightly slower than our OCZ Flex9600 2GB results at 44.7ns in off-line testing.

We were impressed with these settings as they equaled or were just a couple of percent off our upcoming X48 results. However, knowing that tRD settings on this board can be improved at this FSB setting, we decided to fine tune the memory and tRD options to see if we could reduce latencies even further while keeping throughput at the same levels. We just could not reduce our tRD settings below six at 10x450. We have noticed the QX9650/780i or QX9650/X48 combinations behave a little differently when the CPU exceeds 4GHz as the stress on the memory controllers creates several problems when trying to setup the board with low tRD values, high FSB rates, and tight memory timings. Kris will be detailing out tRD (read delay) and how to properly setup the MCH in an article tomorrow that will explain in detail why FSB rates should not be used to justify if a board is good or not.

In the meantime, by reducing our processor multiplier to 9 while keeping a 450FSB for a final 4.05GHz setting (VCore dropped to 1.50V with 1.425V real), we were able to reduce tRD to 4 by slightly increasing our memory timings to 4-4-4-15. This also required us to increase tRC to 26 and tRFC to 48 for stability. Our end results are remarkable for an 8GB memory configuration with latencies dropping to 37.9ns while read rates dropped a percent but write and copy speeds were improved even though we had a 500MHz disparity in CPU speeds. Once again, we see that by adjusting tRD to improve chipset latencies and throughput that our memory performance can be affected dramatically at times.

What does this all mean in the real world, generally you can expect an improvement in memory sensitive applications up to a few percent (3%~7% on average). These improvements are usually important for benchmark results, but the bottom line is that you can properly tune a system for "best" performance without having to run excessive FSB/voltage rates, memory speeds, or spending additional money on low latency memory. We will have application performance benchmarks available in the mainboard review, but for now, we can equal the performance of the X48 chipset in most situations with the 780i when using a 45nm processor.

Does this mean the 780i is a better chipset, not really in our opinion, but if you have a need for SLI you will not be at a disadvantage when it comes to using a 45nm processor on a NVIDIA platform now. On a side note, our next blog update will show how well a $95 650i based board from ASRock competes with the $300 780i boards with the QX9650 and it also runs the QX9770 just fine, something we are having problems with the 780i boards doing correctly right now. However, to be fair, ASUS did provide a new BIOS this week for the Striker II board that has improved QX9770 stability greatly and NVIDIA did not qualify this chipset for the 1600FSB processors. In the meantime, ASUS did a wonderful job in updating the Striker series, we just need more time with it before coming to a final conclusion.