0.0 - Core i7 4700MQ @ 4550MHz - 4550 mhz CPU Frequency

[ice.cold]

Wow, that's a nice overclock for a 4700mq. I haven't seen anyone with an unlocked, non-extreme edition mobile 4th gen before. Nice Find.

Edited April 23, 2015 by ice.cold

[Guest 0.0]

Thanks ice.cold

 

Don't know if it's same for all other mobile Haswell CPU's or desktop as I only have the i7-4700MQ to use the unlocking trick on.

[ice.cold]

It looks like you managed to solve your 1.3V limit issue . I actually have an ES version of the i7-4700mq, though I wasn't able to increase the turbo multiplier any more than +2 bins over the default max (upto 3.6Ghz). How did you manage to unlock your retail version? It'd be interesting to see if it works on my ES model.

[Guest 0.0]

Basically there's a bug with the CPU that is fixed by microcode update so if we set the CPU up before the microcode update through BIOS modification then we can use higher bins.

 

It would be interesting to see how many Haswell CPU's are affected this way, including desktop ones.

[ice.cold]

So from the sound of it, with an AMI Aptio IV bios, all I need to do is just prevent the bios from uploading any microcode (forcing the cpu to use the factory provided one), and then adjust the ratio limits under the Advanced -> CPU Configuration menu? Is that all there is to it? I thought it would be more complicated than this.

 

I'll give this a shot over the weekend if that's it. Do you know if this 'bug' was fixed by a particular microcode revision? Or is it just any microcode that's been released with stock bioses since the 4700mq launch?

[Guest 0.0]

At least until after you have set the new highest bin. There's other errata that should IMO be addressed with the update but you can probably operate like that at least for checking purposes. I don't know what BIOS you are using or how it let's you adjust bins and some BIOS choke if there isn't an update so make sure that's not going to happen. You'll need turbo ratio limits unlocked as well.

[ice.cold]

I managed to get this partially working using microcode version 07 (earliest version I could find; from an early lenovo Y510 bios). It seems to be fixed in microcode version 08 though.

 

My bios allow me to enter any number between 0 and 255 for the core ratios, however this bios only sets the registers up with +2 bins over the default, regardless of how high the entered multi is. However, I was able to run an 8 thread load at 36x by manually adjusting the limits under MSR 0x1AD, and the requested ratio under MSR 0x199. It still dosen't go over 36x using these registers, even if they are set higher than 36x. Did you need to adjust any other registers to use the higher bins?

 

Microcode version 08 seems to just ignore MSR 0x1AD. My bios is for a clevo P150SM, and is AMI Aptio 4 based. Judging from the bios string in your CPU-Z submissions, your MSI bios is also Aptio 4 based, so it's behaviour should be somewhat similar to mine.

[Guest 0.0]

My own BIOS comes with very little in the way of overclocking, some RAM adjustments and a buggy iGD OC, so I have had to add them myself. Perhaps it's easier that way.

 

All these adjustments need to be done before any microcode update is carried out, at least from 06 (tested) and I would expect from the very first one. I'm a little surprised that it would work with 07, maybe have to have a look at that sometime. What OS are you using, are you using legacy or UEFI, can you post without any updates?

Edited May 4, 2015 by 0.0

[ice.cold]

1 bricked bios later and a wiring bypass after the VCC solder pad broke off my mainboard, I can now report success! Here's a screenshot of my cpu @ 42x running an 8 thread 7-zip benchmark using the initial microcode:

 

 

The cpu voltage readout for my cpu is wrong without any microcode updates, due to erratum HSM32. It should be double the displayed value (I used a +125mv offset).

 

I was actually able to use Throttlestop 8 to overclock the processor, as the max turbo multiplier increased by 2 every time the FIVR controls were opened after setting the core ratios to the max multiplier. Quite a convenient bug (or feature), and somewhat ironic since it disabled your overclock.

 

It's funny that intel decided to omit this from their errata sheets. I'm a little surprised it wasn't discovered earlier, but I'm more surprised it actually works on your retail processor, as I would have thought they'd use locked microcodes for them. Still a nice find nonetheless, and worthwhile trying on any other ES cpus that seem to be locked.

 

My bios can be crossflashed to one from a P170SM, which includes hidden settings for voltage control and overclocking of the cpu, igpu, ram etc (some options do brick the laptop), though it does honor the programmed limits and refuses to set anything beyond 36x. Hard-coding the values dosen't work either. I think it would be easier to setup the cpu multipliers in a pre-boot environment, which could also be used to update the microcode. I don't use UEFI, but it should be possible to script something for the biosbits test environment. I am having trouble finding programming documentation for overclocking haswell processors though. Maybe I'm not searching for the right terms.

Edited May 14, 2015 by ice.cold

[Guest 0.0]

Sorry to hear about the bricking and lifting a pad really sucks but happy you have got it working.

 

Yes, the CPU VID reads half without a microcode update so your 0.566 is actually 1.132V. If you apply the microcode update in the OS you will see it magically change to the correct reading. Also IIRC VCCIN override above ~1.795V does not work without the microcode update.

 

Once you have your values set for OC to how you want them you can apply the microcode update and they will stick unless you attempt to write to the OC registers. TS8 currently seems to do this (write to the registers and kill the OC with mc update set) but the author did hint he might change the way it works in beta 2 to accommodate. IIRC without the mc update TS8 limited my max bin to 36, I didn't play around with it too much as I used my own crude software for OC'ing. I think the TS author would be very happy to see your report and screenshot.

 

W8 has it's own mc update that can be applied when OS starts, there might be a hotfix one for W7, not sure. When you say you are not using UEFI do you mean using CSM? I was simply going to give you a rough UEFI program to run for testing OC with say fixed 1.2V if you were having trouble, which could be run before the OS is loaded and OS provides mc update.

 

I'm not really surprised it's not mentioned in the errata. A little bit like a particular memory mapped register being falsely described as an iGD control when it is really used as a maximum core multiplier limit.

 

It would be nice to know how many HSW chips are affected but without people such as yourself willing and capable of testing it then it's hard to say. Perhaps it got fixed in the refresh, maybe not. idk.

 

 

I would be interested in how far you can push the frequency. My own chip seems to start scaling poorly after about 4.2GHz and badly after 4.5GHz.

 

Edit: Tried your trick with TS8 and works the same way here too, nice find. If you want W7 to update the microcode for you after setting OC pre OS boot then Windows6.1-KB2970215-x64.msu is what you want to give update 1A. Never seen any doc's for what you are looking for. All I know (not a lot) is empirically derived with some hints here and there and going by that MSR 0x150 is where it's at if you want to control the FIVR and to some extent maximum multipliers.

Edited May 14, 2015 by 0.0

[ice.cold]

The bricked bios was no big deal really, I've had to take the chip off before and I was able to do that without issue, though I guess it can only be desoldered so many times before something breaks. Luckily it was just the 3.3V line, which can be easily sourced elsewhere from the board. Actually removing the microcode had nothing to do with the bricking - ironically the recovery module decided that it would act out a scene that could have been written for mission impossible, by erasing it's own boot block and then abruptly shutting itself off without warning.

 

I got this multiplier upto 47x with ~1.31V, attempting to set the voltage even a little bit higher resulted in lockups; so it's giving me the same issue you had, probably due to the VCCIN bug you just mentioned. It's likely not stable at this freqency and voltage either, it will probably crash if I try using loads other than the TS8 benchmark. In terms of binning I'd say it's similar to your chip.

 

 

Thanks for the MSR tip, though all the bits from 47:0 seem to be read only, and the rest just crash the laptop when set. While monitoring that register and using TS, I did notice that bits 7:0 briefly showed the maximum turbo ratio when I pressed the reset turbo button, but anything I try write to it is ignored. I think the maximum multiplier is set elsewhere, unless I'm missing the elephant in the room. The Intel Turbo Boost 2.0 Design Guide is what I was trying to find, though it's only avaliable under NDA so I don't think they'll be a public version anywhere. I'm sure you have a much deeper knowledge of the configuration registers for these cpus than what I do, even if you say you don't know a whole lot.

 

I also tested the TS trick with microcode 07, and the max turbo dosen't increase anymore, but the 4C ratio can be pushed to 36x like I posted before.

 

I am using CSM with an MBR partitioned drive, but still send through the EFI file, I'll see what I can accomplish with it. Actually I tried to test the microcode update facilities in the biosbits environment, but it seems to read the CPU platform ID as 00000001 instead of what it should be, so it won't upload any new microcdes for my cpu. I'm not sure if it's a bug in the program or if it's due to my ES processor, so I'll either need to patch the platform check out (to upload microcode version 1C) or just use the windows update. I think 1C disables the buggy TSX instructions, and probably should be used if possible.

 

Update: Just doing a little more research into the topic, there have actually been discussions about bios mods for mobo's with Xeon E3-1231 processors (cpuid 306c3 - same as 4700mq), which say that they can increase their 4C turbo multi's upto the 1C max, by using microcode version 07. These processors might also have the unlockable multipliers.

Link: http://www.win-raid.com/t785f16-Discussions-UBU-Tool-related-Questions-Reports-and-Suggestions-54.html#msg7914

Edited May 16, 2015 by ice.cold

[Guest 0.0]

Have you tried disabling OVP to get above 1.3V? I think you should get better performance than my chip but not sure why your TS Bench is slow for the clocks.

 

I'm a little busy at the mo but let me know what multi, voltage and mode (fixed / adaptive) you want to try and I'll try to put something together.

 

DT chips would be interesting to try out. Perhaps though they need the ratio overclocking feature that some of the mobiles have for using limited unlocked bins. Lots of unanswered questions at this time.

[ice.cold]

Yep, the OVP/FIVR faults was the problem, when I realized that I found that my processor had already hit it's clockspeed wall; trying 48x at 1.5V still occasionally locked up, and I don't think this chip will hit 5Ghz unless I rig it upto some impromptu LN2. The benchmark scores do improve with increasing clockspeed, and there dosen't appear to be any throttling. I don't know what TS scores you get, but mine could be lower possibly because of calculation errors during the test from too low voltage, and possibly also the fact I was only using 1 stick of 1333mhz CL9 ram at the time.

 

Don't worry about the efi program; I thought you were just going to send me something you were already using. I figured out a way to set the multipliers before the OS boots .

[Guest 0.0]

My software is not user friendly, I just recompile with fixed values. Quick and dirty.

 

The mobiles seem pretty much like their DT parts except no IHS and more importantly less pins for the mobile PGA so should try to keep the current down I think.

 

Probably it's that single channel DIMM hurting the score, here's my TS8 beta 1 at ~4.2GHz for comparison.

[Massman]

Bumping this thread so people can get the full story

[kimgnguyen]

Sorry to dig up such an old thread, but you wouldn't happen to have a copy of the pre-tsx fix microcode ?

[kimgnguyen]

Edit to my previous post: I mean, microcode update 7

[tmash]

I have made a guide: https://www.techinferno.com/index.php?/forums/topic/9636-4ghz-i74700mq-hack-intel-haswell-cpu-microcode-bug-hack/