P150em - trying to get 980m working, Prema mods gone

Hey!

I've Googled the absolute heck out of this. Lots of talk of overclocking, excitement of upgrading back when it first came out. Lots of threads from 2015 and excited people linking to Prema mods site, but of course nobody attaching it to the thread or rehosting it... I guess nobody expected Prema to just pack everything up and disappear from the internet. Also pretty awful that archive.org doesn't/didn't keep the site around, too...

I've got a P150em, a rebuild project from a blown power control IC and a bunch of other issues. Its GPU was fried, think it was a 670 or something, giving that nasty Code 10 error. Took a deep breath and splurged on a 980m after a fair bit of research. But the depths to which one needs to crawl down this rabbit hole apparently are deeper than I was able to go...

It don't work. It's identified, but it presents itself with a corrupted SUBSYS ID of 51051558 with the latest (17) stock BIOS. With the Eurocom Prema mod (I ain't gonna say where I got it, but it's nearby and pinned!), it got a SUBSYS ID of a similarly useless 52811558, matching that of a GTX 970m, but with the 980m's device ID.

With Prema "1060" mod BIOS (the only one I've found): PCI\VEN_10DE&DEV_1617&SUBSYS_52811558
With stock BIOS: PCI\VEN_10DE&DEV_1617&SUBSYS_51051558
Of course, being as though the subsys doesn't jive with the proper ven/product IDs, it doesn't install, and this will now be the only Google result for these IDs.


Also pulled the VBIOS but the Vendor ID in the ROM doesn't match any part of the SUBSYS it's presenting, so unlikely that has anything to do with it (though it is an MSI card):



I might end up ripping apart the BIOS in UEFITool and comparing sections... but dear god, if the work's already been done and there's an easier way, it'd be nice to just do that.

I was able to get the driver installed and initialized by modding the INF and disabling driver signature enforcement, but ... the card wouldn't function when called upon. It'd end up rendering with the Intel GPU (Furmark) or just plain freezing (CUDA compute apps), without even starting. That's when I reflashed the stock BIOS.

Where can one even begin looking? I'm pretty far down this crazy rabbit-hole hell, and I really just want stock performance (not a fan of overclocking).

 

aaaaaaaaaaaand now it's bricked, tried to patch that hard-coded ID in the modded BIOS using UEFItool, so I can stop using driver signature enforcement override (can't play games with that, anticheat refuses to run with DSEO). Flashed it, now it's just no-screen, no POST. AWESOME.

Why can't it just honor the card's subsys ID in the VBIOS?! On what planet is it OK to override the card's identification with something from the BIOS?!

Fn+B, nothing. LEDs on (unless I hold Fn+Esc, then just the power LED flashes orange, doesn't turn on). Fan spins, but no evident POST sequence. Just dead. Gonna have to pull off the ROM chip and read/flash it externally. I died a little inside tonight. Prema seems to be Clevo's saving grace, everything always links to that blog or those mods... now that it's all gone, it's a barren wasteland of obsolescence and dead hardware.

 

Likely for driver compatibility.

 

Prema is still around, but he got feed up with people stealing is work..

You can @Prema and he will get a ping to this topic.

 

I can relate and definitely see how that could happen. I'd want to just keep old hardware alive and out of landfills, especially powerful stuff like this.

I've ordered an SOIC8 adapter for my programmer, so I'll pull the ROM chip (thankfully, it doesn't require a full disassembly, just below the MXM card) and see if I can compare/patch it back to the proper modded version. I only changed 2 bytes, sheesh... but being compressed, who knows how much that'll change. Hopefully it doesn't change offsets and ****. Optimally it's just a flat-written binary (1:1 as seen in my ROM file) like the desktop boards I've done.

I'd still need help from the one and only @Prema to get the subsys ID tagged correctly for this thing... apparently you can't just flash a Clevo VBIOS onto it because it's a 4GB card and there was never a Clevo 4GB 980m. Maybe that's why it's applying the wrong subsys ID... but given that the subsys ID it's getting is one for a 970m card (according to the nVidia INF), I feel like that's maybe a hard-coded thing that could be easily changed. I did, after all, find the exact full 32-bit subsys ID (little-endian packed) coded into the ROM, which is what I tried to patch...

(hell, I'd even be happy with the 980m card handicapped as a 970m for power constraints and whatnot.)

edit: compared the original (Prema mod) file & my UEFItool modded file... same size, but the last ~600kb of the ROM file is completely different just with that 2 byte change. Fn awesome. so this'll definitely be fun to put back together, sigh.

 

And we're back

Pulled the BIOS chip (16mbit), found it was supposed to be the first 2MB of the ROM file... it was nothing like the file. Gibberish. Reflashed it as 1:1 of the file contents. Then, with it reinstalled, the behavior changed... now the lock LEDs would flash and fans would roar full speed after about 10 seconds from power-on. Tried a USB flash drive, but no activity LED flash, so I assume it wasn't doing recovery (plus also not clear if P150EM uses UEFI-based recovery or bootable DOS-based recovery).

So then I tore the whole thing apart, pulled the EC chip (32mbit), ran a diff against the second ROM file (4MB), found that it had differences at the top and at the end - but the differences at the top seemed to contain NVRAM data - things like serial and MAC, UEFI boot devices, etc. So I diff'd the original and my crap-patched version, found only differences at the end, and went back and patched only the far address space with the original file. Reflashed that too.

And it POSTs and runs normally again

I'd still super appreciate a hand from @Prema getting the 4gb MSI 980m properly recognized without needing driver-signature bypass and a modded INF, but for now, I'll live with it, and if I get tired of it, I'll sell this 980m card and get a 970m card known to more likely work properly (has a flashable Clevo VBIOS).

Whew.

update: Finding that the 980M just doesn't seem to work right in this PC. Seems to refuse to run 3D rendering, but GPU compute seems to work. Maybe takes too much power. I'm researching swapping it for a 970m and all my problems should go away like magic. One can hope. This might just be the end of the thread, then

 

Cards are normally plug and play for this IIRC, what card exactly did you get?

 

Specifically this: https://www.ebay.com/itm/Nvidia-GTX...e=STRK:MEBIDX:IT&_trksid=p2057872.m2749.l2649

From an MSI. Seems something is weird about being 4GB.

 

Yeah the 4GB card was a bit odd IIRC.

 

Yeah, since I can't get it to run any 3D operations, and it seems to put a HUGE strain on the power system when running GPU compute with full CPU as well (folding@home), I'm going to swap it for a 970m. Just pulled the trigger on an MSI 970m 6GB card locally, seems like an identical board to many other offerings shipped via slow boat (so at worst I'd need a VBIOS flash, but hopefully not even that much). Pretty sure it's smooth sailing from here once I swap the card.

 

Running a full system load is stressful? did you get a wall power meter and check you were not overloading your brick?

 

mfw I look at my brick and realize that, though it's twice as voluminous as a regular 90W brick, the label actually says "120W" and I wonder why it's audibly screaming for mercy.

(I originally got the computer without a power brick, and I'm still struggling to find an economical power brick in the <$40 price range that's not an electronics-frying tin can ... currently it's a really old HP OEM brick with a different tip)

Still, even on battery alone (the battery's actually really good for having a "dec-2012" sticker - 75% capacity), I can't get it to run anything 3D, so there's that. It always ends up using the Intel GPU even though every app I try is set to use NVIDIA in the control panel... or even launched manually with "use high performance" option.

 

Well that's your issue lol, 180w is a minimum.

 

What Meaker said. You need a bigger brick. 120W is way too low. My brick is a 230W.

 

It's proving really hard to find an adapter that doesn't say " Replacement AC Adapter" on the label. A "used item" (good way to find OEM bricks) search for "(clevo,sager) 180W" on eBay gives 2 results, and for just "(clevo,sager) (ac adapter,charger)" gives 9. Everything else is either horrendously expensive or "Replacement AC Adapter" toaster garbage.

I've got a Lenovo 180W adapter I can chop up (yes, with solder and heat shrink, no electrical tape meth-head crap), but it's 20V. I don't suspect there's really a problem with using 20v, as the original was 19, so a bit like 5% more but probably not even that much. That ought to work.

 

You cant find any Delta power bricks? Where are you based?

 

No, I can find them, but they're tooooooo dam spicy. $80 is a little hard to swallow. (SF Bay, CA - btw)

I took care of it today, put a matching tip on the end of that Lenovo brick, and had it burn-in at full CPU & GPU compute load for a good 6 hours today. Zero issues, thermals well balanced (82 GPU peak with graphite+paste, 70 CPU with just graphite pad). Brick gets comfortably warm, no screaming for mercy. A well built OEM brick is better than an overspec'd knockoff

Now I just wait for that 970m to turn up in the mail... seller ain't exactly hustling to get it mailed, so it'll probably be mid next week. And I've got a new P2808B power control IC on the slow boat, probably end of the month or so. I'd installed a P2808A and that's why I've got such wacky power behavior*.

* long explanation for the curious: I originally got the computer with a physically-blown power IC, so I got a replacement... but the wrong one. With the P2808"A" chip, had to do an ugly diode hack to get it to work, power button doesn't function at all, can really only put it to sleep because it comes right back on if the battery is installed, can't turn it on at all without battery (though it runs without battery once on), and if it's plugged in but "off", it'll turn on if I insert the battery, and if the battery is fully charged, I have to plug-in an unplugged AC adapter to get the back-feed from charging the adapter's capacitors to kick the power on. Yeah. This chip is just a "soft power switch", not a power manager - so it's just a digital on/off switch, so doesn't really affect operation once it's running.

 

Bam, the 970m arrived. Works perfect! It auto-installed the driver, as expected. And now 3D rendering works great... super stupid fast. Amazing. 100% great success.

... Well, except for resuming from sleep. When resuming from sleep, the GPU gets stuck in high-performance power mode, just absolutely inhaling battery life. I can see it in GPU-Z, the clocks are at full performance power. I have to disable and re-enable the GPU to get it back to normal, or sometimes I can open and close a 3D program to reset it.

I tried Googling it, but I can't get any results that deal with the "high performance when resuming from sleep" issue, instead dealing only with "can't go to sleep" or "wakes from sleep" issues. The GPU ought to be just _off_... but it's running full blast instead, doing nothing.

Any ideas? I'll come up with a workaround otherwise but figure it's worth asking.

 

Hibernate instead?

 

You can use restart64.exe from CRU to easily restart the drivers.

 

Tried it, but it only restarts the Intel driver, not the nVidia. I don't even want it to restart the Intel driver... that does no good and just messes up all my windows, hah.

edit: oop, maybe never mind, I keep forgetting to give GPU-Z a few seconds before taking the clock speed reading. Maybe it's restarting both, but restarting the Intel (unnecessarily) is really a fair bit annoying. I might just write a little devcon.exe script to do it.

edit: The P2808B0 arrived, swapped out the P2808A1 I'd previously installed, and now everything including the power button works great - but I still have to run restart64.exe each time I resume from sleep, lest the GPU bakes away all my battery. I suppose I'll live with it, but if any future thread archeologists stumble over this with insight, it'd be welcome.