Noel,Backup

Ok, awseome discussion. Maybe I can pull this off. Nick, do you have a specific recommendation for best Mobo, DDR3, CPU combo (including manufacturer and vendor) if you wanted to create such a system? How about case, cooling, GPU(s)? Thanks for any help.Incisal Flyer

> >Noel,>>Backup

Ya'll might want to consider the entire P35 platform (x38-48-et-al) will become obsolete in 6 months. The Nehalem will require a new mobo and DDR3 memory. This new Nehalem based platform will use an AMDish buss system called Quickpath and an AMDish CPU based memory controller. I doubt any of this will help the platform's performance (any more than it helped AMD's application of this same architecture), but it will 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. Noel, enable "Transaction Booster", in the bios then pick a level. Boot to windows and run Memset http://www.tweakers.fr/memset.html"Performance Level" is tRD. Experiment with the level setting in the bios and see what tRD comes out the other end. The board will decide what tRDs are set by your bios "level" choices. How do it do dat? Who knows, just do the math and get "level" to set tRD as close as you can. Tweak-on. Hey, we're counting on you. Great fun to play with.

>Noel, enable "Transaction Booster", in the bios then pick a>level. Boot to windows and run Memset>> http://www.tweakers.fr/memset.html>>"Performance Level" is tRD. Experiment with the level setting>in the bios and see what tRD comes out the other end. The>board will decide what tRDs are set by your bios "level">choices. How do it do dat? Who knows, just do the math and get>"level" to set tRD as close as you can. Tweak-on. Hey, we're>counting on you. Great fun to play with. Thanks alot Sam, I was doing this even as you were posting this message, having finally figured this out. I was looking at a DDR3 article on Anand and noted the commentary about Transaction Booster. I am able to boot to Windows at a transaction booster level of 2 max. I had downloaded MemSet and voila, I have the data. Shows a tRD of 7. So using Anand's formula for latency, Trd = 7 x 1000/400 = 17.5ns. If Nick's recommendation of memory to get to tRD of 5 comes to pass, latency goes down to 12.5ns. This seems signficant, but only to the extent the application exerts the sort of demand that matters. Interesting discussion! I think with all this being said, there is not much point in changing mainboards. Perhaps changing memory might be worth doing as the cost shouldn't be too bad. Downgrading to 4GB is probably real reasonable for me, even with Giga 4.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

I wish to make something clear... the Transaction Booster when it displays values such as -3 -2 -1 0 +1 +2 +3 in the BIOS is dynamic on many motherboards and may not be a completely static setting. It may vary under a load... be carefull using both. I typically turn off the Transaction Booster (set it to zero if its an option after disable) and use the formula to work the value. In a case where memory will not provide the needed change due to limits I would use Memset to take up the slack with the booster disabled. Newer motherboards will allow a full MCH READ DELAY setting like MEMSET along with vMCH. I would opt for Transaction Booster {DISABLED] {0}If available: DRAM Static Read Control {ENABLED}:)

OK, so we look now at Anand's use of Everest in the X48 article, discussing tRD, under the "Real World Results" section . . .an FSB of 400MHz using a 5:4 divider for DDR2-1000 with 4-4-4-10 primary timings at a Command Rate of 2N.tRD of 5 = memory latency of 53.5ns, memory read of 9166 mb/s of 7 = memory latency of 58.5ns, memory read of 8652 mb/s"Not only were we able overclock the controller to 450MHz FSB but we also managed to maintain a tRD of 5 (for a TRD of about 11.1ns) at this exceptional bus speed. Using the 3:2 divider and loosening the primary memory timings to 5-5-5-12 allowed us to capture some of the best DDR2 memory bandwidth benchmarks attainable on an Intel platform." This gets latency to 45.5ns, read to 10,950 or so.Noel's old clunky DDR2-800 tested using Everest, amped up as best as can:memory latency of 60.4 ns/ memory read of 8165mb/secIt would appear I might be able to go to as low as 45.5ns or so latency, and up to 10900 or so read rate. But also, I am running my rig at 3.8GHz right this moment. How this will affect FSX performance, seems from Nick's arguments to be signficant. It's not a giant difference. What I'm not certain about is, does using Transaction Booster give the enough control over tRD as the setting on the X48 ASUS board. If it will, then perhaps this is worth the modest hassle of changing memory modules.If me brother decides to upgrade, I've got some memory modules for him :().One last test: amp up to 4.3Ghz, and run Everest . . .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

Remember Noel, the CAS, bandwidth and the divider all play into this. At DDR2 800 you are going to be limited no matter what.By the way, you are on the right track, get that latency down as I showed in the presentation and you will see a difference. @56ns I saw a nice change but nothing spectacular but at 48ns and the bandwidth increased, all the stops came out.The reason the myth has been perpetuated so much is because those who have worked with memory clocking usually only work with speed, miss something, or, they simply do not have the bandwidth high enough to see a major change.60ns is nothing to write home about... get that down to 55 and under and I know you will see a change if other areas are in sync.:)

>Remember Noel, the CAS, bandwidth and the divider all play>into this. At DDR2 800 you are going to be limited no matter>what.>>>By the way, you are on the right track, get that latency down>as I showed in the presentation and you will see a difference.>@56ns I saw a nice change but nothing spectacular but at 48ns>and the bandwidth increased, all the stops came out.>>The reason the myth has been perpetuated so much is because>those who have worked with memory clocking usually only work>with speed, miss something, or, they simply do not have the>bandwidth high enough to see a major change.>>60ns is nothing to write home about... get that down to 55 and>under and I know you will see a change if other areas are in>sync.>>>:)>>>>>>>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

>Remember Noel, the CAS, bandwidth and the divider all play>into this. At DDR2 800 you are going to be limited no matter>what.>>>By the way, you are on the right track, get that latency down>as I showed in the presentation and you will see a difference.>@56ns I saw a nice change but nothing spectacular but at 48ns>and the bandwidth increased, all the stops came out.>>The reason the myth has been perpetuated so much is because>those who have worked with memory clocking usually only work>with speed, miss something, or, they simply do not have the>bandwidth high enough to see a major change.>>60ns is nothing to write home about... get that down to 55 and>under and I know you will see a change if other areas are in>sync.OK, how about getting specific. I looked at newegg, and all their DDR2-1000 or higher is CAS 5 or higher. What I'm not certain about (are you Nick?), is does Transaction Booster (max of 3) allow manipulation enough to get to tRD of 5? 6? It looks kinda iffy bothering to go to 6 from 7 when you look at the relative latencies. The DDR2 1000 at 450MHz, loose times, with a *manipulated* tRD of 5 was nicely at 45.5ns. Is that going to happen on my X38? Can you make an educated projection, or are you missing to much info to make this deterimination?So, make a memory recommendation, so I can check pricing. It's been a good study so far. Definitely don't want to change mainboards at this time. What DDR2-1000 will do CAS of 4? Maybe not. I see Anand had to "loosen" timings to get to 450, to achieve that latency of 45.5ns. What about DDR2-1066?The other thing I mentioned is definitely valid. I can also dial back some settings (AI, don't really care for it all that much anyway) and autogen density can dial back for me.Thanks much for the help . . .

http://www.newegg.com/Product/Product.aspx...N82E16820145044http://www.newegg.com/Product/Product.aspx...N82E16820146114http://www.newegg.com/Product/Product.aspx...N82E16820145165and these may do CAS4 at 1000-1100http://www.newegg.com/Product/Product.aspx...N82E16820227271definitely CAS3 @ 800I like OCZ.. their forums have engineers who will work with you on subs and setupthere are others but I am not at home right now with my links.. will check and post more later

>definitely CAS3 @ 800Thanks Nick, CAS3 is def not happening with my cheap Mushkin. I tell you though, it is completely stable, all 8gb, with latency of 60ns, which in the context of many other rigs running FSX, is really not too bad.I will check out the recommendations . . .Noel

What do you think of this:OCZ Platinum Edition 4GB(2 x 2GB) 240-Pin DDR2 SDRAM DDR2 1000 (PC2 8000) Dual Channel Kit Desktop Memory Model OCZ2P10004GK - RetailCas Latency: 5Features: 2.15V EVP Platinum Mirrored XTC HeatspreaderHeat Spreader: YesVoltage: 2.1V Model #: OCZ2P10004GK Item #: N82E16820227280 Think it will do CAS4? Even my cheap mushkin at 800 can do that. If I ran this at 450, perhaps it would go at CAS4. The one's you suggested are just too much $$ at this point in the game, I mean, for what it will deliver. This stuff's $96 after rebate, for 2GB modules. I would buy two. I wonder if I can return it if it will not get me adequate latency?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 the guide for best configurations:http://forums.avsim.net/user_files/187235.jpgThe only question I have is whether all DDR2-1000 memory is able to run at tRD of 5. It looks like that is the implication.

Back upThe chart is not absolutes and fact... the chart is the formula theory applied to real world ability, meaning, it represents what you CAN DO -if- the parts WILL PROVIDE in ability/stability and the correct BIOS/MOTHERBOARD programming is present.Your suggestion of shopping cheap and taking a chance is going right back to where you are now with the sticks you are running. Are there modules out there that are cheaper and will perform within the formula? Absolutely. Can you tell by looking at them on the net? No Way.That one is the luck of the draw and a bit of research into what the clockers are using now a'days and the parts bin/serial numbers.I have not been around DDR2 lately so I have not kept up on what is and isn