Ram choices

[toothjockey 3]

Greetings,I'm in the information gathering stage of putting together a new system to be used for FSX with Vista 64.I plan on 4GB (2x2GB) DDR2 memory. Is there much to be gained by going to 1066 instead of 800? Looking at the Corsair Dominator. It seems to me that it would be hard to recover enough benefit from the huge increase in the price of this ram. Am I wrong?Thanks,Don

[mjrhealth 820]

FSX benifits from every banwidth increase, it helps get those textures loaded. So yes you should see a benifit even if only 4 or 5 FPS, but it should help reduce stuttering too.

[Guest Jeffhew]

Hi Don,My recommendations are in my specs below. My system's about 5 mos old now but, should be a good starting point. You really don't need the Dominators to get good performance. I'm not on Vista though, so there could be other concerns. Just as a benchmark, if I push the memory to 1140MHz, FS9 won't run for more than a minute or two before a black screen and a CTD. At 1130MHz, she screams like a raped ape. I know you're looking at FSX but I thought I'd throw that out there.Best Regards,Jeff

[Guest D17S]

Sadly, you are not wrong. A builder will not observe Any difference, in Any way between ram running at 800, or 1066 or even 1600. Save your dough. 1) FPS is a function of the clock speed of a single core. 2) Texture loading is a function of how many core are operating at that speed. 3) Ram's contribution is a function of quantity. 4Gs will help performance over 2Gs. However speed will not matter here. Even running ram at 533 will generate no performance deficit. Build with DDR2-800 ram. The cheapest DDR2-800 ram with a lifetime warranty will provide the maximum performance available. (You Are going to O/C that CPU, right?)

[toothjockey 3]

Thanks for your response guys.It appears though that there isn't a clear consensus on the answer. :)Sam, what are your qualifications (or background)? You seem to really know your stuff. I've seen many of your posts. Is your background in FS or in computing in general?Thanks guys,TJ

[Guest D17S]

We went on for weeks about this. I finally had to grudgingly admit the subjective assessment of improvement was at least honestly presented. It would be like filling one's everyday vehicle with Premium after generally running midgrade. The everyday driver could tell a difference, but it would take scientific instrumentation to prove it to a non-everyday driver in that same vehicle. It was that close. And it was extremely expensive, AnD it took a real background in computer science to make even that minor improvement occur. The technical metiphore that was finally presented described two sets of circular tracks formed as a figure 8, with the tracks only in proximity at that single touch point. Freight trains traveled on these tracks. These track/train sets represent the memory buss and the Front side buss. The goal was to transfer cargo (a data bit) from one to the other at that touchpoint. Both trains traveled at different speeds. The transfer rate more depended on synchronizing those rail car's door openings so a transfer could occur, rather that simply speeding up the trains. Increasing the FSB or memory speed did NoT necessarily help performance. It became a delicate dance where timing the rail cars trumped their absolute speed. As it turned out, it took Massive increases ("brute force'd") memory buss speed increases to make entirely minor improvements in transfer rates (well, duah!). It was about sync, NoT speed. As it turned, out the memory controller only even aTTempts a transfer every 12th car! That's Intel being Very conservative. That function is called rRD. This is a northbridge function and It Can Do Better! If a tweaker wants to play this game, tRD will still fully function at a tRD of about 6. This is called "Overclocking the Northbridge" - NoT the FSB. But even with that, the improvement remains subjective at best. The real debate boiled down to: "Is an only subjectively determinable improvement worth thousands of dollars of hardware and/or hundreds of hours of study?" My view was no. The other side said yes. That's where the consensus broke down. BTW, improving this memory <> FSB data transfer function is what Nehalem is all about. With our current-gen stuff maxed out, we were able to get close to Nehalem's initially advertised memory latency numbers. It didn't matter a whit with our rigs, and I don't imagine it will with Nehalem either. We'll see.My work is with real airplanes with lots of systems, which led to an interest in FS, which drove to an unwilling addiction to optimizing it's platform. Fortunately, it's a very happy set of interconnections.