M15x Nvidia 470M Mod

Yeah, this bests even the greatest of soap operas.

 

Wrong.

Just google "ERROR: MXM Structure not found or invalid". That's the error people get sometimes when they upgrade to a GPU that isn't supported by the BIOS, their POST is halted for 30 seconds and performance is limited to
balanced performance mode.

Here is proof that it's BIOS incompatibility:

He flashed the 280m vbios (unsupported GPU) onto an FX 3700m (supported).

I'm not saying it is a bios problem in this case, but it could be.

 

Is the 460M GTX that much different than the 470M GTX that the M15x wont work properly right off the bat?
You would think the new BIOS would work with both cards, if only it was a simple swap.

 

To get Riva Tuner to accept the card, go to the Riva tuner install folder and open RivaTuner.cfg with notepad.

Scroll down to the Section labeled [GPU_10DE]

At the bottom there is a listing for G100 right?

Since the GTX 470m is GF110 based, you should add a line underneath that and call it:

"G110 = (enter your device ID here)"

save the file and exit. Then open up rivatuner and see if it is recognized.

If it isn't recognized, then you could try adding your device ID with a comma before it on the G100 line of the .cfg file.

If that doesn't work, I would take out your card and look on the physical die to see what the device ID is. It should be listed there. If the device ID isn't the same, try that one.

hope you could follow this. I'm horrible at writing directions.

 

Double Post.

Might be kinda dangerous, but you could dump the GTX 470m bios and set the 2D settings to the 3D Performance settings and flash it. . . . I also did that to my 4870's and I haven't had a problem.

You'd have to make sure you copy the 3D performance settings exactly - including the core voltage, and shader,mem,core speed.

 

It's actually GF104 based.

 

/\ Good catch. So replace "G110 = (ID name here)" with "G104 = (ID name here)"

 

I suggested that before, but he said nibitor doesn't support the GPU yet so...

 

Also get in touch with frynight, he might be able to get it working as his university team has managed a 480m in his M15x!

 

Wrong. It little known, but MXM 3.0 standard have two revisions by now - 1.0 and 1.1 (confusing, eh?)
MXM defines interface between SBIOS and VBIOS, partially on the account of managing the "power states", like changing clocks. I am not sure, if revision 1.1 updated any power-related interfaces, but this is entirely possible.

Then, say, if SBIOS is done for revision 1.0 and VBIOS for new card is 1.1 this could cause subtle errors like this - requiring updating SBIOS for the full support.

 

Hmmm...this may be possible. The mxm-sig website does state revision 1.1 of mxm 3.0 type b July last year so this could be the case that our motherboards are still 1.0. Bit of a kick in the teeth really.

Edit: maybe this could help, a few days ago I had a few problems oc'ing my 260m with evga, what ever I did new (higher) clocks wouldn't stick and everything ran at 2d clocks even though my sliders were at 3d. Restarts and drivers wouldn't fix it until I just turn stealth mode on then off again and I was able to run on full blast. This would happen every now and again where the nvidia card would think it is in stealth even if it wasn't!

 

Someone on the forum (Mexicansnake I think?) mentioned that AW support told him that M15x's that ship with the 460M have a newer revision motherboard and that the older ones are not compatible with the 460M. Don't know how much truth there is to it but if MXM 3.0b is a factor, it could be why it won't work on Inap's system.

 

Well some guy got a 460m working in his M15x that came with a 260m and it was almost as easy as plug and play so I don't think it is a motherboard problem.

Edit: Come to think of it people have different motherboard revisions because some have had problem overclocking with setfsb. Wonder which one inap has.

 

There are different revisions of the mother board for sure but from everything that was discussed in a few different threads it's nothing physically different or limiting, it's small things like the pll. So far there's only been 2 mobo's identified across all the revisions 1 that is OC'able on SetFSB and 1 that isn't. Both of those mobo's were out before any of the Fermi talk so there shouldn't be an issue at all physically.

Edit: The way this was verified is there have been 1 or 2 people who have had an upgrade done to the 460m by Dell using the revision 12 mobo and had them working just fine, the Rev 12 mobo has been shipping in systems since before the 460m was even being discussed

 

Sweet thanks for clearing that up. Still hope inap gets it working though...

 

Yes, that's good to know... so at least the 460m should work. I really want to upgrad my 260m in the next couple of months.... Though 5870 is a good option, I'd prefer a card with Cuda... 470m would be so great!

 

took a break last night, getting back to it. thanks for all the suggestions. so far only the 263.09 works but no 3d yet. everything else including sager drivers would either not install or would cause bsod upon reboot. yes i tried the have disk method and still some drivers will not install.

 

Mind bringing me up to speed, on why you expected the 470M to work in the M15X?

This machine has no GF104 support, right? I'm surprised it even booted.

 

MXM standards are not chip-dependant, in fact whole purpose of MXM is to enable correct interoperability of any MXM video board with any MXM system board. Therefore it's perfectly fine to expect possibility of things working
Problems often arise when manufacturers a) don't implement MXM fully/correctly b) use propietary extensions and c) due to different versions of the standard

 

didn't really expect anything, just testing since the 460m was supported and many people here wanted to know if it was possible.

^^^^thanks Ingvarr.

 

perfectly summarized!! it really sucks that there's a standard which actually is great, but implented in not such a great way...

Does a 460m of eruocom/clevo work in a M15x? Because if it does it's really crazy that a 470m doesn't imho.

 

ichime is using a clevo 460m in his setup.
alot of support for 460m and 480m, but 470m doesn't have much support atm.

 

Honestly I'd say it's probably just a matter of time, the 470m is probably the newest card on the upper end laptop GPU scene and as such has little to no support. This has to be a Vbios or driver issue. I'm wondering if it's a custom Vbios that's needed with this, have you tried to extract the Vbios to look at it and see if there's anything proprietary in it yet Inap?

 

All the MXM standard guarantees is that they'll fit in the module. That has little to do with plug and play interoperability.

There's little precedence set for random cards working plug and play, outside of core families. Any G92 likely works in a notebook slot with G92 support, but for GF104 to work behind a GF106, a BIOS update is requisite.

They're too different to reasonably expect the 470M to work properly.

This is 100% an issue with the M15X BIOS' support.

 

But if it were a BIOS support issue I would think one of two things would happen

1)absolutely no post as the M15x doesn't have another GPU to default to to continue the boot sequence
-or-
2)the card wouldn't load at all not even on 2D clocks

So if those are true then it's back to the driver or the Vbios, and I doubt Nvidia is going to release a card to the public or even a vendor without a driver that will at least work(even if it's not the best) which would then lead to the last culprit the Vbios.

 

here is the vbios, if anyone wants to check it out:

Download GF104.bin from Sendspace.com - send big files the easy way

 

I repeat, all chip-dependant functionality should be contained in the VBIOS, exposed to SBIOS via chip-indenendant MXM tables.
Problems arise when manufacturers cut corners and put chip-dependant functionality in the SBIOS, what they dont suppose to according to MXM standard.

I can not post NDA-covered documents, but this public one summarizes core of interoperability pretty well - http://www.mxm-sig.org/file.cfm?doc=CF89E942-F4B8-7D82-089893AFA60B59E4

It is possible that particular 460M works just because it uses different VBIOS, with more standard adherence, or just older revision, more compatible with particular M15x SBIOS.

Another possible culpit - MXM allows some "manufacturer-specific pins". If particular card uses such pins (customized for particular vendor), this could make it non-interoperable with pure-MXM or different vendors.

 

really? Alienware is doing a system exchange for me but they upgraded my video card to a 460M (laptop originally came with the 260M). I should get the new M15x within a week and I will find out if what AW support said is true.

 

Well the temp is good!

 

i keep coming back to the possibility that it's really just an .inf file problem. Based on our prior experience with the 5870's, the only time the Subvendor was identified in GPU-Z as Dell(1028) was when a Dell card was installed. When a Clevo 5870 was used, it was reported as a reference card -- ATI(1001). But with Inap's new 470M, it's being reported as a Dell card when we know it is not. The modded .inf files include only the Device string (10DE-0E30) for the 470M, with no subvendor strings. That usually works and let's drivers be installed without regard to the specific card. I wonder where the Subvendor ID is coming from and whether that is interfering with the operation. I'm out of my element here, but that Subvendor ID feels wrong.

 

^^^^Ditto, the other thing throwing me for a loop that I may just be wrong on are the clocks. The core clock is at full speed for the 470m on the settings, but the shaders and ram are a decent amount lower is that due to the manufacturer setting them that way or is something wrong with either the reporting or vbios?

 

hmm did you try disabling CUDA and PhysX(and anything else extra nivida has on there) inap?

 

yeah everything is friggin screwy, heheheh here is another screen shot with nvidia inspector and gpu-z. inspector showing 1500mhz for memory and gpu-z showing 750mhz.

 

Different voltages too.

Can't open that vbios file. What are you using?

 

i used gpu-z to save it. and using NiBitor to open, but it not working correctly either, it show card not supported and the info is all screwy.

 

I understand what the specs say. But the precedence has been set, and it says otherwise.

How many high-end gaming notebooks (AW, Clevo, MSI, etc..), over the last 5 years, have accepted an MXM chip which was unrelated to the chips that manufacturer did support? The answer to that question (next to none) is the precedence from which one should extrapolate the chances of a new, unsupported GPU properly functioning in their machine. The theoretics of the MXM spec rarely hold up in the wild.

 

AFAIK Radeon Mobility HD 5870 was used in platforms designed long before it was out (and only for 2xx chips in mind), so this sets pretty good precedent.
I don't really know of other good MXM 3.0 examples (full BIOS interoperability was one of goals of 3.0 standard, earlier 2.1 don't count).

 

Is there any other way to extract the vbios other than GPU-z, I know you can save a backup using RBE and a few other ways with ATi cards so there has to be another way to extract it, I'm getting the same exact problems when trying to open/modify it using no less than probably 5 different programs.

 

waiting on a new driver atm, but everything is pointing to the M15x bios. sorry guys things not looking too good.

but here is something strange, a screen shot of gpu-z of a 470m in a clevo. showing the same memory clocks as mine.

 

memory clock should be at 1500Mhz according to nvidia, 750Mhz is the half, but as it is Double Data Rate, doesn't this sum up to 1500Mhz?

 

yeah i think that is why its reporting it like that.

 

Yeah its the way the bandwidth is calculated. Its like dual rank or summat like that that reads half of what it should be. I was wondering if I should start a thread at the nvidia forums about why there isn't 470m driver support?

 

yeah it really sucks that there aren't more drivers our there to try. gonna gave it another few days, but things don't look so good.

 

Well put. In theory all MXM III B cards should work on all MXM IIIb motherboards, with the given acception of cooling and watt consumption being more than the PSU can output

 

Dumb question, do you still have any of the other cpu's you had Inap? Probably not a power consumption issue but it may be worth a shot to throw a lower consuming cpu in there to see if it opens up the 3d clocks to narrow down a 940xm/470m problem. If it's too high of a power consumption the bios may have a heierarchy of what to change first, doubt it, but it never hurts to try.

 

From some pages back I put forward maybe contacting frynight as he and his team moddified the bios to work with the 480m. Have you tried physically reinserting the card? Cleaning the connectors etc.

 

naw i sold off my extra cpu's, kinda wished i didn't. but using your 240 psu with it. but i think the power shouldn't be a problem since clevo/sager is using the same psu.

just outta idea's right now, waiting on a reply from fynight but i think this mod is a no go without a M15x bios mod.

yeah i reseated the card a few times, killed off a new tube of ocz freeze already, the die is frigging huge, hehehe

 

But the 240 psu isn't BIOS supported so I doubt it's gonna pull any more than about 150 is it? If you're brave enough there's a couple i3's over on ebay for about $75, wish I had a cheap CPU I could send you to test with.

 

man this card alone pretty much tapped me out, heheheh
if this doesn't work in the next 2 days, then i'm returning it.

 

Yeah if we compare your GPU-z shots with one from the w860cu:

We can see some subtle but probably significant differences. However, the memory readings are the same.