I for one, get the following scores using AIDA64 Extreme Edition, version 2.85
@ 288 FSB
Memory Read : 7652 MB/s
Memory Write : 6001 MB/s
Memory Copy : 5562 MB/s
Latency : 68.6 ns
CPU Queen : 11173
@ 288 FSB with 3-3-3-9, 2T timings
Memory Read : 7788 MB/s
Memory Write : 6034 MB/s
Memory Copy : 5685 MB/s
Latency : 66.7 ns
Compared to yours (please, check your results again with v2.85), I get better memory write and copy values (1 GB banks VS 2GB's, I think), but lower Read and much lower Latency scores. As a side note (and as expected), the CLKBUF's have a small negative impact on latency.
Maybe I'll get a couple of those HyperX sticks
Tempting, I agree! These sticks sure shine when operated under 500 MHz (beyond that point, I would recommend the G.Skill's). Also, make sure the shop or reseller will be willing to offer you a refund in case your system refuses to take 1T CR. My guess is that Asrock don't build their boards as consistently as the bigger manufacturers do, hence results may vary with similar components.
For example, I can imagine many 4Core users don't have to alter CLKBUF's from Advanced Memory to obtain a stable mild overclocking. As for me, I need to have these cranked up somehow (starting @150-300 ps)... Having in mind my value DDR2 sticks are pretty standard (exact same SPD values as Corsair's).
Let's talk briefly about Memory voltage as set to High.
Quoting
controsensi.it, Vddr voltage measured at VT3 point, without Mod, is 1.86v with "DRAM Voltage" set to Low; 1.96v with "Auto" and "Normal"; 2.06v with High. Usually, EPP (Enhanced Performance Profiles) benefit from 2.1v (on average), which I believe most sticks can easily take. At worst, yours wouldn't need the extra juice. Note that the guys at tt-hwardware.com did run them successfully @390 MHz with 2.1v, with the same 3-3-3-8 ultra tight timings.
('et plus' means 'and more')
One thing I noticed, in your bios settings, you sync the PCIE Clock with CPU (at a 295 FSB). Any reason why? In my case, I also sync PCIE with the CPU @ in my case a FSB of 288. I find that if I lock it to 100MHz, my computer will freeze (vpu error).
I am running my system at...
304 FSB right now (guess what, playing with the CLKBUF's again, currently set @300/300/450/450 ps)!
@ 304 FSB
Memory Read : 8070 MB/s
Memory Write : 6335 MB/s
Memory Copy : 5908 MB/s
Latency : 65 ns
I'd take the AIDA64 results with a pinch of salt though : "Memory Latency benchmark test uses only the basic x86 instructions and utilizes only one processor core and one thread". Apparently, AIDA64 team are working on a new version of this benchmark.
Well, the answer to your question resides within it : If I don't do the PCIE / CPU sync thing, the overclocking causes the desktop to fail loading itself properly (garbled display). I stumbled upon this setting while reading some posts here and there, don't have an explanation to backup that choice of ours (other than it may have something to do with Asrock's implementation of the PCI-E, taking into account the hybrid nature of this range of mobos). I think the VSTA boards corresponding setting can be incremented by 1 Hz.
What puzzles me further is the lack of a few secondary timings readings via CPUZ or AIDA64.
As you may notice, AIDA64 chipset screen doesn't show tWTP, tPTP,tRD.
Oddly enough, these "misc" values are hidden from us.
Other than the presumably optimal BIOS settings, a rather tiny GPU, and PWM enhanced cooling, there could be a reason why I am getting 300+ FSB when others cannot :
My main HDD is neither an IDE one, nor a SATA one. (it's a Seagate SCSI Ultra 160 one and as such, it bears its own controller).
Inversely, this could be the reason why Asrock engineers provided us with the intriguing IDE Drive Strengh "feature". SCSI drives have mostly disappeared from the market, but latest SSD drives may offer a similar level of decorrelation. Which brings me to my advice towards Hlafordlaes, if I may : First, get that brand new SSD drive you evocked earlier on, then the performance memory sticks.
doubleOseven said:
I just ran the Lightsmark2008 program.
Another GPU intensive benchmark comes to mind :
Unigine's Heaven benchmark, featuring hardware tessellation.
It is DirectX 11 based, so needless to see, I cannot run it on this machine.
Also, it's not free but the free version will do the job.
I am glad I passed the 300 FSB limit, but GPU upgrade is still the priority.
Thanks to nightspydk for bringing back the subject of timings.