Tough call but if it were me and I was going to purchase very soon, I would be aiming for the Asus P5E3 PREMIUM which added important afterthoughts in upgraded engineering to the DELUXE (which was a dud) or when available the Rampage Extreme.And do remember that 100% support for Nellie will require a shift in motherboard platform, although the memory can still be transferred to it.

>certainly be more future-proofed than a DDR3 mobo based on the>current 775 socket platform. If one seriously wants to upgrade>their performance, the best bet may be to just sit back and>cope with a minimal deficiency now, waiting 6 months for the>next-gen. The payback will almost certainly be outstanding. >We're ALWAYS in that same situation. :) Take any point-in-time in the FS timeline going all the way back.I do not know if I will wait until Nehalem (which I think for mainstream cpus will be more like 8-10 months away, unless AMD is able to release something to push Intel into acting faster). Perhaps I can. However, you can see what I am running (3700+, 7800GT) and it is about the minimum I would want for anyone running FSX. It's past time to upgrade. I doubt I could get acceptable framerate with a 22" or 24" monitor at any native res with my current rig. Sometimes I'm glad I still run a 17" Trinitron at 1024x768.But since I've become an 80%/20% VC/2D flyer, you can see where the wind is blowing.RhettAMD 3700+ (@2585 mhz), eVGA 7800GT 256 (Guru3D 93.71), ASUS A8N-E, PC Power 510 SLI, 2gb Corsair XMS 3-3-3-8 (1T), WD 150 gig 10000rpm Raptor, WD 250gig 7200rpm SATA2, Seagate 120gb 5400 rpm external HD, CoolerMaster Praetorian

Nick that was quite the effort. That qualifies as a term paper. It took me awhile, but I finally made it through all your threads and links on memory tuning. I don't know how you find the time to make all the posts you do, but I enjoy reading them and thank you. Not sure if I understood it all, but my questions should reveal that.In your tests above I understood you to say that you were using DDR3 memory at 900mhz with a tRD of 12 to represent fast DDR2 memory. However you said that the Patriot memory above should achieve a tRD of 6. So it seems that fast DDR2 memory should be able to achieve the results of Test 2. Am I missing something? It appears that you can achieve the same Trd delay with DDR2 or DDR3. DDR3 just seems to offer more bandwith due to being able to operate at a higher frequency. Is this correct?Reducing the wait states makes sense in improving the memory efficiency and performance. However, doesn't increasing the bandwith on the memory buss above that of the CPU buss just make the CPU buss the bottleneck?As much as I'd like to have a $400 X48 motherboard, a $1000 CPU, and $400 worth of Cellshock memory, the funds aren't available. So it looks like a Q6600, DDR2, and a P35 motherboard for me. With that in mind, do you know which P35 motherboard would give me the most options in optimizing the Partriot DDR2 1200 memory you pointed out above? I did notice that ASUS makes one that can take DDR3 or DDR2, allowing the use of the faster DDR3 once the prices come down. http://www.newegg.com/Product/Product.aspx...N82E16813131188Thanks again for sharing your knowledge and time here. As Sam said, this forum has gotten a lot more interesting lately.Ted

>In your tests above I understood you to say that you were using DDR3 memory at >900mhz with a tRD of 12 to represent fast DDR2 memory. However you said that the >Patriot memory above should achieve a tRD of 6. So it seems that fast DDR2 memory >should be able to achieve the results of Test 2. Am I missing something? You are correct and because DDR2 1200 @ CAS4 is faster than DDR2 900 @ CAS5, you should be able to do much better but remember, the test 2 results for flight with the scenery load set and balanced up for DDR3 was not that impressive. None the less with some traffic and scenery tweaking that would fall into place, especially with a CAS4 rating even at DDR2 1000 if you had to down-clock. At that point I would also test for 1T ability and may even take that over DDR2 1200. At 400FSB, DDR2 1000 5:4 and 1T CMD you are still running very strong. 1T is a very overlooked ability today in order to get higher bandwidth. I always strive for success @ 1T and will usually sacrifice some bandwidth to get it. It will also serve to lower latency but its something that must be tested to see which, the bandwidth @ CAS or 1T CMD, delivers the better result. >It appears that you can achieve the same Trd delay with DDR2 or DDR3. DDR3 just >seems to offer more bandwith due to being able to operate at a higher frequency. Is this correct?That is correct. It all works together and although the scale for the amount of dataflow which can be processed without restriction issue is significantly increased woth DDR3, the same principle for clocking DDR3 is used for DDR2.MCH READ DELAY or tRD is not the holy grail.. as an old slogan goes..

>Your suggestion of shopping cheap and taking a chance is going>right back to where you are now with the sticks you are>running. Yes, and no. Don't forget, with these $39 2GB modules, rated only at 800Mhz, they have no trouble dialing down their rated 5-5-5-18, to 4-4-4-10. I'm thinking there is a high likelihood their 1000Mhz modules, actually their REDLINE upgrades, should run those timings as well, or have a shot at it anyway. Therefore, running those at 400MHz FSB, there is a high likelihood of hitting a decent configuration that should net latency of 12.5sec x 4, 50ns. That's decent. I think I may be able to pick them up at newegg, test them, and if their cool then we have a nice cost effective solution. If not, they have a refund up to 30days I believe, and will verify. If those don't fly, then maybe consider an upgrade of modules to something pricier. I'd also prefer to stick with 2gb modules. Mine are extremely stable at a cool 1.80v.Well, as I orginally explained, I accepted Sam's portrayal of the FSB and MCH interface really was sort of a passive conduit: as long as you have not exceeded the raw carrying capacity of the FSB & MCH there are no issues. I think there is theoretical and empirical basis for proposing there are timing issues that critically affect performance and smoothness. Unfortunately, I didn't appreciate this when I purchased my mainboard. I went for cheap DDR2 believing it was a non-issue. Even so, the machine runs really well. Shoot I'm getting 3DMark scores over 15,000 which is decent in itself. FS9 tricked out maximally runs totally smoothly at high frames. So, it's . . . OK that it isn't totally perfect. But, for a couple hundred bucks I can get another 4 or 8GB of memory that may improve the picture enough to warrent the minor hassle. I would consider a mobo and memory change were I to sell off my current mainboard, which may happen in a year or so when my brother will want to upgrade. Thank you for your insights.

Give it a shot Noel.. if you can return them you have nothing to lose and everything to gain, even if its only in the learning experience of working with different modules and getting a 'gist' of their 'personality' and how they respond with the board/BIOS. Do make sure the board can handle their voltage requirements.You may get lucky. Those old CAS5 1066 Crucial Ballistix modules that were hitting 1200 and tRD5-6 and were the gems to the clockers were only about 70-90 bucks with a rebate at the time.Remember bandwidth is just a frequency, it's the data flow you are trying to port, it's speed you are trying to increase, and, reduce the time the 'RED LIGHT' is 'on' you are after. You have much more CPU and FSB ability than you do memory bandwidth flow ability in DDR2. My test shows quite clearly even the DDR3 1440 bandwidth could be higher/better, effectively resolving the last small bit of issues seen in test 3, and that was only 3.6GHz @ 450FSB.Good luck and good hunting

"Some of those double sided 1066 Ballistix may still be floating around which were able to do CAS4 @ 1000 and CAS5 at 1200-1250 running the Crucial spec of 2.2v but unfortunately the high quality runs are pretty much depleted now. The better sticks use the Micron D9GMH chips and the new ones which started hitting the shelves in February are just not hitting the numbers, at all."From this comment Nick, it would appear you are proposing FSB is quite signficant in the performance equation. I think I would have to give the nod to Sam for suggesting FSB is really, from a practical sense, a non-sequitur issue. From Anand's article:"Without a doubt, 550MHz FSB represents the coup de gr

Noel, This thread and what I have posted is not about being right or wrong.. Its about what -is- right by electronic engineering standards when it comes to memory speed and latency in proper use of FSB/STRAPI have already posted many times in this thread it is not about the FSB or the CPU.. You have more FSB/CPU than you have memory bandwidth for FSX when FSX is being pushed to deliver and no DDR2 system is going to match that of DDR3 for data flow ability without some serious voltage and cooling applied. You can get close to equal results of DDR3 1440 at DDR2 1200 4-4-4-10 assuming tRD is 6 or lower, but thats comes with a cost of heat and most likely vMCH and vdimm increases, and, very possibly a NB voltage increase to maintain stability as it did in my tests last year.Unlike DDR2, DDR3 memory allows FSB to be raised stable with the chipset design and should remain between 450-462MHz until DDR2 2000+/1600FSB processors are the norm, at which time the optimal FSB is then raised again to 500MHz and above. The chipsets will then optimally perform at those FSB/STRAP speeds.The charts are quite clear and when the memory divider/STRAP are factored in with the memory speed, that is what makes the difference, not raw FSB, not raw CPU and not raw mem-speed. Alone those elements provide restricted results, together and in their correct use the results are elevated, by far.and one other thing too.. I have also warned many times in this thread that the use of more than 2GB can invoke timing changes the GUI readouts will not show. I highly suggest you test your system with only 2GB of memory, and, make sure to use the correct slots for that test before moving to a 4GB analysis.Assuming you can meet the terms of the formula with the memory at 400FSB and lock it down @ tRD6, each MHz higher in FSB over 400 will then provide a boost you would not get otherwise. In that, 405-410FSB w/formula -stable- is stoking the fire and not wasting bandwidth. That is why the charts show 450FSB for DDR3 to calibrate, and, 462 is actaully the hard "sweet-spot"At 462 my memory would have been running 739MHz or DDR3 1478... which drops it to tRD6 on the P5K3 Deluxe (above DDR3 1475)@ 5:8 divider (333 STRAP):450/8x5=720MHz462/8x5=739MHz462FSB means I would need to raise Vcore to remain at 8x and raise the fan speed on the Thermalright 120x, or, reduce the multiplier to 7.. which is not bad because the resulting loss in GHz would not hinder the FSX result after crossing the tRD6 threshold and approaching tRD5.Unfortunately that Kingston memory will not run 462 CAS6 stable however I can boot it, and, I can use a vMCH volt mod I did apply to the motherboard since its not available in the BIOS but the sticks get very hot without special cooling, so, it's not worth the risk or everyday use with those modules. I can install special heat sinks and do it but this is not about -extreme clocking-, its about using the hardware and not having to go to extremes for that use. Go for the memory/FSB/STRAP settings first while maintaining the multiplier that will hold CPU at default GHz.. then raise the CPU multiplier, last.. and watch those temps.And without FSX I fail to understand how you may be 'gauging' the results... With your current hardware, FS9 is not saturating your bandwidth and crushing your CPU. You should however see a marked improvement in audio/video encoding.:)

Nick, I get it with DDR3, this is not my area of concern. I will not be changing mainboards unless as I say, my brother wants to upgrade his old P4 platform, which won't be real soon. With my X38's BIOS as it is, I believe my only direct control over tRD is thru adjusting Transaction Booster (MCH Read Delay) +/- 1-3 clocks. I have no idea without trying out the modules I mentioned, what sort of tRD will be allowed. It will be trial and error, and verification with MemSet. I need more than 2GB with my machine, AND . . . FSX is not the only program I use, nor will it be the most important. 4Gb is the bare minimum, and I don't want to use 1GB modules because I want the headroom (for audio production) for 8GB.Having said that, I have to pose this again, which I don't recall was responded to in a meaningful way on your end. The whole visual quality & smoothness & complexity = performance position, which who could disagree with, depends on two things: processing demand (apparently this is a dynamic variable in FSX from what you are saying), and processing capacity (which is fixed, but optimizable).If I am a happy camper dialing back some AI, traffic, autogen density, there hopefully is a point at which, for example, terrain texture clarity is approaching maximum, even with my clunky 15ns of latency per clock cycle, when other demands are dialed back. In your humble opinion, given what your testing has shown, can you make a prediction about what sort of slider settings in FSX I could expect to use to keep terrain texture sharp? This IS the important part for me: smooth & clear performance, even if some objects are not there. That doesn't bother me, as I have said, I dial back some of this cuz i like the appearance of being less cluttered in FS9. Any guesses here Nick?With your Test 3 outcome, latency should be approaching 9.7ns x 4 = 38.8ns. That is really impressive by comparison to my platform, and presumably from your testing, it matters. I think the best I could hope for with DDR2-1000 or 1200, would be around 50ns. 50ns vs 60ns is about a 17% difference, not really too much. Put another way, if I dial back 17% of the demand of processsing in FSX, what will this translate to in slider positions? This is really where the rubber meets the road in my situtation . . .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

Nick N, everytime I post to you I feel like an average 6 year old talking to Steven Hawkings about advanced mathematical concepts and particle physics but here goes. Based on what you are recommending (don't scrimp on the CPU and memory and especially don't scrimp on the motherboard) how does this look? ASUS P5E3 PREMIUM/WIFI-AP@n intel x48 chipset ATX form factor 3xPCI-E(x16)/1xPCI-E(x1)/2xpci/4xddr3 w/sata2 raid,dual lan(gb),1394,usb 2.0 & audio (cpu type:intel - socket 775) from Mwave for $384Patriot 4GB Kit PVS34G1600LLK for $540 also from MwaveIntel Quad QX 9650 from Newegg.I know that this is just raw material and would need to be tweaked and set up correctly but would this be a reasonably good foundation to start with? Any recommendation on GPU configuration with a set up like this running FSX/Acceleration with Vista 64bit. I was thinking of triple head 2 go with 8800gtx/sli, I know the case and cooling system will be important if I do this system. Any help or advice from you or the others in the thread would be appreciated.Incisal Flyer

I guess you missed my post above in response to your questions, especially SLihttp://forums.avsim.net/dcboard.php?az=sho..._id=37371&page=What you selected is fine. The Premium version of the P5E3 family is a much better selection than the deluxe

Are those shots of you in constant motion? I can never get my 7cm roads to remain crisp while flying faster than about 70ktsThen again, your system is probably twice as fast as mine in general....What is your system btw?

Yes, just as the benchmark shots were in constant motion at 250KtsThe system parts/specs which were used in the demonstration and was the source for all of the images, were listed in the first thread of this presentation but here they are again:Asus P5K3 Deluxe WiFiIntel QX6700 @ 8 x 450 FSB (3.6GHz)CPU STRAP 333CPU Vcore 1.45vThermalrite 120 Extreme with a single 120mm fan8800GTX / Thermalright HR-03 cooler (+90mm fan) running 641core 1052memory (2104)3ware 9650SE 4LPML PCIe 4x RAID Controller 2x WD 150GB RAPTOR (FSX DRIVE)1 ST380817AS Seagate Barracuda dedicated to Windows XP x64 SP22x1GB Kingston PC3-8500F DDR3 1440 (720.0 MHz) (Ratio - 5:8) CAS-6 7-6-15 @ tRD 7

>Asus P5K3 Deluxe WiFi>Intel QX6700 @ 8 x 450 FSB (3.6GHz)>CPU STRAP 333>CPU Vcore 1.45v>Thermalrite 120 Extreme with a single 120mm fan>8800GTX / Thermalright HR-03 cooler (+90mm fan) running>641core 1052memory (2104)>3ware 9650SE 4LPML PCIe 4x RAID Controller 2x WD 150GB RAPTOR>(FSX DRIVE)>1 ST380817AS Seagate Barracuda dedicated to Windows XP x64>SP2>2x1GB Kingston PC3-8500F DDR3 1440 (720.0 MHz) (Ratio - 5:8)>CAS-6 7-6-15 @ tRD 7Nick, what is your estimated latency with this configuration? Is it 7 * 1000/720, or is it 7 * 1000/450? I thought Anand's article gave it as tRD * 1000/FSB?I don't understand the strap, the divider, concept. Can you give me some insight on this? My BIOS doesn't show a divider per se, but does have the strap settings as 200/266/333/400.

AH HA!Now I see you are getting the gist.. yes, it is tRD x 1000/FSBSo that places the trd latency on that system I used at 15.5 which is a bit high, isn't it. I mentioned with that board/BIOS I must hit 1475 to drop to tRD6 OK, now look at my CALWI. http://forums.avsim.net/dcboard.php?az=sho..._id=37304&page=At CAS6 1440 I am running right around 8.25ns (respectfully). That plays into this as well however it is also a bit high for the performance mark/lantency of tRD6. tRD 6 would demand a CAS CALWI of 8ns or less to be on the high end of tRD6. That is why I would need to push FSB to 462. My CALWI is high therefore I need to increase FSB to compensate.Those tRD setup charts only get you in the ballpark. The BIOS/chipset may demand small changes to meet the theory of the charts.Now, here is why I said AH HA!... If my CALWI is high, and, I am not running the optimal tRD which is 6 or 13.3ns... then a very defined reason for my performance increase in the presentation was..... Memory Bandwidth and CALWI So that bandwidth and CAS is more important than you may think. Having it sit 540MHz higher than the DDR 900 test runs and a single tRD level lower made a significant change. I raised CAS from 5 to 6 doing that so the true 'major' difference between the DDR 900 tests and the DDR 1440 test was in fact,