780M SLI Throttle [M18xR2]

So I got 780M SLI from my friend's dead AW 18. His set-up had an issue where it would throttle from 849Mhz to 405Mhz, go back to 849, then throttle again to 405 in a cycle. He disabled SLI to fix this...

Now for me...I had the exact same issue: https://www.3dmark.com/fs/19768101

I flashed the slv7 vBIOS again on both GPUs, same issue.

First thing I tried was undervolt GPUs by -50mV and it seemed to work?
2 runs: https://www.3dmark.com/fs/19768244
https://www.3dmark.com/fs/19768301

I wanted to try and fix this issue so I tried this method: http://forum.notebookreview.com/threads/alienware-m18x-r2-with-980m-possible-throttling-fix.791358/

it seemed to fix my stock voltage throttling:
https://www.3dmark.com/fs/19775350
https://www.3dmark.com/fs/19775304

Next I tried overclocking LOL which worked at 1.050v & +150 Core: https://www.3dmark.com/fs/19775399

Next I tried 1.100v & +200/+200 and it throttled again: https://www.3dmark.com/fs/19775559

Bringing it down again seems to not throttle.

So I am wondering, is this a driver issue? Was this an issue with M18xR2? I thought it worked fine. PSU is not shutting off so shouldn't be a power issue?

@Mr. Fox @Johnksss @Raidriar @Rengsey R. H. Jr. @jerryzago @GodlikeRU @vulcan78

 

Yup. I am surprised you don't remember this, bro.

There was a very clear line of demarcation when NVIDIA cut the nuts off of 780M SLI. It was the start of my abandonment of Alienware laptops. 880M and 980M just got worse. It might be a coincidence, but it started in June of 2015. Windows 10 was released to the public in July of 2015. It also roughly coincided with the release of Maxwell GPUs. Things have gone downhill since then, but what they did to 780M SLI is unforgivable. The behavior was the same with 780M SLI in the M18xR1, M18xR2 and Alienware 18.

I released these videos in 2015 and had a fair amount of dialogue with people I knew at Alienware and Manuel at NVIDIA. Waste of time. NVIDIA claimed they could not replicate the issue even though it affected all three of my systems with 780M SLI and the systems of other 780M SLI Alienware owners at the time. Alienware didn't care about their EOL products, so there was no help there. The didn't sell either M18x with 780M SLI and didn't care about the 18. They said it was NVIDIA's fault since everything worked fine with all prior driver versions, and blew it off because they just didn't care.

Fun fact: all of the 780M MXM GPUs that I owned worked flawlessly in SLI in a Clevo regardless of driver version. They only ran like feces when installed in an Alienware. Same was true of the 880M and 980M GPUs. All of these GPUs ran like a top in a Clevo, single or SLI, and they were a hot mess when installed in an Alienware. And, @Prema and @SLV7 vBIOS mods didn't help at all. So, that was the final nail in the coffin that pushed me to Clevo, and Alienware's shift to BGA filth sealed the coffin of Dellienware hate in slab of concrete for me.

The only solution for 780M SLI was to use outdated drivers, and when doing so the GPUs performed flawlessly. 880M and 980M didn't work well with any driver version. All the while, Clevos with the same GPUs were running circles around the Alienwares, never missing a beat using any driver version.

Some of the weird tweaks and gyrations helped the Alienwares malfunction less, but they never worked 100%. There was always something that wasn't right. Probably some kind of hidden cancer in the Compal EC that was never activated until NVIDIA flipped a switch in their drivers and activated the EC cancer. Before the virus-infected drivers hit NVIDIA's servers, 780M SLI in an Alienware was, literally, an unbeatable laptop hardware configuration.

These videos show precisely what you experienced.




@Papusan remember this drama.

 

A problem that was never resolved.

 

Followed almost immediately by the bricked display panels getting their EDID corrupted. Those days were like a piece of hell on earth.

 

Ah this was that issue, I thought it was just 880M SLI.

Do you guys know which latest driver version gave full performance? Doing the trick from 980M SLI seems to be working at stock and mild OC so I might stick with it but if the other drivers provide even better performance and no tricks needed I'll downgrade. I'm using @j95 425.31 drivers right now which are the last drivers ever for all series below 9.

I'm already enjoying the performance boost from 680M SLI with these. Already reached 1060 performance easily.

 

I think you will find the answer in the video. Early 2015 drivers.

 

Thanks, so for anyone else that ever sees this thread, last good driver is: 345.20

 

Just a little update.... so 345.20 is for Windows 8.1

I can run 780M SLI perfectly fine thus at stock clocks using the reboot trick from the 980M SLI thread.

I can also do a small overclock and still maintain clocks. However, medium overclocks cause the system to do the 405Mhz throttle no matter what so far.

This is all using latest drivers so 780M SLI works fine as far as I know. I will run a full on gaming session later to confirm 100% but so far, firestrike, 3dm11, all benches I ran run full clocks whole time.

Next goal is to beat my overclocked 680M SLI set-up and fix physics score (Fresh windows 10 install has all the spectre etc mitigations enabled). Can match up a properly working single 980M

https://www.3dmark.com/compare/3dm11/13393947/3dm11/13125125/3dm11/10778042

OC 980M might be doable too. 980M SLI quite faster in a properly supported system:

https://www.3dmark.com/compare/3dm11/10743618/3dm11/10760182

 

Crazy stuff, I was ever since convinced that my 780ms perform normally, and when I compared them with my 1060 they really seemed to underperfom, but never went into this more deeply, now I know better!

 

Can you run Firestrike and 3DM11 with your 1060?

I have some runs too but I wanna see how it performs in M17xR4 connected directly instead of my Optimus trick (M18x).

You could also try my trick and see how your 780M SLI performs. Not sure if it works the same R1. Actually I'd definitely like to see how your 780M SLI is performing in the M18xR1.

 

Atm I only have 2 slightly OCed runs (GTX 1060) on my mobile (~ Core +100; Mem + 200).

FS: https://www.3dmark.com/fs/15914940
3DM11: https://www.3dmark.com/3dm11/12852203


Fresh run (stock 1060, 3920xm), today:

FS: https://www.3dmark.com/3dm/37777929?
3DM11: https://www.3dmark.com/3dm11/13401106

Will run FS and 3DM11 on m18x, but have no unlocked BIOS on my M18x (yet), don't even know if that trick also works for the r1, so it will be a "stock" run.

 

Interesting, so my slightly OC 780M SLI is keeping up with a OC 1060 in PEG: https://www.3dmark.com/fs/19775399

Both at stock 1060 is a bit faster: https://www.3dmark.com/fs/19775350

Are you running RAID on your M18x? I have a updated unlocked A05 BIOS for the M18x I can send you.

 

Lol .. in M17X R4

Look at Desktop GTX 980 stock .. maybe a little throttle

https://www.3dmark.com/fs/19876747

 

Is it throttling for sure? If so what's the throttle reason?

 

i'm thinking maybe gpu driver. Can you recommend whats the best driver?

 

It shouldn't be throttling. Performane seems fine, let me find my 980 run from P870DM. Use latest drivers: https://www.nvidia.com/Download/driverResults.aspx/148586/en-us

Found it: https://www.3dmark.com/compare/fs/19876747/fs/9765515/fs/8148698/fs/16862474

Added my OC 1060 as well

Your performance isn't bad. We're comparing a 120W MSI GTX 980 to a 200W GTX 980 w/Prema vBIOS.

Use MSI Afterburner to see if it's hitting power, thermal, or other limit.

 

No RAID, single drive, think it's on stock A02 atm.


Really running well, your 780ms that's what I was expecting from mine mine had bout 8500 GPU Score, iirc.

I'll run a FS in the next days.

 

Okay, I'm down to try and fix yours too. 11k should be the GPU score.

Process should be Flash to stock A05 > Unlocked A05. Whenever you are ready for that.

 

Ok let's try to rock it, gonna flash stock A05 in the next days.

Only 1 time ever unlocked a BIOS, in my m17x r3, that was easy going, from stock A12 to unlocked was really peanuts, hope it's not that hard in the m18x, always a bit scared of unlocking

 

Hadn't booted up the old lady for 3, 4 months now and when I did today she quit that with 5 beeps. Powered her down and restarted, booted up completely, but the date and time in the BIOS were reset, the settings were still ok. Must be the BIOS battery quitting, correct?


Edit, happened again, so it was the BIOS battery, luckily had a spare one, and I was on stock A05 already.

 

Okay, try and run 3dmark firestrike or 3dmark11 and let's see where it's performing. I'm on my Mac right now, when I am back on my Area-51m I will send you the unlocked BIOS. Process should be identical to your M17xR3.

 

Had done 2 quick runs late last night, smhow forgot to post them.

FS: ~ 9800 GPU score
3DM11: ~ 14000 GPU score

Older driver, 34x.x or 35x.x
Running on Win 7.

 

Alright, my goal is to get you to

11k FS GPU & 15k 3DM11 GPU using the latest support nVidia drivers.

You are running A05 stock already right? Would you like to flash A05 unlocked? If so I'll send you file in PM.

 

Yes it's the stock A05, hope I won't brick it
EC is 1.05
ME is 7.0.4.1197 if necessary.


Is there any favor in using the svl7 vB, am not planning to OC these anyway, but uV is also possible for better temps, iirc, correct?

 

I've gone ahead and sent you in PM the updated A05 unlocked BIOS & the 780M vBIOS.

Let's see how this works out.

 

Thanks for all mate will post the results.

Atm. we have bout 40-42 C ambient, thus I have decreased testing and benching a little bit, for the next days.

 

Ambient temps went down to around 25 C, so I started a bit of tweaking.

Flashed the unlocked A05 BIO, set to defaults and then set all back to my old settings.
(smhow get an "Atheros..." Message now after every POST, which I cannot get rid of)
(strangely my CPU's sometimes hitting the 90 C barrier now)

Flashed the unlocked vB on both cards.

Ran 2x FS, yet still on the old driver: 385.69

stock: 10653 GPU points
https://www.3dmark.com/3dm/38064935?

-87,5 mV: 10681 GPU points
https://www.3dmark.com/3dm/38068519?


Next step is: installing the latest available driver, 425.31

stock: 10 148 pionts
https://www.3dmark.com/3dm/38069661?

-87,5 mV: 10 191 GPU points
https://www.3dmark.com/3dm/38069988?

 

Wow that is much better, you are 8% away from my R2.

Try this:

Start up laptop into BIOS, go to the video configuration and set to "IGPU" and let it load to windows. Then start>restart
Go back into BIOS and video configuration to "PEG" and try benchmark then.

Idk exactly where the video configuration page is on the R1 bios, but go around the tabs until you can find it.

 

Yes better than before.

Ok i think this setting must be in the 2nd "Advanced" tab. Will post a pic before changing anything.

 

Good choice, be very careful and only select iGPU & PEG. Choosing others may require a CMOS reset.

 

just curious, does this throttling issue affect the Alienware 18.
im running the latest nvidia drivers, and on 3dm11 i get 15K on GPU, 9K CPU, 12.9K overall
if I use 345.20, would it scale up? also when OC'd? Because I did notice that it didnt scale much when overclocked.

 

According to @Mr. Fox it also affected AW18.

Actually....now that I think about it, this did affect my friend's AW18 as well, which is where my cards are from.

I'm not sure if the M18xR2 trick will work for 18 but the older drivers should restore performance.

 

And Azor with Dell and Alienware did nothing! To fix it.

 

Yeah, why care when you can make $$??

$$ makes the world go round...

*sigh*

 

Honestly, 780M SLI is still very relevent, especially in games that support SLI. I didn't realize back then when all these cards were new, how powerful they really were.

This is my final upgrade to the M18xR2 and I'll be keeping it like this until it dies.

 

Me too, will be keeping my m17x/m18x. For me they stil serve well, have enough power for what I'm doin with them, a lil bit gaming, surfing and a lil bit of tweaking every now and then.

 

So I went looking and found this setting, must be this one to change, correct?

 

Yup that's the one!

 

Ok, so I did the switch and back to PEG and then ran FS, I skipped the CPU tests as it tends to peak 90s now, need to check that afterwards.

FS, stock: 10282 GPU points
https://www.3dmark.com/3dm/38151236?

 

When you run the test, can you check using something like MSI Afterburner w/on screen display to see if the GPU clocks ever go down during benchmark?

Also have you tried 975mV or 950mV?

 

Yes, will use AB next time.

Tried to run FS with undervolted GPUs but strangely the screen was running without colors, so had turned black and white

Then I resigned for yesterday.

 

Installed AB and ran FS, couldn't see any throttling, clock was just at 849 MHz all the time.

 

That's great, I wonder why your score is lower than mine by a bit though. I doubt going from 2920XM to 3920XM makes that much of a difference. Mine is OC to 4.3Ghz.

Either way I think your cards are running much better than before.

 

CPU really do help with the score ..fyi..

 

@Mr. Fox Does Windows 7 score much lower on GPU score than Windows 10? Maybe a 800 point difference, have you experienced that?

Yes but we're talking about a 2920XM vs 3920XM in a 99% GPU-bound test. There's almost a 1,000 point difference. Not to mention he is running windows 7 which gives a higher CPU score actually.

This shouldn't be that much of a difference in the GPU test. I will say his performance has gone up quite a bit (he said he was getting 8k score before).

Even my AW 13 R3 with 7700HQ (4-cores) scored around the same as desktops with the same GPU and 8700K 6-core CPUS: https://www.3dmark.com/fs/13653218


I'm more curious why undervolting is causing @M18x-oldie to get a B&W picture. When I undervolt my 780M by a bit, they are rock solid so I'm sure it would help his M18x as well.

What are you using to undervolt? Have you tried nVidia Inspector?

 

Yes improvement is there, that's cool

Yes, I'm only using nV inspector for undervolting.

The b/w picture only occured in the last 2 undervolted runs with the new driver. Maybe a driver issue??

Atm the CPU is running 3,9-4 GHz all cores due to getting a bit hot, prior it was 4,3 single, 4,2 all.


Btw, is it safe to use Inspector and Afterburner at the same time?

 

Yes, what I would do is use inspector to only modify the voltage, then use afterburner for clocks.

However, I now use Afterburner for everything after unlocking voltage option.

I'm uisng same drivers as you so maybe try afterburner for undervolting.

 

Ok, had asked cause the option for undervolting is greyed out in my afterburner.

 

Try this:

 

Thanks for that hint, managed to unlock the GPU voltage settings in AB.
Ran some FS:

stock: 10246 points
https://www.3dmark.com/3dm/38283333?

-50 mV: 10122 points
https://www.3dmark.com/3dm/38283224?

-25 mV, Core + 100, Mem + 200: 11233 points
https://www.3dmark.com/3dm/38283424?

-25 mV, Core + 125, Mem + 300: 11639 points (best run!!)
https://www.3dmark.com/3dm/38284137?

-37,5 mV, Core + 100, Mem + 200: 11343 points (artifacts after FS closes)
https://www.3dmark.com/3dm/38283634?

No b/w picture this time!
But the last run caused black artifacts when FS went back to Desktop after the run was finished, so I suppose that's not a healthy setting at all.

Final run: no uV, Core + 150, Mem + 350, max PL: 11489 points (no artifacts, clocks fluctuating now & then)
https://www.3dmark.com/3dm/38283977?


So finally managed to break the 11k barrier, but uVed and OCed, not at stock settings (yet).