920XM records broken with the old M15x! Success!

I have managed to iron out instabilities on the new chip. A few of you have been asking how my 920xm is faring so I thought I would try to break some records. I have shattered the records. I have managed foremost to attain 32x at stock BCLK 133mhz for a total 4.256ghz across all cores very stable under load. 33x results in a instant BSOD under load.

For an hour or so I have played with the BCLK with varying results I have managed to bench at sometimes over 4.3ghz and other times (mostly) crashes. Considering 32x is rock stable I see no need for anything else. I did however manage to submit a record clockspeed of 4.52ghz if you look below and click on it you can see full detail.

This means I finally knock SVL 7 off top spot for wprime 32m time and highest clockspeed record. It seems difficult to register at hwbot but I will figure it out.







Now it is time to fine tune this chip and eventually get some water cooling planned

 

My 3630QM @ 3.2 GHz is still faster.

 

You can't touch my wprime score though. Best score there is 7.3 secs. It really depends on the test.
Due to added instructions and efficiency the 3630qm can win at times but other times the 920xm does pretty well to keep up and sometimes even beat it.

Even in passmark another benchmark that makes use of the improved instructions etc on newer chips your chip is only a mild 4% faster at 7600 points vs the 7300 I managed yesterday.

Doesn't sound like much but for a chip that is ancient by computing standards not bad for it to stand toe to toe with chips 2 generations later.

As a platform that can support even the 980M it is likely my M15x will stay out of the trash longer than the Y500. This is truly the point of this demonstration to prove that the platform can be tweaked to greatly reduce bottlenecking of high end GPUs. To have 3630qm benchmark power and more raw horsepower than perhaps even faster chips is a testament to how badly Intel have advanced these past 5 years.

 

Were you able to do this with just the pin volt mod? I know you don't have stock cooling but how much do you think you could achieve on stock cooling? What kind of temps are you seeing under load? Well done sir.

 

On stock cooling you cannot even attempt the pinmod. This requires additional cooling. On stock cooling you are much better off running at normal voltage as you would be able to hit higher stable clock speeds

As for temps under load. I have time now for some tests. I will attempt 1024m throttlestop that represents a moderate load but not nearly as strenuous as Prime 95 that I dare not do lol. At 32x.

Hmm for anything longer than about 30 secs for full load the system halts on BSOD. I got flashing scroll and num lol. Temps were about mid 80sC so I think the VRMs are getting too hot. Mike's watercooling would do wonders!! This in ambient 23C.

I could put the thing out side in the garden where it is about 7C and it might finish the test but what would be the point other than to prove I need better cooling haha

 

What cooling solution do you use?

 

Well my 3940XM @4.2GHz will blow your balls out of the water so I don't even need to do a bench .. Also King of Interns won't be a match for me .. But great job getting this much out of a 5 year old laptop .. I salute you

 

Bravo well done mate...........i would not even dare too try such core speeds but than again i am not a benchmark guy 27 stable is good enough for me.

we are proud at our King :thumbsup:

 

Well done King! I'm surprised your M15x hasn't exploded in a shower of tiny black sparkles! 32x on four cores is beyond insane! Imagine if you had CAS 7 1333Mhz RAM too!!!

 

Impressive work! As for the newer chip guys - get yours to over 100% overclock, then chime-in

 

Pretty freaking difficult when they're already much faster and closer to the ceiling to begin with.

 

The overclock ceiling is almost always one's cooling

 

And power. In that sense, newer chips clock much higher for the same power and cooling. I was simply pointing out that going by percentage increase alone is misleading if the baseline is so low to begin with in the case of the 920XM. Imagine if I had 1.6 GHz i7 and overclocked it to 4 GHz...ZOMG 150% increase!

 

Or voltage....also octiceps sands and ivy and way more efficient. The 920xm was clocked quite low but the best they could have done was 2.5 as a baseline.

 

I'm most impressed/surprised that your laptop isn't on fire yet, knowing what a desktop 920 is capable of. 4+ GHz is your benching OC, there's no way you're running that 24/7 and during gaming with simultaneous stress on both CPU and GPU.

 

Nice work King Of Interns, I also have an old m15x that I am thinking about tinkering with. I read your fan mod. What is the pin mod that is being discussed?

 

Why is it misleading? 920XM is a hot piece of silicon, a REALLY hot one. I can run mine @2.7GHz across the cores, but that's for gaming. Running wPrime and the clocks hit 2.1GHz in an instant (insufficient power limits) and temps go up (that's why they are insufficient ). Granted 8740w doesn't have cooling to write home about it, but even in my brother's M17x-R2 with it's massive fan and radiator the CPU wasn't of the coolest running @3.6GHz. So yeah, it's almost always the cooling and of course voltage, but one can't apply more voltage if there isn't proper cooling

As far as the new chips go, it was always been like that - newer, faster, more efficient or at least up until now. Haswell wasn't very impressive and rumors say that Broadwell would be even less. Anyway, feel free to overclock yours to the same level, as you said, you'll need less cooling and less voltage for the same achievement

 

It's misleading because the percentage increase alone sounds more impressive than it actually is compared to more recent unlocked chips. And saying "oh Intel has made no progress last 5 years because CPUs can't overclock by over 100% anymore" is even more out of context.

And about the power and heat, that's exactly right, which is why I'm extremely surprised that KoI is able to clock his 920XM past 4 GHz and not burn his laptop/house down. 4+ GHz is stretch even for desktop 920's, and that's with high-end dual tower dual fan air coolers or liquid cooling.

I wish my 3630QM could overclock, but the i7-x6xxQM are completely locked. Actually I don't, because my CPU and primary GPU share a fan and CPU cooling isn't great as it is.

 

I remember my i5 750 going to 3.8 with all power saving features on with 25£ xigmatek aegir, not too much volts needed, something like 1.25 which was just a tad over stock. For 4ghz i neede to either kill powersaving stuff or bump voltages, over 4ghz some serous volts were needed. Around 1.5v for 4.5ghz which ended up being aroud 200w cpu power draw.

Over 4gh they chug trough power like those amd fx space heaters but it was good times.

Knowing what i know from desktop about overclocking that generation this is really impresive achievement.

Knowing that i had mbo with 12 + 4 phase design i wasnt afraid as i found it was able to provide 275w to cpu without sweat. But this kind of power draw on a notebook mbo which usually have 3-4 vrms would be scarry for me

 

Of course, in recent times it seems that everyone is talking percentages, and make even few millimeters differences in laptop's dimensions look huge. On the other hand it was always taken into account when it comes to overclocking. So again, why is it misleading? I can see your point if we talk about thickness of given notebooks. Unlike whether or not you'll feel those 2mm (OMG 25% difference, LOL) in thickness is up to you, but overclocking and benchmarking were always about numbers Also we work with what the manufacturer gives us, it's not like I said "Hey Intel, I want 500%-or-something overclocking CPU", now did I? Am I clear enough now? Also there are very few CPUs capable of getting those 100% at all, let alone on air

I feel you there, my 5920G has the same setup, quite the pain, but I upped its game by the end of the project. There's always room for improvement, it just takes "some" time and "some" money ("some" = a lot)

 

There is no luck in it guys. This is a special 920xm.

I cherry picked it and reserved it after gotmilk shared that he got it to 3.6ghz on stock voltage all cores under prime 95 stable. With 200mv on top it is pretty much guaranteed to make at least 4ghz across all cores.

The VRM's are the limiting factor here. If I can cool them properly I could probably run at 33x and perhaps even 34x on 1 or 2 cores all day.

As for all the negativity from octiceps it is a shame. The 920xm doesn't actually share it's history with the 920xm it is based on the slightly newer Lynnfield I think. It isn't a mobile 920 desktop even if the nomenclature might suggest it does. I am running this 24/7 with 32x across all cores. I don't run it at 100% much it can't run for long as I already stated at 100% but it is stable for the most part including games as they don't press the cpu to 100% for any length of time especially at such a beasty OC.

Anyways the 920xm is heavily QPI bankwidth limited. If it shared similar bandwidth numbers with the desktop 920 at a whopping 4.8GT/s it would absolutely smash the 3630qm at the clocks I am running no offence. Unfortunately it is gimped at 2.5GT/s which is why performance is limited. Intel haven't made much progress in terms of performance in the last 5 years. AT ALL. Only with efficiency.

The reason the newer mobile chips seem to perform so much better is due to much increased bandwidth. Otherwise there isn't a lot in it clock for clock perhaps nehelem to haswell up to 40% difference. Just check cpu mark the i7 970 at 3.2ghz get 8500 points. I get about 7300 at 4.2ghz. Hmmm imagine I had 4.8 GT/s to play with now......3920xm suddenly wouldn't look so good haha!

As for power I can combine my 220W and 240W(330W adaptor) adaptors together when I need it. I have plenty of power thank you.

 

Bro, I'm not being negative. I just think you're being disingenuous for suggesting that Intel has made no progress performance-wise the last 5 years, esp. when you haven't used any of the newer stuff. Efficiency = performance, even more so when laptops are concerned. I won't argue over it; the real world results all over the web bear that out.

And you're absolutely right, although they're all Nehalem architecture, Clarksfield is based on Lynnfield, not Bloomfield like the desktop 920. But Lynnfield and Clarksfield use DMI like the Sandy Bridge and later processors, not QPI, and perform just as fast as Bloomfield while consuming less power.

 

I did a little research and it shows that QPI on the bloomfield processors feature considerably higher bandwidth vs DMI. I don't know what this translates to exactly but the bloomfield chips scored considerably higher clock for clock than my clarkesfield chip. Other than the bandwidth I can't see a huge difference in architecture.

Perhaps the 8500 score on the 970 is OCed but I doubt it. If it really scored that at 3.2ghz then it would be at around 10K at 4.2ghz or more just like the 2900xm and 3900xm at 4ghz+

Of course at much greater cost in terms of power and heat. Efficiency does mean performance in that you can push closer to the ceiling. However my point is if you push the nehelem close to their own ceiling with proper cooling of course the difference is not as great as it should be. By this point haswell should be at least 75% faster clock for clock IMHO.

 

It doesn't translate into anything for you. QPI is designed for ultrafast communication between different physical CPUs in a multi-socketed system or for lots of high bandwidth PCIe devices. Makes sense as X58 is the enthusiast/server/workstation platform.



If you’ve only got a single-socketed system and aren’t running a lot of high bandwidth PCIe devices, then QPI is overkill. So why not get rid of the X58 I/O Hub & QPI and put PCIe on the CPU die itself? One less chip on the motherboard means cheaper boards. Great idea, huh? That's exactly what Lynnfield and P55 are: A cheaper, just as performant alternative to Bloomfield and X58.

===>

You can look up any number of reliable benchmarks (keyword: reliable, as in not PassMark) from reputable sites (e.g. AnandTech, Tom's Hardware) and they will show you that, outside of rare instances where Bloomfield's triple-channel memory edges it ahead of Lynnfield's dual-channel memory, Bloomfield and Lynnfield/Clarksfield perform identically clock-for-clock.

 

Can i ask about the temps ?

What's is the max temp recommended to 920xm?

While idle~~ or in internet navegation i'm always got 65+ about 65 to 70 max.

While playing got TOO HOT 80+ ALWAYS, some games i got 90+ . Is this normal?

with 740MQ i never reach 80c while playing.

 

Until I get water cooling sorted I haven't got a max temp lol. Guaranteed it hits 100C in no time though at full load.

The 920xm can operate at 95C all day without issue at full load. I have done this many times in the past with video encoding and the like. The 740qm doesn't compare because it is MUCH less powerful. With proper OCing the 920xm can double the performance of the 740qm.

Anways 80 - 90C is totally fine. If your max temp hits 100C though it means your chip will throttle and you get less performance. Therefore you want the max temp to be maximum 99C.

Not joking the 920xm is a TOUGH cookie.

 

Something like this you can forget with the new famous Alienware15 with soldered hardware and crippled power supply. HaHa

 

True that.. That's why my next laptop is the Clevo P770ZM... i7-4790K FTW

 

Especially if it's binned.

 

Think about this !!! That new Alienware 17r2 is already outdated. This autumn or winter comes a new generation hardware and software .. New faster processor generation (maybe Skylake, gtx1080m and windows 10). Itis guaranteed to be changes .. Alienware 17r2 live in overtime already ..

 

Hey guys, look that.
I really really dont know what to do, never can reach best performance for my 920XM cus it just looks like a stove.

Please, help me D:

 

You need to change the set multiplier tab to "turbo" and then alter the TRL/TPL settings to OC ie increase the turbo multis.

Currently according to the pic you have the max multi locked to 17x. Once you change this your chip will behave normally don't worry.

Rubbish! The M15x is much older and you simply need to put a 970M in and OC the XM and it will perform fine in games.

 

i dont have turbo option on set multiplier tab, just from 9 to 15 and the last one 16T. i have locked to 17x cuz if i set more than this it get too hot like 100c and just turn off m15x.

 

Hmm I guess it must be the version of throttlestop you are using. 16T must mean the same as turbo on my older throttlestop.

Yes the chip gets hotter but you should be able to run the chip at 24/25/26/27x easily in that order from 4 - 1 core active on stock cooling and keep temps under 90C.
Open her up. Repaste with ICD 24 and try again perhaps.

 

Yep I'm trying Throttlestop 6 on Windows 10 and I have to set Multiplier to x 17 T. Same performance as Throttlestop 5 on Windows 7

 

What is ICD 24 ? I repaste it but i dont get nice temps, always when i'm in gaming i got 90~~98c, if i try top games like, cryses3, daying ligth and etc my m15x ALWAYS TURN OFF cuz its too hot..
i dont know what to do anymore! D:
Kings of Interns can i know how is you water cooling mod for m15x?
Anyone know any mod for coolers and etc? - PS i dont want transform my 15x into a desktop, so any easy thing to do.

 

Look up M15x hybrid cooling mod. I hope to copy something like that.