How does "X" compare to the MSI 975 implementation?Quote:
Originally Posted by macci
We're ready for the boards anytime ;)
Printable View
How does "X" compare to the MSI 975 implementation?Quote:
Originally Posted by macci
We're ready for the boards anytime ;)
managed to push my E6600 + Badaxe rev304 a bit harder but still 2x 512MB PC5400ULv1.5 at 2.2v hehQuote:
Originally Posted by bachus_anonym
http://fileshosts.com/intel/Intel975...4min37s_tn.png
Wow eva getting there! What did you do just tighten up timings and run 4:5?
i think what ram you use effects clocks and ultimately times
on 1067fsb with 2.2v vdimm 1304 bios
2x256MB Centon PC4300 Advance Micron D9 fatties = max 382fsb
2x512MB OCZ PC5400 Platinum XTC = max 378fsb
2x512MB Corsair PC5400ULv1.5 Micron D9 fatties = 421fsb
as you can see only corsair ram would of allowed me to do 9x389fsb in the first place..
for the 14m 37s run i loosened timings from 1:1 4-4-4-5 to 4:5 5-4-4-9 at 2.2v to run 4:5 for ram @486mhz.. cas 4 will reboot probably lack of vdimm heh
Went for my new personal best 32M time. I think this is one of the fastest times under 3.9ghz and 4.0ghz at 1:1 ??
http://fileshosts.com/intel/Intel975...9s609ms_tn.png
Heh, I guess we're getting off topic :lol: But I'm glad that there is consensus to 1333fsb being a memory performance killer and as macci said again, most likely running 1333fsb does clock-down memory frequency thru ratios. Just no software to show it :p: If the latest Everest 3.0 does not show RAM speed as he said, there there is definetely something off :fact:
nah don't think 1333fsb strap is running down clock divider otherwise my ocz p5400 plat xtc would of booted at >423fsb 4:5 upclock...
i had tested all day yesterday
with 1333, 533, 667, and 800
800 is not going to post for sure
667 is not running at 4:5, i think* it is runing 1:1 or 5:4 divider
533 is not running at 1:1 for sure but not sure the divider
I'm pretty sure that 1333 with 667 divider is 1:1. According to CPU-Z, same sticks of Teamgroup D9 2X1GB do:
- 420Mhz 3-3-2 @ 2.6V with 1066/667
- 525Mhz 3-3-2 @ 2.6V with 1333/667
If 4:5 is actually 1:1 at FSB1333, than in both cases the RAM is running at 420Mhz.
Rant mode on
Guys its nothing to do with clocking down the ram if you are in 1333 mode...its ALL ABOUT CHIPSET INTERNAL LATENCY
Do i have to follow you guys around and tell you like 7 times that its chipset related? come on....!!!!
Reduce the internal latency on the mem controller and the ram will work better. Find a way to force 1333 on the CPU strap but force 1066 on the NB strap and you will see ;) Asus obviously have found a way to do this.
This has been the same on all Intel chipsets since 845...you must see this nothing has changed.
Rant mode off...
nice and clean vmod :toast:Quote:
Originally Posted by Supertim0r
@ Super vdd measure half way down thread here:
http://www.xtremesystems.org/forums/...d.php?t=100125
Tony, but be realistic... Do you think, DDR2-1200 8-3-3-3.0 is even possible? Or DDR2-1150 8-2-2-3.0 ??? I don't believe that at all. Come on, no relaxing chipset latencies will do that, period! :fact: Everyone believing fatbody can clock to DDR2-1150 8-2-2-3.0 should be slapped :lol:
Look at the memory clocks P5W DH is showing and search for "WRs" on BadAxe, 1333fsb. You'll might change your mind.
:peace:
get on skype I will explain so you can also....Quote:
Originally Posted by bachus_anonym
Can anyone try Corsair XMS2 Xpert on the Bad Axe with 1333 strap ? Just to clear this up.
thanks ;)Quote:
Originally Posted by Dynasty
http://www.xtremesystems.org/forums/...d.php?t=105703
i hope this helps
A couple years ago I received some Platinum Revision 2 Enhanced Bandwidth Limited Edition DDR2 Modules from Alex Mey at OCZ - Some of the very first modules. These were equipped with Micron Fat Body IC's and I was able to push the memory to 7-2-2-3 @ DDR 960 on a DFI Board with modified bios....Quote:
Originally Posted by bachus_anonym
So, I am personally rather sure that you could see DDR1150/1200 with more relaxed chipset latencies... :)
Just as I was talking to Tony I will say what I think here as well.
I'm not convinced that on Intel DDR2 can clock 30% better just from relaxing latencies between chipset and memory. There is so much chipset can do and so much RAM can do to allow for better memory OCs. In my mind, bringing AM2 example is a good one. Think about it...
AM2 is not restricted by latencies that on Intel mobos chipset brings when communicating with memory and CPU. Therefore, it might be safe to assume that DDR2 should clock on AM2 just as good as on Intel platform, If not even better. That assuming what Tony said, that chipset can make a WORLD of difference in memory OCing. I know it makes but I don't believe in such drastic changes as we see on BadAxe and 1333fsb strap, at least with monitoring software that we have on hand.
Now, with that said, something needs an explanation, don't you guys think? Why on AM2 new D9xxx does not seem to go higher than DDR2-980 5-3-3-3.0 @ over 2.7v so far but according to BadAxe results on 1333fsb strap, the same memory already "did" DDR2-1200 5-3-3-3.0. Then those same people (that tried BX) switched to ASUS P5W DH and we are yet to see DDR2-1000 5-3-3-3.0 on that mobo?
Also, I think that none of the software (CPU-Z, Everest and others) can properly detect 1333strap and therefore, like macci, hipro5 and few others are now saying - memory ratios might be different on 1333 than they are on 1066. I'm sorry, but I think believing that memory suddelny clocks a HEAVEN (not talking 10-20Mhz better but over 100MHz) better just on one freaky board will not convince me :nono:
Full respect to everyone :peace:
EDIT: PERFECT example! Same memory, same voltage but SPi32M @ SocketAM2 vs BadAxe @ 1333fsb strap 4:5
or Max clocks --> Shamino's 6400C3 @ 2.7v on SocketAM2 vs BadAxe @ 1333fsb strap 4:5 and 6400C3 @ 2.5v
vs Tyrou's BadAxe @ 1333fsb strap 4:5 and 6400C3 at 2.75v
Yeah, right! :rolleyes: Someone pleasssseee show me that on P5W DH or P5WD2-E Premium or new MSI or DFI Infinity.... :)
I agree bachus something isnt right. I know my mem can do quite well but I didnt expect 440fsb 1:1 3-3-3-8 at only 2.2v just doesnt add up to me. FUGGER ran this mem for awhile on his 955 and never got any results that I am... I hope that there will be a fix or something sometime.... Should I send you my setup so you can do some more testing maybe you can figure it out:)
@ Tony thanks for the explaining this.....
Its just the issue that you are trying to compare memory overclocking from one memory controller to another...simple fact is you can not.
Intel clocks ddr2 a whole lot differently than AM2 does, if you qualify DR2 on AM2 you can damn bet it will do a lot better on an Intel rig, same ram etc. Ram that will not do TRP at 3 on AM2 may do TRP at 2 on 955 or 975 at the same exact speed...this is a fact as we have seen it.Remember our VX UTT, modules that did 280 2-2-2 on A64 939 would not even do 220 on 875 Intel with the same vdimm so there are differences, its just DDR2 at this time clocks a lot better 2t on Intel over AMD.
DDR1 on A64 clocked a lot better than on Intel, but at this time it seems to be reversed for DDR2 although the AM2 memory controller is a lot more efficient even with higher latency on the modules.
Then you move to chipset latency, you hit the faster strap and the ram needs to drop to say 4-3-3- to clock, then you hit the slower strap and suddenly you can do 4-3-2 or even 4-2-2 at the same clocks...its weird i know but it has to do with the internal clock and latency of the memory controller on the NB.
Faster strap = lower strap IE running 800 instead of 1066, slower strap = higher strap so 1333 instead of 1066 etc
Intel NB need to be treated as a CPU. You have different strap ratio's that we need to call multipliers. As you clock up the fsb on your cpu you also clock up the fsb on the NB, eventaully it hits a limit and will not go any further. So you change the strap and it is able to start clocking again as in theory its clock speed has been reset to default. The only issue is it also sets the latency.
So you up the strap and lower the multi on the NB so you lower the NB clock.
Now PAT was just this as i said before, it increased the clock of the NB and this is why Asus had an ass slapping off Intel over the 865 chipset. Asus set the strap to 533 and changed the internal latency (available in 865 tweaker) and gained a huge performance boost, the only issue was no way would your BH5 clock as high unless you did all sorts of NB vmods etc so the board had a chance.... this is where Hipro5 had awesome success as he applied mods to the board that would allow the 533 strap on the IC7 and still have huge clocks with BH5...
So pick up here on the similarity, with the 533 or 667 strap on the IC7 you could run your ram at moderate overclocks but the performance was huge. You then set the 800 strap and the BH5 would do 30fsb more but performance was not as good...ask yourself why??? you downclocked the NB ;)
Now we are seeing some huge clocks with tight latency on Badaxe, the reason for this is 1333 strap on both the CPu and the NB...its probably set real slack latency...this is why you see 3-2-2-0 ddr1100 etc but performance will not be as good as 900 at 3-3-2- or 4-3-2 on the 1066 strap as the chipset is downclocked.
So some important rules to abide by here:
Don't compare intel to AMD memory overclocking...they are NOT the same.
keep the chipset strap low and the memory latency low and you gain a ton of speed, but you need an awesome NB, great memory and a good CPU to do it, and maybe a ton of mods ;)
One last thing, will someone please find and add the chipset latency tweak to the 975 tweaker, this will allow you to see what im partially talking about; While this will not increase the clock speed it will reduce the latency to the chipset.
To continue.
For the ultimate Intel 975 board, you would need an I/O chip between the CPU socket and the NB. The boot strap set by the BSEL pins on the CPU do not now go directly to the NB they go to the I/O chip. Bios Controls this strap and you are able to force any CPU strap you require.
Now the NB has to see the CPU strap but we have an I/O chip in the way, so we also control independantly thru bios the NB strap.This will allow us to force a NB overclock.
Normally the CPU strap is a direct link to the NB, some do have jumpers that allow you to force a different strap but the ultimate would be bios control over both.
I'm curious about something... Is there anyone willing to try how memory OCs differ at 800fsb vs 1066fsb strap on BadAxe?
Say, get your Conroe to start at stock 266FSB and then start looking for max memory OC at 800fsb strap. Then switch to 1066fsb strap and see if there is any difference in memory clocking and if there is then how big? How far Intel CPUs max out at 800fsb strap? 300MHz or so? That could be an indicator If 1333fsb in fact can make such a dramatic difference, if we are to believe in those huge OCs.
I had FCG do this already, the only issue was he did not document results...Quote:
Originally Posted by bachus_anonym
well, with correct DDR2 1060 4-4-3-4, the 3.8G E66 with FSB1066 32M should be like this
E6600 with initial FSB1066 with GSKILL DDR2 800 2GBHZ
http://www.iamxtreme.net/cpu/e66/100_4344.JPG
compare with my old initial FSB1333..
http://www.iamxtreme.net/cpu/e66/100_4335.JPG
interesting 20s faster than my 3810mhz time of 13min 49s with 1067fsb strap and 2x512mb sticks heh