|
|
Will you be overclocking a P45 mobo? READ THIS!
Author: Richard SwinburnePublished: 12th May 2008
Overclocking Intel's P45 is said to be more complex than it has been on previous boards - you're going to need to do a lot of specialist fine tuning.
Overclocking has been progressively getting more difficult since the 975X chipset was introduced, but it wasn't until more recently with P965 and P35 that more advanced options have been released by Intel as the antiquated front side bus runs out of steam.
If you want to buy and overclock on an Intel P45 board (or X48, X38 and P35 for that matter), Leach believes that you must learn how to use:
- GTL Reference Voltages
- CPU VTT and its relation to GTLs
- Clock Skews
- CPU PLL Voltages
It’s a case of spending a lot of time increasing the CPU VTT (not over 1.35V – you’ll kill the CPU) and CPU PLL (not over 1.7V, because again you'll kill your CPU) and tweaking the GTL Reference voltages for the CPU and North bridge to be around 61-63 percent of VTT for 45nm CPUs and 67 percent for 65nm.
This is particularly noteworthy with quad-core processors because, if you’re finding two cores drop off under Prime95 load, it’s probably down to the fact that the two CPU dies are not identical and while core one and four can hit the FSB you’ve set, the other pair of cores are having trouble. Tweaking the GTL can sometimes give them better stability.
Leach even went as far as to say you’ve got to find points on the board and check the actual voltages with a multimeter, because we’re talking some extremely minute changes and if there are elements of vDrop from the board or vDroop when the CPU loads the BIOS can be inaccurate.
Next, it's strongly suggested that you also play with the clock skews – we know MSI and Asus will have these on their P45 boards, Asus has them on its current X38 and X48 boards, while both DFI and Gigabyte also have them on their X48 boards. Basically as the data has to jump from the front side bus domain to the memory domain this window becomes smaller and prone to more jitter, the faster you increase either the front side bus or memory frequencies and timings. By adjusting the skew you can realign these clock signals and suddenly stability should return again – a good board will have less jitter in its signal generation and finer skew adjustments than one that's not as good.
The kicker that this is a completely blind art – you’ve quite literally got to sit there for hours and tweak the nuts off the board trying combinations of GTL and Skew settings until you find something that works. You can calculate some of it, like the GTLs, but throwing all the elements in a pot will still require a considerable degree of discovery. However, as soon as you change the front side bus, memory timings, the CPU (no two CPUs are identical, even if you buy a “Q6600 G0”), the memory sticks (there are different tolerances between batches of the same product, never mind different products!), update the BIOS, or even if you’re using the same board as someone else there’s no guarantee that one set of settings will work on another board.
Addendum 14th May 2008: This is not to say you won't get anywhere with just a quick splash and dash, you certainly will, but by learning the new options available and using less voltage - so therefore it gives less heat - you can get the same effect as the brute force method. By combining the two, you can go like a Duracell Bunny: even further, for longer. Speaking to Asus product managers today, they again reaffirmed that just 1.25-1.3V on the northbridge and playing with the CPU VTT and NB/CPU GTL factors can offer a whole more world of performance margins, especially if you are considering DDR3.
So there you have it, prepare to invest in some serious time if you’re upgrading and want to aim for that serious overclock. With this in mind, is there even any point in review sites “reviewing” overclocking any more, since it's unlikely that the end user will be able to replicate it because of equipment diversity? Let us know your thoughts in the forums.
Related articles on bit-tech.net
Comments (39)
Stats: 0.083 seconds
It's not that, it's just we've reached the limit of the front side bus.
Oh, please. Bloomfield is not all 'Extreme Edition', and of course it will be expensive at launch, everything is expensive at launch. Come 09, prices will be perfectly sane. Please don't buy into everything Fudzilla says.
Would it be better if they swapped to a hypertransport (?) ala amd?
Good old times of:
Google: Nehalem CSI
Apart from the obvious geek factor, is there any real need to overclock processors these days? If you've already got four cores running at 2GHz, can you really notice the difference that a small amount of overclocking makes to everyday computing?
And besides, if Intel does lock overclocking in future processors, I've still got my graphite pencil handy :D
Part of the mass appeal of Intel's Core 2 lineup is such that for enthusiasts and overclockers, the Core 2 offers low-price parts that can be overclocked to within the performance of high-end parts that cost many times more; all the while outstripping the performance equivalent to AMD budget offerings considerably. If Intel lock out overclocking on their Nehalem range bar the Extreme Editions (or whatever nomenclature they take up), we could see a very slow takeup of their new processor in the enthusiast marketwhile cheaper and more overclockable Core 2 parts are available.
I've been saying this BEFORE Fudzilla - since I first saw the 2 SKU approach. I've been talking to others who have shared my feelings recently too.
It would make very little sense for Intel to deliberatly stick two fingers up to it's enthusiast market (which lets face it only amounts to around 1-2% of their total revenue). The only thing they would really stand to gain by doing this on purpose is a bad reputation amongst enthusiasts and the "hardcore" overclocking community.
Simililarly, it would make little economic sense for them to re-engineer nehalem's architecture to "allow" for overclocking, as it's such a niche market to them.
The way I see it is simple:
Either nehalem will be a big enough leap over the current generation that overclocking becomes of little concern or, failing that, there's always the green team :)
Either way I think, as enthusiasts, we shouldn't get our knickers in a twist just yet.
it's not a small amount. e4300's go from 1.8 stock to 3.0 overclocked. q6600 go from 2.4 stock to 3.6 overclocked. that is 166% and 150% of stock speed, and makes a huge difference.
still, is it necessary to know these new things or can you still use old overclocking techniques? either way, I hope bios coders make an option to overclock with these new techs and have an old school overclocking option along side.
Let me explain more in depth what i stated at the MSI launch event.
FSB has limits, with quad cores the second set of cores usually does have issues scaling on high fsb, you can dial this back by adjusting GTL ref voltages in relation to the VTT voltage applied to the CPU. It is important you do this as it can open the tolerance window wider for the cores to scale...so CPU's that stopped at 380fsb may be able to scale over 400 etc.Also remember the NB also sits on the FSB and as such some adjustment of the GTL voltage may also be needed to keep the working frequency window open also...so you have CPU cores to worry about and the NB its self.
Memory speeds also suffer from clock misalignment at high frequencies, DQS skew can help re align these differences and allow data transfer to work properly at these higher memory and FSB speeds...remember there are 2 domains that the MCH controls/works with...the memory domain and the FSB. data has to move from one to the other.
This effects all Intel chipset boards, NOT just P45, everything from 975 onwards really is effected by all these tweaks as all are capable of running quad core cpu's and as such FSB issues are more apparent to them.
Let me give you an example of some issues i was seeing recently:
Customers were complaining they could not run the new 1150Flex II at 1150MHZ on the Asus Rampage Formula, upon testing I too saw issues getting the ram to clock any higher than 1030MHZ just like the customers. I then proceeded to play with the MCH Skew option in bios, advancing both channel A and B to 300ps, this allowed the ram to clock all the way to 1180MHZ and also as a byproduct the CPU fsb also was clocking higher.
So without the skew the ram was worthy of an RMA, BUT some fine adjustment had it working faster than rated speeds.
As you see to get the best from an Intel platform you have to tune the board correctly, fail to do so and you will think you have components that are either "just not good overclockers" or "working below rated specs"
As FSB rises it is really important you know how it works and apply the tweaks that help it rise so you get the best from all your components.
Intel could hardly care less about the enthusiasts trying to get a bit of extra performance out of their chip. The VAST majority of their equipment goes to OEMs like Apple and Dell, neither of whom do any overclocking. They don't care about a few hobbyists not paying top dollar when the big guys are doing tens to hundreds of millions of dollars in business with them.
As the engineer (read: someone who is paid a lot to know this stuff inside and out) has stated, there's just so much complication that it's hardly worth the time.
the speed these chips are running is already blindly fast, 500Mhz is no where near slow! even the average PIC or AVR are running fast enough at 8Mhz, computers are just getting way too complicated now. if only they can start from scratch, without all the backwards compatibility rubbish, then we can have simpler chips, and simpler chips means easier to hit higher speed, it's also cheaper to tweak/produce.
That sentence might be the main reason why Intel might actually WANT to lock people out of overclocking unless they buy the expensive CPUs? Why pay for the top of range extreme edition when you can overclock the (relatively speaking) very cheap Q6600 and get very similar performance?
They already do it to some extent with Core 2 with the locked and unlocked multipliers for standard and extreme CPUs respectively, fixed clock and overclockable is the next logical step...
All AMD has to do now is continue being overclock friendly and Intel is just shooting itself in the foot, far as nahalem goes anyways.
The LGA1160 or w/e it is that has the supposed lockout is the first Intel motherboard to support IMC, GPU and integrated PCI-Express.
If you look at the Bloomfields, they only have integrated DDR3.
I'm thinking that if the overclocking lock is real, it might have something to do with the mentioned features. Cause IMC, GPU and PCI-E all have different frequencies and Intel might not have figured out how to clock them differently through the bios? That would also be a good reason why they're coming out up to 6ish months after the normal Bloomfields - you have to get a lot of stuff right for everything to work (processing wise).
As much as people say that Intel is all about the business, we're starting to see that the enthusiast is a bigger and bigger part of the market; look at IDF. We saw watercooled and overclocked Skulltrails. If that's not sign that Intel is turning an ear to overclockers i don't know what is.
p35 have this problem. all motherboards have this problem at 500Mhz. and because p45 is now 400Mhz as standard, 500Mhz is the overclocked speed rather than then 400Mhz overclock most of us use on p35.
the reason AMD doesn't have this problem is explained very clearly in the article
I don't profess to know either architecture particularly well and I simply stated that I'd read elsewhere that nehalem prevents overclocking in an almost "coincidental" manner. A side effect of the new design if you like.
Never did I claim that the mem controller being on die was the sole reason behind this, though admittedly, I could've probably worded it better.
The fact remains, I can't see Intel making great efforts to either encourage or discourage overclocking - as has been stated several times, it's just of such little interest to them in the grand scale of things.
1) they are a small % of volume but they buy products with a high asp/high margin.
2) they influence a lot of non-enthusiasts in their choice of product even if they will never use what makes that product a good one as far as the enthusiast is concerned.
*intell don't mind overclocks as they are not significant anymore and don't cause embarasment like the celrons been faster than the early p3's due to the lack of slow off die cache.
It irks me that no one gives references, making it hard to rule out an "echo chamber" effect. I wrote Intel for "support" on this, and they suggested I execute a non-disclosure agreement. I kid you not. This may be the reason no one is willing to say how they know. But you don't all agree!
Of course, there might be no correspondence between cores 0,1,2,3 in Linux and cores 1,2,3,4 in Windows; someone might have told Microsoft that 0,1,2,3 was a "standard".
References to the article? My own testing with P45 and X48 and previous testing with P35. Talking to industry overclockers and performance enthusiasts and also getting emails from several companies making P45 boards giving me documents explaining jitters, overshoots, vtts, gtl, skews and data eyes. The BIOS on the basic Asus P45 board mirrors that of the DFI X48 and Asus Rampage - compared to the P35 it's leagues ahead.
It's not like you can't overclock with just leaving them alone - you can - but it's the difference between an average overclock and a good one with solid stability :)
I was actually only pressing for a reference for the core numbering scheme. No one can provide one, who isn't under an NDA.
EDIT: Addendum, I've taken out the Nehalem comments until I can find something concrete, but everyone I've talked to in the industry has either echoed what I've said about the 2 SKU approach or at least shown hesitation. I think I'll join them until it becomes more apparently what Intel's motives are, considering it's historically shady nature when it comes to new features. It's a shame FUD posted something the same day, I didn't want to be lumped in :(