Very interesting, keep up the good work !!!Happy landings...

Thanks for reporting back Noel. Looks like the higher speed and more mb capacity is better than the low calwi so far. It'll be interesting to see the results when both are running with 4 gb.Have you tried seeing what kind of speed you could get out of the OCZs if you relaxed the timings to match the Mushkin?Ted

Not yet Ted. But I will try as many permutations as possible and come up with some sort of conclusion on what role CALWI optimization has primarily. I think we can all agree, even the least technical of us, that highest bandwidth and lowest latency delivers the best promise of best performance as defined here and there. However, what role these variables have versus raw CPU & GPU horsepower, is in my mind, still up for debate, and truly hasn't been answered well. As I've pointed out before, Nick's narrative description of his various testbed performance perfectly fits my machine's, even with my mushkin at 4-4-4-10/800mHz. The 5-5-5-12/1103 seems better, but they are both very good. The other issue that some have pointed out is that it is also about caches, and other engineering features (I don't understand well) processors such as Penryn offer versus AMD's built in controller in what that brings to the table. Sure is a fun hobby tho!QX9650 w/ Retail HSF|8GB Muskin PC-6400|ASUS P5E|EVGA 8800GT @700|Seagate SATA 2 x 4|Seagate Cheetah 15K.x|XP Pro|Vista 64--soon to be installed

Here's one permutation:http://forums.avsim.net/user_files/188305.jpg1060Mhz @ 4-5-5-15 (1066 not stable)5:6 dividerCALWI value: 7.5nsEverest latency: 5.28ns - 53.3nsEverest read: 9230Pretty dang good!Now here's the interesting part . . .Ok, i'm in the hallowed CALWI zone now. Bandwidth has increased by roughly 30% over mushkin at 800Mhz.I think it's maybe . . . 10-15% better in terms of texture load update rate, translating to better texture clarity/detail. I think if you did a double blind test on me of what I see at this superior memory latency and memory "performance", it would be hard to say if I could identify which mem config was in place. Seriously, I think my guess of very good x 1.15 = very good +. It really a little academic, esp in the context of all other factors affecting performance & user satisfaction. Both work well. I think, yep, not sure yet, the OCZ CALWI'd and respectably bumped in bandwidth IS better, but truly, it's not a lot to note really. I think CPU power truly is the big player. But, that's just my opinion, based on limited knowledge. Setting up the sim properly is big. The memory issue I will have to downgrade to relevant, but only to a degree.I'll keep trying out different scenarios n see if I can get a more accurate sense. 10-15% "better" is also, signficant. And for what I paid for these modules, it wasn't a horrid spend. I'll probably keep these. These modules, so far, do very well, esp for the $$.CR of 1T doesn't fly with all four DIMMS in place.Noel

You can also see now the Everest latency even though you have entered the CALWI range remains about the same as you saw in the past, meaning, use of that as a reference in performance is superficial at best. You won

>You won

I edited my post.. I misread your memspeed originallyYou are in CALWI @ 7.6 but you are not in the high performance zone..If you were unstable @ 5:6 it is not a valid test result because the instability can be presenting lower performance.. this is not a race car on nitrous and if it hits 300MPH for 5 seconds, its in the zone and you saw what it can do.. this is electronics and if they are unstable you can not confirm a test result as valid.Noel, your focusing on one area.. to an extreme. There is nothing magic about CALWI unless the rest falls into place and even then the result is limited to the final factor, memspeed and tRDThe differences can only be seen if all factors are in place. Not 1 or 2 Noel, all of them..Based on what you have seen in increase you can extrapolate for if you had the sticks ruinning 1T, and, if you were able to increase memspeed another 150Mhz+You can now see the potentialI stated quite clearly in my presentation that I was able to expand DDR2 performance to a point where it was in range of DDR3 however to do so required voltage increases and a memspeed of over 1200MHz @ 1T. What you are doing at the bandwidths you are working with is not going to show you anything dramatic

>I edited my post.. I misread your memspeed originally>>You are in CALWI @ 7.6 but you are not in the high performance>zone..>>If you were unstable @ 5:6 it is not a valid test result>because the instability can be presenting lower performance..Actually, I think 1055Mhz @ CAS 4 gets the 7.6ns latency, and I was stable at that. So, the test was valid. Using stock vDIMM, I could not maintain 1066Mhz CAS 4, but did for many minutes. Well, if you are saying top performance in every relationship = best possible performance, I don't think you would get any argument from anyone there. My contention has been, and even more so now is, that there is nothing quantum about any of this optimization. The better the ENTIRE environment is, the better the performance. I think it's 100% a continuum.At 1055Mhz, CAS 4-5-5-15, 4.17Ghz, I just took a flight from KMRY to KSBP with ASX weather, and all the usual addons. The Cessna 441 Conquest is beefier than the default models, and the display settings were close to maxed out, as much as I prefer anyway. It does 250 knots as well. It was, 98-99% of perfection the whole way, complete with fog layers and amped up traffic from UTX. Smooth as silk, textures crisp, detailed and stable. Move that into a very dense terminal, it might drop to 92-95% of perfection. 92% of perfection, is still. . . very very good. In fact, the "untrained eye", might not notice the difference between 92% of perfection, and maybe the 98 or 99% of perfection in that setting one might get from the fully optimized environment. Heck, I'm sure there are other places in the sim I haven't even touched as far as optimization goes since I'm so new with it. Speaking about that, what are bufferpools, and what impact do different values have?My conclusion comes down to this: DDR2 or DDR3 connected the rest of my particular parts? It will matter, but not that much. Certainly no need to either purchase $600 worth of CAS 3 memory capable of doing 1000Mhz, or lose my X38 mainboard for a DDR3 board. It just doesn't add up.Memory latency & bandwidth optimization is just one piece of a bigger environment. It plays a role, but only after several other issues are addressed, including CPU clockcycles, GPU, file subsystem optimization, lean boot ups with unnecessary services disabled, optimal drivers, etc. Even the lowly mushkin at the lowly setting of CALWI 10ns, 800Mhz CAS 4, could deliver high 80's to low 90's% of perfection. And I'll say it again: let's crank up the PMDG 747 were it possible in FSX, put all sliders including traffic to 100%, and then fly at 500 knots at 3000 feet over Low Angeles & up the cost. How will the fully optimized DDR3 testbed perform? 100% of perfection? I kinda doubt it. If so, by God you're on to the Holy Grail. Put another way, the fully optimized DDR3 testbed gets you able to process a little more complexity with the same high quality performance than the moderately optimized DDR2 system, but in the end, it's very arbitrary, since I'm sure we could stress the DDR3 system to a fairly poor performance level, if not now, then with the next set of complex addons. As for showing things dramatic: 92-98% of perfection leaves very little room for dramatic . . . This OCZ at 1055 is really quite decent and is stable. I'll plop in the old mushkins n see if I can tell anything different. It was working well too at 1103Mhz. But I think the OCZ's do better. Haven't really tested this enough to know for sure. Too bad not 1T. Will higher vDIMM possibly bring 1T into the realm of possibility? I had it running at 1T while at 3-4-4 at 800Mhz, until it crashed. I do have a question you might know about: the OCZ I bought is rated 2.4V maximum to retain their warranty. As a practical matter, what is a typical level beyond that one could probably still not cause at least near-term damage, provided the DIMMS are kept very cool? My sense, with very little exp to back it up, would be: 2.4v will probably fly quite long term with these. Maybe 90% will travel 3 years or more, and by then, people won't exercise their warranty anyway? Add 10% overvolt, which would be .24v, we are now at 2.68v. I would guess this would stress things, but I kinda doubt it would be catastrophic, but as I say, I haven't a clue. I've always figured 10% over anything electrical *should* still be in the realm of expected fault tolerance, else it would simply be too fragile to be useful. So, as a practical matter, how much more over 2.4v should be safe (with no guarantees of course)?Thanks for the useful hints on adjusting and optimizing a specific memory subsystem. I learned enough to make some nice improvement, and really didn't have any clue prior to this discussion as far as getting the most out your memory, DDR2 or 3.This just in: Everest to 52.2ns/9480mb/s. Thought I'd take a chance on AI clock twister, after changing to 4-5-4-15. Doesn't want to go below 15. Pretty impressive DDR2, for $73/2Gb, and there's a rebate of $20 on one kit. Now, if vDIMM over 2.4v would be safe-ish, who knows, maybe another .05% improvement :()Cheers

I was really hoping you'd get the 744 to run. I spend 40 hours a week with those dumb things and it's the only airplane that has any significant interest to me. In the sim, I'm not looking out the window much, but it would be nice to have TCAS go off every once in a while. I have to run with AG/AI completely off. As far as sight seeing, Ground Environment was the Real performance break-through miracle. We sure need to say thanks for that. Thanks!I expect we are attempting to Manually blaze a trail that Intel is in the process of Automating with Nehalem, and then Sandy Bridge. We are being introduced to things to come. Given the $$s and the technical understanding, a portion of that future promise might be available now. So far though, I do not understand the technical properties well enough to make it happen even if I Did want to spend the dough . . . mainly because I've been lazy. I also have buckets of cents, but not so many dollars. The technical explanation exists. That's what's so great about technical things. It is black and white. Maybe I'll dig in at some point. However even for me this is just, well, too much effort for not enough reward. "Speed costs" and in this case, it's not just dollars.In the old days, it was a real challenge to O/C a CPU. I was amazed when my Q6600 just ramped to 3.6 with the mobo in Auto. I got the added performance and didn't have to become a rocket scientist to do it. Ohh Thank-You, somebody! I expect this memory performance will be one of the next frontiers. We've seen an aspect of the future, but the main performance increase we can expect will be Sandy Bridge with 32 (essentially Core2), hyper-threaded cores . . . in 2 years. This is a magic time for computer geeks.

>I was really hoping you'd get the 744 to run. Sam, which 744, PMDG's? I'm not all that fond of 747's so was going to wait til 737 NGX comes along . . .If you are using PMDG's 747, how does it fly Sam?>I expect we are attempting to Manually blaze a trail that>Intel is in the process of Automating with Nehalem, and then>Sandy Bridge. We are being introduced to things to come. Here's Phenom with its on board mem controller vs high latency C2D:http://forums.avsim.net/user_files/188319.jpgOther design features help C2D compete n beat AMD, features that help attenuate latency issues.

So guys from your testing what do you think the answer to the subject of the post is? Do you think you would notice a difference in FSX using either the OCZ 800 mhz at cas 3 or perhaps these 1200 mhz Patriots at cas 5?http://www.newegg.com/Product/Product.aspx...N82E16820220282Also what I think I hear you saying is that regardless of what memory you get, it will be noticably better with 4 gb over 2gb. Is this true?If 4gb is the way to go, is 4 x 1gb preferable or 2 x 2gb?Ted

Core 2 Extreme QX9650 4166 MHz [ TRIAL VERSION ] X38 Dual DDR2-1057 4-5-4-15 CR2 51.9 nsIt will do better, maybe under 50ns.Here's my take: if the Patriots will work as advertised, seems like that would be the best performer. Except for the price. The OCZ's are performing very well, but maybe I'm lucky, at about $30 per GB, or so last time I checked. I think you have to go for 4GB any way you slice it. I think cost is a big factor. If these OCZ's fry I think I'll let it go and just run my mushmems at 1103Ghz and call it good . . . Decent memory is becoming quite cheap which is handy . . .

Its not clear whether the Asus P5k will even run memory over 1066mhz. Sounds like 4 gb of the OCZ 800 is the way to go.Thanks for posting all your findings Noel

Sure Ted. I have to think this mainboard does well with 4 sticks in place. So, kinda lucky really. Edited: for what it's worth, I'd have to say the OCZ at 1050+ at 4-5-4-15 is the best so far, tho 3-4-4-10 at 800 was pretty good too.QX9650 w/ Retail HSF|8GB Muskin PC-6400|ASUS P5E|EVGA 8800GT @700|Seagate SATA 2 x 4|Seagate Cheetah 15K.x|XP Pro|Vista 64--soon to be installed

As you may have gleaned, computers are the hobby and heavy jets are the profession. The Level-D 767 and the PMDG 737/747 are the only models around that seriously recreate heavy jet system's LOD. They have modeled a (virtually) fully functional FMS. (We professional users of the real deal nit-pic them unmercifully, but it is really pretty good). Our current gen PCs have an advantage though. A quad core is 100X the HP of that generation's FMC computing power. PMDG had to program a delay into their FMC's keyboard response to more accurately simulate its real world response. To keep in-cockpit frame rates flyable, all AG/AI must stay off. The left side of the scenery slider screen can stay at 100%. I get low 20s on the SADDE6 final to LAX's 24R and 30+ at FL300. Core 1 stays pegged to 100% and the other 3 bounce along loading scenery. (Having multi monitors is great). Sometimes I'll get tired of programming flight plans and emergency procedures. I'll pull out the P-51 and go terrorize the country side. It looks great at 400mph. In other words, 10% won't do it on Anyone's machine. These theoretical assertions may be technically accurate, but real world actualizations by other than heavily funded computer scientists are nebulous, at best. The need is simply beyond what is currently available - to this audience- . Our hopeful nature has consistently allowed imagineering to break an otherwise indeterminable tie. We need a software fix. For right now, these are all just interesting insights into laboratory experiments. Extending hope is a wonderful thing, but sadly, reality tends to intrude. I find presenting these technical possibilities in any other way a recipe for disappointment.