>The 9800GX2 has the same memory bus and amount as the 3870X2 ->2x512 bits so why wouldn't the two perform equally if that was>so critical?The 3870x2 has a memory bus of 2x256 bits, which is the same as the 9800GX2's. This is considered a 512 bit interface, as 256 for each GPU.Noel

Nick,Whilst I don't think my benchmark is the panacea of FSX benchmarking, it does remove the human element of subjectively measuring performance improvement which, in my experience, has been grossly in error from some of the posts I have seen in these forums (ie. statements like - I upgraded my video card from last gen to latest gen and now get double the FPS everywhere - yeah right!). Now, you seem to know what you are talking about and likely can tell the difference without some form of semi-scientific method, but the rest of us, me included, likely can't.Re memory settings, I don't have many options with my Corsair XMS TWIN2X2048-6400 memory, which defaults to 5-5-5-18 @ 400MHz. I run at 9 x 390MHz to give my 3.5GHz overclock, and the lowest I can go on memory timings is 5-4-4-10 at 1:1. I haven't benchmarked the difference yet, and will do when I get a bit more time. Is this about the best I can set my memory to?Gary

All I was trying to determine is whether the 3870X2, the top 8800 cards and 9800 GX2 cards are reasonably equal in FSX performance. It seems we have to be guarded with any independent reviews so I was just trying to suggest an alternative.I still believe that for a given processor speed and equal option settings in FSX, AA and AF any differences in fps would be down to the graphics card.I think we'll just have to agree to disagree on that.

I should also have included the same OS of course!

What a thread - just to chime in using my own experience:I had dual 7900GTXs in SLI, and noted that with FSX, I would get a decrease in FPS in any mode but using SLI FSAA (where one of the cards does an AA pass), and then only at high resolution (1600x1200 and above). I also noted that having SLI didn't help much in any game I have except for a few titles that actually could multi-thread the rendering pipeline (something hard to do in windows, as rendering threads are very much controlled by DX layer - and that means, 1 rendering thread per window unless I'm wrong, at least in DX9 I think this is correct). The driver level seems to make the biggest difference but I've concluded SLI is more about making money than realizing effective benefits to the average user. Sure, there are a few outlying examples that can dispute this (see the quad AMD box and Quake), but they are very rare examples. Meanwhile, I'm back to a single GPU (albeit a fast one), as it seems the better way to go than two slower ones together, or even to fast ones for that matter.On the subject of reviews, it seems that most video reviews are skewed one way or the other, not necessarily meant that way. For starters, the GPU benchmarks are nearly all but useless as the drivers do one thing right, they are optimized for all the benchmarks (3DMark, etc...).Then, FSX is almost never on the benchmark, probably because not only it's a hardware hog, but even the best of the cards out have a tough time with FPS (the end-all measure don't you know). Throw in 7,000USD (that's what, 200 euros these days? :) ) and you get a whopping 20FPS at KSEA, less if you pack it with AI. Not a shining number exactly. Yet, you see nothing but Crysis benchmarks out there, and I can only wonder why that is, outside of the fact it's probably "the" king hog at this point, that does show a different number when you pop a new board in (or a different driver).I maintain that for FSX in particular, the benchmark is especially the sum of all components, not just one board in the system. It's just hard to make comparisons with the plethora of hardware and driver combinations out there, and any key component can make such a difference in the actual experience. FSX is just way too "blah" for a review, especially with the cutthroat competition out there as to who gets the "First" hardware review out, to the point we're now reviewing engineering samples, not even the actual product...Cheers,Etienne

GaryI think perhaps the way I explained benchmark results in reference to FSX may have placed a thought out there your work is not valuable. That is not what I meant. Anything that establishes a test pattern, if used correctly, can produce results that may be analyzed. Key words,

Nick,Thanks for going to the trouble of putting together such a detailed response - an outstanding effort!I am still having a hard time understanding why RAM performance cannot be benchmarked in FSX. In my mind (and I do have a computer/electronics background), each component can influence performance both indivdually and from an overall system perspective. As you state, at the system level there are sweet spots for system performance that minimise wait states between various components, primarily between RAM and CPU, but also on the other busses. Surely, if you tune a system to a sweet spot as you outline above, the software will run more efficiently. If frame rate is not capped in FSX, there is a set amount of activity that the executive must accomplish each frame. So if a frame can be accomplished more efficiently then more frames can be processed per second. ie. FPS is a valid measure of improvements acheived by better balancing a system. You can certainly see this for CPU speed increases, so I don't see why RAM is any different.Having said that, there seems to be very few attempts out there to equate better RAM performance to anything practical. The only reviews I have seen are on AnandTech, here http://www.anandtech.com/memory/showdoc.aspx?i=2900&p=6 and here http://www.anandtech.com/memory/showdoc.aspx?i=2989&p=7, where synthetic memory benchmarks show a notable improvement but real game benchmarks show neglible gain (< 3%)from having super fast RAM versus el dodgy RAM that most of us buy. And this is in more "simple" FPS games that don't have quite the lookahead texture loading job that FSX poses to a system.Granted, they aren't doing the memory balance trick that the AnandTech links you provide above, but I can't find any performance assessment out there of what this balancing act actually achieve. If you have any links that you can provide in this regard it will be much appreciated.Sorry, I am not trying to be argumentative for the sake of it, rather the engineer in me resists the thought that something that can be so quantitatively designed and deterministically function such as RAM can't be quantitatively measured. Video cards are another matter, as when all the engineering is said and done, it's got to look pretty to the human hanging on to the output side of the monitor! If that all doesn't make sense, then I guess we'll just have to agree to disagree :-)Re my RAM options, again thanks for everything that you have provided here. I realise that there are better options out there than the RAM I currently have, but I am not planning on changing it anytime soon so I was more looking at what I can do with what I have. As previously advised, my options are limited and, to espouse further:1. I am at my max multipler already, so 390MHz is the max bus speed I can go without dropping CPU speed.2. My memory config is 2x1GB and 2x512M (ie. full slots), which I know is not good and it seems my FSB is limited to about 420MHz as a result.3. My CAS latency is pretty well stuck a 5, as setting it to 4 results in a system lock anything over about 290MHz, even with extra voltage.4. Anything less than 1:1 memory to FSB ratio pushes my memory speed above its 420MHz fail point if I want to keep my CPU at max speed.So I guess my question really comes down to: is 5-4-4-10 better than the default 5-5-5-18 @ 390MHZ 1:1 or is there a better setting (sweet spot) that may give me a lower CPU overclock but a better overall system performance (that preferably I can measure with something, if not FSX ;-)).Gary

GaryYou sort of answered you own questionsI don

@Ray.....This is not the first time I've seen a conversation between someone who has studied the internal workings of a system, and someone who hasn't.Its two different languages. Those of us without knowledge of the science can only perceive the system from the external evidence; its easy to become frustrated if we are told the external evidence doesn't produce useful data.And so, its tempting to say, heck...lets reject the intenal analysis...cuz unless I reject it, I'm left with only two choices...and I don't like either choice.1. study the internals myself so I can overcome the absense of valid external reports or 2. do without the learning I seek.While I sympathize, Ray...Nick's analysis is quite compelling, and it would say you have to choose either choice 1 or 2. Rejecting either and claiming knowledge based on simpler concepts appears not to be a valid choice.I've been musing on other interesting element that appears to me in this thread; that is the general desire for validation of our choices after we purchase a fairly expensive toy. Haven't we all looked at data AFTER our purchase just to re-assure ourselves that we didn't screw up? I have...whether its a car or a video card or whatever...we all want to believe that we make wise choices. Many folks can only afford to make one purchase...so we don't have the opportunity to buy all the available cards in a given segment and do valid testing on our own machine, and its easy to find yourself wanting to have the forums validate your choice for you by having everyone acknowledge your purchase optimized your experience.You can remove this need, but you do it all in your head. You need to just enjoy your purchase...and I'm sure its awesome. Whether it optimizes your experience is irrelevant. Its a sunk cost, you're having fun, and I'll bet your fsx is super.This is really a variant of choice 2 above.Best,Bob

And just a quick follow up...I had one person I went through and set up everything (BIOS+WINDOWS+FSX)... Everything worked PERFECT (he could not believe the difference) except for texture blurs and annoying stutters in turns.The frames were there, the fluid flight was there (with the exception of turns), the scenery was all there in all its glory but the textures were blurry. We tired bufferpools and every other tweak I could think of.. then finally:I had him recheck the memory timing.. He mistakenly swapped tRAS for tRFC. (24 for 15) As soon as he placed tRAS where I specified the ground textures came in razor sharp and ALL the stutters ceased. No one can tell me memory speed and timing has no effect on FSX. I will and it does. And let me upgrade my comment about Aces getting blamed.. The issues around all this DOES go back to Aces and how they designed the sim from FS2000, up. My plea to Aces is they start over and this time make sure that all systems can meet baseline rendering without having to jump through hoops to figure out what

>This is where DDR3 begins to shine into the holy grail area.>Purchasing CAS 8 sticks @ DDR3 2000 and running them at CAS 6>at DDR3 1600-1800 is by far wiping DDR2 off the map for real>world performance. Not only is the data better optimized with>DDR3 (with no clock), the higher memory speed is being used>instead of wasted by wait states. As time goes on and the CAS>rating is achieved with higher speed memory product, DDR2 is>dead.Nick,Sounds like you understand the subject matter. Tell me though, how does this translate into better performance, using your definition of what you experience in terms of smoothness, clarity, etc?Here's my dilemma: Sam suggested I need only DDR2 memory, because I would be using the pricey QX9650, so did not need to maximize the FSB since the FSB is rarely if ever the bottleneck. This last part I'm not sure is true. Do you know for a fact what sorts of bottleneck potential my processor has with the bandwidth limits of an FSB of 1600, or even 1333? I have run a few benchmarks (of varying performance capabilities) and I see the 1333 actually outperforming 1600. This was of course not strictly a memory benchmark. I am using pretty high latency DDR2-800. I am not using FSX, I just know that FS9 performs maximally loaded with PMDG and tons of addons with all sliders to the right, 1600x1200x16xAF4xAA, and the sim is completely smooth at all times, maintaining the frame counter at whatever I set it up to 70FPS, sustained, in dense terminals with 60% traffic. Anyway, I see absolutely no difference in anything with either FSB setting, but I DO see the predicted differences at different CPU clockspeeds, at least in fps. It's always very smooth.Anyway, what do we know about real world mismatch of FSB vs CPU where the FSB cannot handle what the CPU is sending or grabbing due to lack of bandwidth. Surely these values have been quantified so as to supply a high degree of credibility that optimizing memory performance makes any difference whatsoever on FSX. From the Anand quote:"We were really surprised at the gaming test results. We really did not expect the bandwidth improvement of P35 to have much impact on gaming results, but Far Cry showed a 2% to 5% improvement in performance just comparing P35 to P965 under the same conditions. It really didn't matter whether P35 was running DDR2 or DDR3; the improvement was essentially the same."OK, now herein lies an issue worth noting. "2 to 5% improvement" really is approaching insignficant. 5% of 30 fps is 1.5fps. This affects nothing of signficance. And 5% was the upper end of this range. Tell us more Nick, sounds like you know the subject matter well, and thanks in advance. I know my stock memory timings on this very low cost DDR2 is 5-5-5-18, if I recall. Do you recommend I try to improve the timings? My P5E is just using its auto configure option in the BIOS. I have just changed the FSB and multiplier to get the clockspeed I'm seeking.3DMark06 runs over 15,000, I forget how much over. Noel

Nick,>As for FPS and FSX and benchmarks, that

I was only using Seqsquashtoo as an example and he is not the only one who has witnessed this behavior in the sim. There are people who are pilots here who are not prone to excitement or over reaction who see the same results. And that behavior is something that can be found all the way back to FS2000There are quite a few things we are told never to do..Do not raise TBM, do not use unlimited frame rate with high TBM, etc, etc, etcThere is difference between the science/theory and the end result based on the hardware the application is run on. And as Phil has stated they could not test their product on the possible hardware combinations and settings.. therefore -they don

NoelI wish I had time to dive into this and actually run you though a setup of your system, but unfortunately I have a delivery of textures I must turn my attention to for GEXn Europe and my time is limited.First of all I do wish you and everyone else would lose the word FRAME RATE. In games that number has value, in FSX/FS9 that number has limited information tied to it.If you already getting the performance you are from FS9 it appears to me you are set up fine for the application of choice.3DMark can never be used as a compare against any game for performance. Just as the frame rate in MSFS presents limited information, the score in 3DMark has little real world bearing on games.When I say a 20-30% improvement, I am not talking about frames I am speaking of the total estimated resource increase to the title of choice.I already posted above to Gary what AnTech said about memory speed, low latency and game response. Auto configuration in a BIOS may not be bad... Auto settings and how well they play depend on the BIOS programmer and the memory in use. I have seen several systems over the years in which the BIOS was configured by a pro such as Victor Wang, whereby the {AUTO} settings would have the correct math behind them in conjunction with the motherboard components to produce excellent results. That is typically not the case but it does happen.This is the chart you should strive for in CAS/DDR Speed (first image)http://www.thetechrepository.com/showthread.php?t=195The GREEN lines represent where the CAS value has the greatest impact at DDR. You should strive for DDR speed within the GREEN range posted for the CAS value. It can require a Vdimm increase if the memory product does not meet the spec of that chart. That increase may or may not be dangerous based on the engineering of the modules and must be researched/tested if those modules are not designed for it.The higher you can go @ CAS with DDR the better the result.. just because CAS 6 is listed for >1333, if you can do CAS 5 at 1333, which is the goal of faster DDR3 modules as time progresses, then you are moving the presented performance scale, up.Once that is achieved the next step is to raise FSB to where the chipset/strap have the greatest impact the CPU, and, in that process raise the CPU speed to the highest stable value it will maintain @ CAS/MEMSPEED. With DDR2 this is typically 400Mhz as a starting point. Older systems may not be able to achieve that... and actually there is a way to trick the old 965 systems into doing it by increasing the Northbridge voltage and dropping the strap to 800... but that is another story LOL I could go on and wish I had the time.Now, newer motherboards are making available the MCH Voltage which up till recently was not available without modifying the motherboard.. it allows another voltage tweak to the modules and like Vdimm requires the user understand the limits.The best clocks on high memory speeds are achieved by working the MCH voltage in reference to CAS however it is not necessary and the user can find the best combination possible by following some basic rules and charts which I posted above in links to AnTech.. here is a chart layout which shows what they have tested as 'all around' best for DDR2, and I agree...http://www.anandtech.com/mb/showdoc.aspx?i=3208&p=9However those memory speed numbers are just a start in my book and they can be improved, both FSB and memspeed. The majority of the success comes in the initial purchases.. motherboard/BIOS and memory.I do not see how you are going to evaluate a performance difference with FS9 running as it is now but you are the best judge of how your system is functioning over anyone else. FSX would be a much better test of the result however seeing an accurate result also means FSX and the system are set up to deliver without other factors tainting which is why I would like to run the tests myself. As metioned to Gary above, as soon as I can I will present some test results.. I will try for this weekend. :)

Gary, I do have a system I use here for playing and as I mentioned to Ted I do want to present some numbers for everyone to chew... my problem right now is time.What I will do is use your benchmark in those tests and relay the result information along with the stats. Since the test presents results which you feel will relay information perhaps since my systems are fine tuned in every respect the results presented with my observations between default BIOS, increasing CPU alone and then apply mem clock techniques will provide data which demonstrates the use of high memory speed as a source of FSX improvement. As soon as I can, I will do that! :) CPU speed is definitely the focal point in performance.. no argument there. Properly optimizing the flow of data around that speed is what makes high memory bandwidth valuable to the title, or any other application in use.