Precision M6700 owner's thread

[Aaron44126]

40 minutes ago, Goophy69 said:

Hi all, I'm thinking to upgrade the graphic card in my M6700 too.

Would like some cross checks to just make sure I have all the info and consideration update to date.

 

If you have the IPS / 10-bit display panel, then it is eDP.

 

Quadro M5000M is the easiest upgrade.  You just drop it in and mod the INF.  Or, there is an easy GUI tool that will do that for you now (nvcleaninstall?).

Also, you can use GeForce 980M, it might be easier to find / a little bit cheaper and it is basically the same thing.

(There is also Tesla M6, but you must use that with Optimus enabled, which isn't possible with the IPS panel.)

 

Some users reported newer drivers not working with Maxwell GPUs in the M6700.  I took the M5000M out of my Precision M6700 in August 2021 and I never ran into that.  There aren't a lot of M6700 users left (especially with "unsupported GPUs") so I am not sure what the current status is.  If you run into trouble, you might want to first try loading drivers from before that time.

 

If all you really care about is HEVC (H.265) video decode, then you could use M3000M / M4000M / GeForce 970M and be fine.  I believe the video decode unit is the same in all of these Maxwell GPUs and performance doesn't depend on vRAM or the number of CUDA cores or anything like that.

 

Pascal still has the VBIOS issue.  You must flash an ES VBIOS on.  You need to do that in a different system or with an external flasher.  Also, when I tried it, I had stability issues (periodic BSOD) — looks like you saw my posts about that.  Not sure if it was just something off with my card, but @jeamn ran into something similar with GeForce 1070 in Precision M6800, as I recall.

 

Turing does *not* have the same restriction with the VBIOS.  HP's RTX 3000 MXM card has been confirmed to work in Precision M6700.  However... the card isn't standard MXM layout, you'd have to do some modding of the heatsink and chassis to make it fit.

[Goophy69]

On 5/29/2024 at 10:40 PM, Aaron44126 said:

Turing does *not* have the same restriction with the VBIOS.  HP's RTX 3000 MXM card has been confirmed to work in Precision M6700.  However... the card isn't standard MXM layout, you'd have to do some modding of the heatsink and chassis to make it fit.

I just searched RTX 3000 MXM looks like there is a Dell Alienware version that looks like the same shape even though slightly different layout on the capacitors stuff (those would be covered by thermal pads so probably no sanding needed).

But the $300+ price is show stopper 😑

 

M5000M is selling for $130+ while M3000M is ~$35 and M4000M is ~$65

The M3 and M4 both outperform K5000M (well, M3 is a little behind on texture but not much that I'd worry about),

so am leaning toward M3 and M4 for now.

 

Didn't know the Geforce equivalent would be cheaper.

Quick search looks like 970M is ~$80 while 980M is ~$150

both more expensive than M4000M and M5000M,

back to M3 and M4 for now..

 

Thanks for the quick response BTW 👍

 

Now pretty much settling with M3/M4 depending on price

 

On 5/29/2024 at 10:40 PM, Aaron44126 said:

 

Some users reported newer drivers not working with Maxwell GPUs in the M6700.  I took the M5000M out of my Precision M6700 in August 2021 and I never ran into that.  There aren't a lot of M6700 users left (especially with "unsupported GPUs") so I am not sure what the current status is.  If you run into trouble, you might want to first try loading drivers from before that time.

Similar on the current setup.

With Dell's old driver, NVENC API version is only 8.0,

HandBrake the software I often use to transcode video requires 12.1.

With newer driver downloaded from Nvidia, handbrake can utilize the GPU for H264, but 10bit display output setting in the Nvidia Panel won't work. It falls back to 8bit after clicking Apply.

 

With the need of H265 acceleration, the relatively low price ($30~60) of M3000M and M4000M are pretty attractive now.

Wouldn't refuse a M5000M if by magic an auction comes up with ~$80 or so.

Anyway, thanks for the quick recap.

[Aaron44126]

1 hour ago, Goophy69 said:

With newer driver downloaded from Nvidia, handbrake can utilize the GPU for H264, but 10bit display output setting in the Nvidia Panel won't work. It falls back to 8bit after clicking Apply.

 

My M6700 had the LVDS 8-bit panel so I am not familiar with the ins and outs of how the 10bit works.

 

I do have this tidbit though.  Another display option offered in the M6700 was a "3D" panel which allowed for using NVIDIA's 3D Vision feature with apps/games if you had appropriate glasses.  3D Vision would only work if the NVIDIA driver could validate your setup, and it was more picky than it is for just installing the driver.  This feature would break if you tried to install an "unsupported GPU" like the Quadro M5000M or GeForce 980M.  ...You could still use the system, and you could still use the "3D" display panel, just the actual "3D" functionality of it would not engage.

 

All this to say it wouldn't surprise me if there is trouble getting the 10-bit display to work with the Quadro M5000M or the newer drivers.

 

Though I'll also say that Dell also offers new-ish drivers that support the Quadro M5000M (+M4000M,M3000M) — you can grab them from the support page for the Precision 7710.  You'll have to mod them to install on the M6700 (like you would have to do with any NVIDIA driver) but maybe the Dell magic bit that allows the 10-bit display panel to work would still persist in those ...?

[Goophy69]

4 minutes ago, Aaron44126 said:

Though I'll also say that Dell also offers new-ish drivers that support the Quadro M5000M (+M4000M,M3000M) — you can grab them from the support page for the Precision 7710.  You'll have to mod them to install on the M6700 (like you would have to do with any NVIDIA driver) but maybe the Dell magic bit that allows the 10-bit display panel to work would still persist in those ...?

Great hint, will give this a try after the M3/4/5000M upgrade.

 

Was searching through 7780/7770/7760 all the way down to 7710..

7740 seems to be a sweet spot ~$350~400 a nice CPU and GPU upgrade as well as tons of speed upgrade for peripherals.

but well, can't resist a $30 easy upgrade for now.🫠

[TheQuentincc]

GM204 doesn't really support H265 decoding, it's not efficient and would still load quite a bit the GPU. Sadly there is no MXM-A GM206 card so your stuck with Pascal for your needs.

Also I'd like to point out the HD4000 does support partial H265 decoding. Intel made a drivers when H265 was released and I was able to decode 1080p H265 video with my HD4000 rather than my CPU (which was a time was a 3610QM), from CPU at 90%/35W to not even 10%/10w, I was using MPC-HC back then. I know that the main issue is app support, I was able to get it working on like VLC when it was working on MPC-HC...

[Goophy69]

26 minutes ago, TheQuentincc said:

Sadly there is no MXM-A GM206 card so your stuck with Pascal for your needs.

From quick online searching the GM204 might have the so-called partial H265 support.

Sounds like a combination of hardware + software..

I'd love to go with Pascal P3000, but the VBIOS with M6800 requirement is a bit overkill, and the possible instability.

 

So for now I'm just waiting for a good deal on M3000M 🫠

 

I upgraded the display to the 10bit IPS panel so the intel graphic is auto-disabled by the BIOS; can't use the internal graphic sadly.

[Meronio]

Hi, 

what is actually the fastest mxm gpu card which works with the m6700? The latest mxm cards are nvidia 2080 rtx.

Also some user here is using an i7 4900MQ in the M6700. I thought m6700 has a FCPGA946 socket. i7 4900MQ is FCPGA988? 

[Aaron44126]

53 minutes ago, Meronio said:

Also some user here is using an i7 4900MQ in the M6700. I thought m6700 has a FCPGA946 socket. i7 4900MQ is FCPGA988? 

 

This was an incorrect post, you can't use that CPU, I think he was actually talking about a M6800.

 

GPUs have been rehashed recently.  You can use GeForce 980M or Quadro M5000M without trouble.  With Pascal, you can only use P3000/P4000/P5000 with "engineering sample" vBIOS (must be flashed separately, you can't do it in this system).  Even then there might be stability issues.  (I tried P5000 myself but got rid if it because it would cause a BSOD once every day or two.)  No other Pascal cards will work (Quadro P5200, GeForce 1070, ...).  With Turing, it has been reported to work if you use an HP Quadro RTX GPU but they do not have standard layout so you will need to mod the heatsink and chassis just to get it to fit.

 

With Pascal and up, LVDS is not supported.  If your laptop has an LVDS display panel you must use it with Optimus enabled.

[Goophy69]

29 minutes ago, Meronio said:

what is actually the fastest mxm gpu card which works with the m6700? The latest mxm cards are nvidia 2080 rtx.

I went through this (sort of) just a month ago, feel free to check the posts just above yours. The easiest is M5000M.

The newer may have extra work.

 

I was planning to get the M3000M since I don't really need the higher power GPU and vRAM but just the newer encoder and decoder, but later decided to order a M2200 (smaller size card) to just give it a try.

The M2200 has GM206 chipset instead of GM204, and various online documents hinting that the GM206 has better decoder than the GM204.

The card is still on the mail, should be arriving Monday~Tuesday this upcoming week.

Logically a smaller MXM-A card should work in a MXM-B slot, just need attention on the heatsink, height/thickness on various components, etc. Will see how it goes anyway.

 

If you're after the raw GPU power, it would be GeForce 980M / Quadro M5000M or the newer RTX / Quadro equivalent stuff. The wiki links above have a nice list.

 

38 minutes ago, Meronio said:

Also some user here is using an i7 4900MQ in the M6700. I thought m6700 has a FCPGA946 socket. i7 4900MQ is FCPGA988? 

M6700 indeed has a socket, replaceable CPU; though I think for the 3rd generation CPUs only (2nd Gen likely works too though no real incentive to install a 2nd Gen in the M6700).

4th Gen also has socket version (you can upgrade from something like 4700MQ to the 4920/4940MX for example), but I don't think a 4th gen CPU would wok in M6700, can't seem to find any confirmation.

If memory serves, couldn't recall any success on this subject either (across various product line).

Not much performance improvement between 3940XM vs 4940MX anyway.

I think 3940XM (or the lower tier 3920XM) is the maximum that the M6700 can take.

[Goophy69]

Update on the M2200 (GM206 chipset, MXM 3A smaller size card, Dell version DPN 7024K).

 

Card arrived today, K5000M out, M2200 in without any issue.

No need to mod heatsink, chassis, or inf file; the latest driver from Nvidia website installs just fine.

 

The only real mod required is the thermal pads.

The location of the vRAMs are a bit off, need to add more or fold-and-reuse the old one.

Load tested about ~10 seconds, seems okay. (haven't redone the grease so just functional test for now)

Both CPU and GPU fans auto control fine.

 

Overall, an easy upgrade.

 

BIOS shows unknown video card with correct amount of vRAM

 

Heat sink with old thermal pads for K5000M (missing a strip on the top edge that's stuck on the card)

The modification needed is 1.5~2mm for the new vRAM.

I used 1mm + 0.5mm (the only thickness I have available atm) that seem to mate fine.

Could probably move the old vRAM pads a little toward the core, and cut a strip to stack on the lower plane.

Should have enough since there are only two RAM chips instead of 4.

Also needed to push the arrowed piece upward about 2mm and put a 2mm pad for the R82. (keep the really thick old pad for the FET).

The rest can be discarded.

 

GPU-Z before driver re-install (the old driver for K5000m didn't work automatically)

 

After driver re-re-install:

 

 

10bit setting appears working on latest driver:

 

Brief benchmark/load test,

 

For brief load testing with AIDA64, with the CPU,FPU and GPU all 3 ticked, watt meter settles at 125W.

Sadly forgot to run some tests with the K5000m; from memory it was like 175W with CPU+GPU loaded.

 

Looks like a few transient PerfCap during the benchmark, reason PWR..🥲 Maybe hitting the 55W limit?

 

Might just install the K5000m back and run a benchmark just for comparison..

Still need to cleanup the old pads, and re-paste anyway.

 

--updated 7/5/24--

in normal use the GPU appears to work fine, steady at ~75W on load.

I'm suspecting that the vBIOS of the card was designed for 7520 which probably has 50W thermal design and 75W boost hard limit on vBIOS.

The transient PWR PerfCap is likely due to a software setting limit in the vBIOS.

It already outperforms the k5000m on every aspect, so I'm going with it as is.

 

Tip for thermal pad,

get some 0.3mm ahead of time 🤪

 

[Goophy69]

Just for comparison sake, here's the benchmark for K5000M

 

 

idle at 35W

GPU loaded at 110W

CPU+GPU at 138

CPU+FPU+GPU at 155

 

Overall still a very nice upgrade even though a smaller size card 😃

 

Another 3 pictures just for future reference.

Please excuse the crudity of the thermal grease; just wanted to put something there for a quick boot test, and not fry it before it thermal shut-down.😂

 

 

 

 

[Goophy69]

The caps between the R22 on the K5000M likely don't need thermal pads;

the 3 caps on the M2200 are likely air cooled too.

Just to confirm, the Nvidia heatsink on Precision 7520 has a little cut out on the heatsink for the 3 caps, maybe just due to the height of the caps but I don't see any height difference from the top of the cap to the green PCB plane of the core. Maybe that cut-out is for some extra air flow?

Might just put a thermal pad on the tantalum as well.

 

Looking at the lump of grease after separating the heatsink from the M2200, I'm suspecting the copper didn't really touch the die all the way down.

The impress of the R22 on the pad also feels a little too hard.

The stock vRAM pads on the K5000m heatsink are 1mm, might be better to substitute them with 0.5mm pads, so the copper should make a better contact with the core.

 

This pretty much wraps up the upgrade.

 

Hopefully this opens the door for other MXM 3a cards for future upgrades.

[Goophy69]

Some correction:

 

After a few more trial and errors, I realized that when tightening the screws, the heatsink won't touch the core evenly.

The core imprint on the copper looks like this:

 

The tallest R82 seems to the culprit at my first glance, so Dremel time..

 

 

However, after sacrificing the heatsink, the core and copper still don't mate evenly..

This time appears to be one of the R22 that sits somewhat higher..phew..

After some brainstorming for a solution I found some copper plates from like 10-years ago when they were used on north and south bridges, time flies 😂

 

Adding a 0.5mm plate to the heatsink effectively raises the core a little, giving more room for the pads on other chips.

Now the paste makes a perfect imprint

 

 

I also tried a 0.75mm plate, but the 0.5mm seems to be enough (core touching copper, and still leaving a gap between the heatsink and other components).

Adding a copper pad also makes the Dremeling unnecessary.

 

Wasn't really an easy upgrade by now; result not bad anyway. Didn't know the old copper pads can still find their use nowadays.

[David55]

Does the M6700 have a spill resistant keyboard that channels liquid away from the motherboard and/or other water sensitive components? I accidentally spilled a cup of water onto the left side of the keyboard. I powered down, and removed the battery and AC plug. Is it safe to turn back on? Should I disassemble and try to dry off? Please reply with any info and/or suggestions! Many thanks...

[Goophy69]

3 hours ago, David55 said:

Does the M6700 have a spill resistant keyboard that channels liquid away from the motherboard and/or other water sensitive components?

Yes the M6700 does have this design.

For water, I would still disassemble at least the keyboard and likely palmrest too to examine if any liquid gets onto the motherboard.

I mean the seam around the keyboard is NOT sealed, so the water could have leaked through..

hair dryer, sunny day outside for a full afternoon etc should take care of the residue.

water only.

sticky syrup-like stuff would need alcohol

 

-updated-

From memory I saw a suspicious hold on the bottom, and the back of the keyboard has plastic sheet sealed, all signs of water protection, but can't recall seeing that hole on the keyboard..

Might need to disconnect the keyboard to check.

[David55]

20 minutes ago, Goophy69 said:

For water, I would still disassemble at least the keyboard and likely palmrest too to examine if any liquid gets onto the motherboard. ...

Might need to disconnect the keyboard to check.

Thanks. I dug out a brand new keyboard replacement that I ordered years ago and never installed, only to discover that one of keyboard screws is stripped and I could not remove the old keyboard. Might try to remove again another day. (Somehow I did just before manage accidentally to rip the Caps Lock key off the old keyboard.) From what (little) I could see inside, nothing looked wet. I keep the M6700 propped up at a substantial angle (on a stand for the back edge to allow for easier air flow). Hopefully, this might have helped with the spill. I did find what looked like most of the water collected along the bottom outside edge of the chassis/frame. 

[Goophy69]

System has a reboot pending anyway so I checked the bottom and keyboard.

Back of the keyboard is indeed sealed with drain hole on the bottom left corner (top left on picture cause it's flipped up),

that cut-out corresponds to the channel hole on the bottom left corner,

 

 

that hole goes to the optical bay,

the hole in the top right corner of the optical compartment.

(the hole inside the little square form guard between the optical and memory also looks suspicious, but I don't see anything near that location on the keyboard)

 

The seam around the keyboard is not sealed

so if you have the laptop tilt, hum, this could actually help the water to overflow onto the motherboard...😅

 

If you cut the power quick enough (before it flows to anything that impacts the voltage), and dry everything completely before the next power up (and before the residue can have a more permanent impact on the electronics, like rust),

it should be fine.

Pure alcohol can absorb water, then evaporate together, can somewhat help to accelerate the drying.

Good luck anyway.

 

-updated-

for a stripped small screw, the regular extract tool won't work.

my favorite is light-duty thin dremel or maybe heavy duty utility knife to deepen the +,

or just make a fresh deep -, then regular screw driver will work.😏

[David55]

21 minutes ago, Goophy69 said:

...

-updated-

for a stripped small screw, the regular extract tool won't work.

my favorite is light-duty thin dremel or maybe heavy duty utility knife to deepen the +,

or just make a fresh deep -, then regular screw driver will work.😏

Thank you for the input! I am (perhaps irrationally) afraid to take a drill near the laptop. May I ask what you think of using electrical or other tape, or supergluing a screw driver to the stripped screw?

[Goophy69]

2 hours ago, David55 said:

Thank you for the input! I am (perhaps irrationally) afraid to take a drill near the laptop. May I ask what you think of using electrical or other tape, or supergluing a screw driver to the stripped screw?

using some abrasive disk to cut a - with hand works too if electrical not feasible.

 

Tape is soft so probably won't work; the screw cap is probably too small for tape to adhere to.

 

not sure about regular super glue, normally not strong enough either.

those acrylic/epoxy type (2-component, cold-weld putty, etc.) in theory should do the job,

might need some prep work like scratching the screw cap for a better adhesion etc..

Still think making a - on the existing + is by far the easiest; I had successes with this method on stripped large head hinge and touch pad screws (those 1/4" diameter head with 1.2~2mm thread 2mm long screws with tiny +)

Those $10~20 cordless rotary tools powered by 18650 work great on this type of work, easier to maneuver; 120v is too heavy.

Can also cover the surrounding with a few sheets of thick paper / glasses cleaning cloth, tape them in place, they can withstand the few occasional slips.

 

-updated-

depending on how badly stripped that screw is, and how tightly it's screwed, tape between the screwdriver and groove might work, worth a try.

before cutting wheel, can try cold-weld putty too. The putty doesn't need to stick to the screwdriver very firmly, but needs to stick to the screw head to create a good "thick" layer for the new groove to take place. When cured, the screwdriver essentially drives the putty layer.

Plastic film around the screw head can "shield" out the surrounding in case the putty gets squeezed beyond the round screw head.

The liquid type glue-like 2-component epoxy works too; though this one really binds the screwdriver to the screw.

[Goophy69]

I have this dremel stylus from years ago when it was the only few options available.

 

 

then the market was flooded with all kinds of gadgets, lithium powered stuff..

now you can get something similar in the $10~30 range; even dremel itself now sells their cordless tool for $35 retail price, technology advances so quick 😂