Asus crosshair III with ddr3 1600 - need clarification

http://usa.asus.com/product.aspx?P_ID=2Y4dQFaJ6gPN18cQ&...

make sure that RAM is on the MEMORY SUPPORT LIST before you buy anything.....

1600 on the mobo is just a sales gimic, your amd cpu will only support 1333 and your cpu will under clock the ram to run at 1333.

The OP can buy 1333Mhz ram and overclock it if needed, 1333 and 1600 is basically the same... Lower timings is what matters not MHZ.....

with this ram at 1600mhz its clock ratings are 8-8-8-24. at 1600mhz or 1600 cycles per second the it takes .000625 seconds per cycle. that equates to .005-.005-.005-.015 latency in seconds. now with the ram at 1333mhz the clock ratings are 9-9-9-24. with 1333 cycles per second it takes .00075 seconds per cycle and .00675-.00675-.00675-.018 latency in seconds.... not to mention that the cpu now has a larger pipe to talk to the ram and other devices.

BOTH mhz AND latency are important!

SDRam has a max of 133mhz but has super low timings of 2-2-2-3. where ddr3 is 1600mhz with a latency of 8-8-8-24,, which do you think is faster??

PS: data sheet for ram can be found here:

http://www.ec.kingston.com/ecom/configurator_new/PartsI...

this post was in the wrong thread  

you cannot compare SDRam to DDR3, they are totally different architectures hence the bandwidth differences, the point is that 1333MHZ at 7-7-7-21 is much faster than 1600Mhz at 9-9-9-24 so in this case MHZ is not as important as timings, and the fact that you can raise the FSB as well to increase MHZ if desired.

Another example, I have a set of OCZ Plat. 1066Mhz @ 5-5-5-15 but my G.Skill 800Mhz at 4-4-4-12 runs circles arouind the OCZ's... So no, more MHZ does not mean faster....

wow... wait just a min...

the ram he has listed is 8-8-8-24 at 1600 and 9-9-9-24 at 1333, if you dont belive me, use the link and go to the manufactures white paper on the product. second your not accounting for the increase in processor performance with increased fsb, and you obviously didnt understand my post where i converted the latency's to seconds......

your 1066 ram takes .000938 of a second to cycle once, to retrieve data it takes 5 cycles, thats .00469 of a second. your 800 ram takes .00125 of a second to cycle once and 4 cycles to retrive data. thats .0054 of a second. your ddr2 800 takes longer to retriev data than your ddr2 1066 and thats all there is to it. you like so many other people think that latency is static, but its dependant on the clock just like every thing else in the system.

DDR is short for DDRSDRAM!!! ddr is based of the architure set forth by SDRAM as aposed to DRAM!!

DRAM = Dynamic Random Access Memory
SDRAM = Synchronized Dynamic Random Access Memory
DDRSDRAM = Double Data Rate Synchronized Dynamic Random Access Memory

finalyy your cpu is sitting there waising cycles at 4 GHz because it can only get thigns to work on at 960 MHZ....



PS: i have to appologise as i forgot about Mhz so ddr2 1066 is actually doing 1066000000 cylces in a second, the math will still show the same result you will just have to add 6 0's after the decimal.

lets make this easy to understand...

a ram module has a latency of 1-1-1-1 and a speed of 1hz
to access data from that pice of ram will take 1 second as 1hz = one cycle per second and your ram takes one cycle to retrieve data.

another module has a latency of 2-2-2-2 and a speed of 2hz
to access data it will take 1 second as 2hz = two cycles per second and your ram takes 2 cycles to retrieve data

so these two modules have the same speed, but now you have to look at the cpu.

your cpu is lets say 4hz and 1bit and you use the first ram module
if your fsb is only 1hz, your cpu will get and process 1bit of information on its first cycle and sit idle for the other 3

now you use the same cpu but use the second ram module
your fsb is now 2hz and your cpu will get and process 2 bits of data, the first bit in the first cycle, and the second bit in the second cycle, and the cpu will only waste 2 cycles.

in the real world there are other things talking to the cpu (every thing in the system, prioritized by IRQ's) that take up the wated cycles.

ok, if you are 100% sure of your numbers, then why is it that I get higher Sandra and Vantage scores using my 800Mhz as opposed to my 1066Mhz? Both sets have been tested for errors using memtest so we cannot say that maybe a module is to blame....

im saying that he should find the highest mhz ram this cpu will support with the lowest latency. he should not get ddr3 1066 low latency if his cpu supports 1333 low latency....

ill talk with you after lunch

Agreed, my point is that if the board supports up to 1600(OC), he would be better off with 1333Mhz (lowest timings possible) as opposed to 1600 (higher timings), thats all.... but I still don't understand how my 800Mhz gets higher scores vs. my 1066Mhz...

The only time my 1066Mhz scored higher was by lowering the timings to 4-4-4-15...

im not big on benchmarks, so i dont understand how the program works, but my bet is that i just looks at the latency and says this one is lets say 5 so it gets 10 points where this one is 6 so it only gets 7 points. then it looks at speed and says this one is 800 so it gets 8 points and this one is 1066 so it gets 10 points. if thats the way it works,, but i truly have no clue, it would rate the 800 as having 18 points and the 1066 as having 17. but if you look at the cycles per second, the amount of data per cloock and the delay in clocks (latency) it takes to access that data, good higher speed ram will out beat excelent lower speed ram, especally when you look at its affects on the cpu. he accidentally bought 1600 ram and was wonderin if it was going to work with his setup and the answer is yes as the ram controler will underclock the ram to wahtever the cpu can support, allso in most cases (not this one) when you underclock the ram from 1600 to 1333 or what ever your doing, the latency can be decrease and the system will run stable so if you 1600 is 9-9-9 and you drop it to 1333 usually you can getaway with setting the latency to 7-7-7 or what ever. again though i really have not clue on the benchmarks that rate by a point system, i perfer to use things that rate by throughput, where you should see an increase with the faster ram and therefore faster fsb. but in this case the fsb cant make 1600 (800 actual) as the cpu does not support this. the mobo saying it supports 1600 only means that they dont limit you to a certin speed. ram is controlled by the cpu's on-board ram contoler not the mobo, so aslong as the bios and cpu don't limit anything you could set it to wahtever youd like.

" im saying that he should find the highest mhz ram this cpu will support with the lowest latency. he should not get ddr3 1066 low latency if his cpu supports 1333 low latency.... "

going on what i said above, if his cpu supports 1333 low latency, he can buy 1600 low or normal latency, under-clock it and set the latency to the same or lower than the standard 1333 low latency. in this way when he over-clocks via the fsb (and then multiplyer like you should) he will not be caped by the ram, and if he ever buys a new cpu that supports 1600 he can save from buying new ram to take advantage of the higher speed.

sweet, thanks for the feedback...

Btw, I wasn't trying to act like an arse.... Just misunderstood one of your posts... And the fact that I didn't read where the OP stated that he had "bought" the ram already...

lol probably my post #2   it was porly written, i to was gettin kinda nausty, calmed down over lunch though  

have you played with the tlb patch with your phenom?

No, as far as I know the TLB issue was with the older agenas and phenoms. And the fact that most users that had this issue got a quick fix with a newer bios update.

lol right, thats what i though, but i found this article that says windows, starting with vista sp1 applies the tlb fix to ALL Phenoms...

http://forums.guru3d.com/showthread.php?t=309518

i have a 8400 and had the bios set to disable the fix, but i was getting por performance in games, did this patch (below) and i get 60 fps solid in bishock where i got 30 before. CSS i was getting 40 and now it never goes below 100

http://www.rebelshavenforum.com/sis-bin/ultimatebb.cgi?...;f=13;t=000111

im not sure if the p2 part is true, but since you have one you could bench it  

Yea that was an issue with Vista and not the CPU itself. Most users that had the TLB bug got a quick fix with a Bios update or by using the TLB disable tool . As far as I know the TLB bug was found on the B3 stepping models, that has been fixed with the newer phenoms.

i have a first gen phenom 1 (with the tlb problem), with the tlb bug fix turned off in bios and on is vista i get shitty performance, with it "off, off" i get excelent performance, with it on, on or on off it get crapy performance. i was wondering if this was the same with pII's

ahh gotcha...

yea you have an 8400 right?

right, but there posting results of PII's getting better performance in w7 after running the software disabler.

yea guess I will have to look into that.....

I just thought I'd get back to you all since I have enjoyed reading all the comments this post has generated. I appreciate all your advice and information. The fact is I just wanted to make sure the RAM would work in the motherboard. I bought exactly what I posted, as well as a phenom II 955 and it all works fine. I have no experience overclocking so I'm going to wait on experimenting with that until I have more free time. As far as performance increases in latency vs Mhz you all know better than me. Lets just say it's all significantly faster than my old system with (this is embarrassing) was an AMD 64 3200 with 1 Gb of DDR 400 on a P.O.S. micro ATX board. So I'm happy for now, and if I can make if faster later without making it unstable or increasing the power draw unreasonably then maybe I will do so. Mostly I use Revit and do a bunch of multi tasking stuff rather than play a lot of games, so maybe this system was overkill, I don't know. But again, thank you for all the replies and keep it up if you'd like, it's all very interesting and I'm learning a lot.

Kind of awkward how I got the "Best Answer" but as long as your system running fine that is what counts. Let us know if you have any issues. And special thanks to kerdika for helping out as well...