1070 gtx in an m17x r4

Thank you for the fast response, I'm not afraid of modifications, I already done it for another laptops:
I installed a 675m on a clevo D900F (first laptop with a computer CPU) .
ASUS ROG G73SW with a 770M.
So I am confortable with it, I just want to upgrade the laptop for the next 3 years. And Yea the price will be something to think but I will be able to do it on April

Enviado do meu BLA-L29 através do Tapatalk

 

Another update, real quick since it's a bit off topic. This week I installed a WiFi AC card, the Intel 7260HMW.



Since the stock antenna is optimised for 2.4GHz and WiFi AC uses the 5GHz band, the antenna needs to be replaced.

I saw someone who decided to drill the chassis, and mount SMA connectors for external antennas. That's definately interesting, since external antennas means way better performance, the possibility to connect big high gain antennas and stuff like that...
But it's also a modification that is visible from the outside, something I did not want to do. And it is also less practical, considering each time I would put the laptop in the carrying bag, I would have to remove the antennas, and then put them back in place each time I need WiFi.

But I guess it's all about compromises here, less ease of use, for more performance and versatility, or the other way around. I went for the other option, and decided to replace the integrated antennas inside the display assembly.

I had this little Asus antenna from my desktop computer motherboard. I do not use it since I ran ethernet cables across the house. So I took it apart and salvaged the two antenna boards.



I desoldered both coax, on the Asus antenna, and on the Alienware. The braid was soldered at other places, I just removed it, it should be fine.



I did not have any heatshrink where I was that day, but I'm going to put some on the exposed braid of the coaxial cables to prevent any issues.



It's working fine so far, but I've seen better performance. The speed decreases quickly when I move a bit further away from the router.



I guess these antennas are definately not the best. It was a nice try but I'll see what I can do to improve that a bit more later.
I think there is enough room where the antennas are at the moment, maybe I could fit higher gain 5GHz antennas.

 

Something must be wrong with your heatsink. I am running the same stresstest and i reached my max reading 10 minutes ago at 78 C. I am counting 25 minutes and still running at 77-78 C. Will give another feedback after i reach the 40 minutes mark

Gesendet von meinem SM-G935F mit Tapatalk

 

Ok I'll just try something. I'll move the laptop to another table, with nothing behind the laptop.

Right now I have a wall 25cm behind. Oh, and it's 28°C ambient temperature, I forgot to mention that...

No idea if it changes anything here. The difference is huge.

 

I dont even have those 25 cm. It must be your ambient temp or heatsink.

Gesendet von meinem SM-G935F mit Tapatalk

 

Ok, I'll have to investigate that tomorrow. That's weird indeed.

What ambient temperature do you have? I doubt that's the only reason anyway.

 

Oh, and what are you using to get all those infos on the screen? I only have the default text from Furmark.

 

23°C.
But thanks to this test i can say now that my rig is stable and the mods were well done.
Reaching 40 minutes and its still stable @78°C.

Gesendet von meinem SM-G935F mit Tapatalk

 

RivaTuner and HwInfo combined.

Gesendet von meinem SM-G935F mit Tapatalk

 

Well, I was running Furmark, the temperature reached 82°C in about 5mins.

Then I just raised the back of my laptop by 3cm, the temperature went down to 75°C really quickly and it just stays that way. 12mins and counting. I'm curious to see how far it will go that way.

But your laptop was just flat on your table, so that's weird.

 

Yep, flat.


Gesendet von meinem SM-G935F mit Tapatalk

 

So, final result is 78°C stable temperature after more than 50 minutes of Furmark 1920x1080 8x MSAA.
Notebook on a flat table, 23°C ambient temp.

I think this one goes to the HS provided by @woodzstack. For now

 

I have a slightly higher GPU Core voltage of 0.850V to 0.863V, but the difference is really not huge compared to yours.

I also have 5°C more ambient temperature, that sure doesn't help, but it does not justify 11°C difference with the laptop sitting flat on the table.

I'll recheck everything tomorrow.

 

that what I do, help out where I can !

 

sda

Aight, lets call it a day! Make sure the core makes proper contact with your sink. When i saw the full copper sink i really had my doubts, since it feels way more stiff compared to the copper/alu hs. I still have it here, but i didnt even bother to try it out yet.

 

I replaced my intel wifi card with a Killer 1103, which can handle 5Ghz Networks at 450 Mbps just fine. It has 3 antenna slots. You have an extra antenna cable that you could connect on a 1103. No mods needed.

Gesendet von meinem SM-G935F mit Tapatalk

 

looks like a 4th heatpipe could be added to the CPU cooler if the frame and cover/lid was relieved a little bit.

 

Hi guys... Got back from GDC yesterday. I'll post some numbers and pictures re:5-pipe build in the next day or two. Sorry for dropping off the face of the earth: We had to crunch out a build ahead of the con.

That said, those GPU temps ( @etienne51) sound high... I've not run anything more intense than the TimeSpy benchmark yet, but I've not seen temperatures higher than 65c. I don't tend to run furmark on laptop parts... I'll see about that...

 

All right; Time for some figures-and-such.

Performance tests

Tests performed at 41/41/41/41, 220/250 (figured I'd keep with the benchmark set by A-Gits earlier to keep things comparable).

Timespy


Pretty much as expected. Graphics score wavers between 6000 and 6039.

Firestrike


Again, pretty much as expected.

Notes on Performance
1. GPU Boost 3 is... weird. Under 100% load, the card seems to cap out at it's boost frequency using ~0.90v... but under balanced load (about 50% GPU load and some sort of secondary stress like VR or lots of shader effects) the card will spike the voltage to 1.00v-1.06v and throw the clockspeed to 2000mhz+. It's not temperature dependent. Using a forced air cooling pad I ran Timespy at 58c... no dice. However, throw on a VR game like Arizona Sunshine or Narcosis, the card can be running at 65c at 2100mhz.
That said... I've heard that EVGA Precision can tune the frequency curves on Pascal cards but I'm not touching that app with a 10ft pole... Too many reports of bricked screens on 17xs and 18xs...
2. When in a low power state (see "acting weird") the card will ceiling at its unmodified clockspeed (1442mhz) and refuse to budge. Hibernating and waking clears the state and returns the card to normal. Not sure -exactly- what triggers the state. Shutting down certainly does... and so does hibernating off power in my bag sometimes? Bit of a mystery on that one still. At least it can be reliably cleared.


Mods
Heatsink mods cuts in orange
Aluminum cuts in red
Plastic cuts in green

Base/hatch - Ugly Dremel work on display



GPU Heatsink And Bay Mod - Less-ugly Dremel work. Note here that I had to buzz the corner off the GPU fan to get it under the larger pipe cluster.



3/4 Profile on GPU Heatsink - Wherein I show that I nicked one of the heat pipes. Also partially displayed: An ugly M14x fan mount using brass retention clips.
Not displayed: Pad heights
Memory: 0.5mm (I neglected to record this actually. It may have been 1mm. Whatever it was it made the memory match the height of the coils).
Coils: 0.5mm
VRMS (and similar-height units): 2.5mm




Notes on Heatsinks
1. Using Kryonaut
2. Didn't bother lapping the heatsink




Temperature/Timing/Voltage Data

Timespy
Voltage range: 0.8120-0.9000v
Temperature Ceiling: 69c (flat on desk) 65c (cooling pad) -- Oddly, I get about the same temp (65c) if the laptop is balancing on my knees/lap, so it has more to do with airflow than forced air.
Clock range under load: 1662-1865 (220mhz oc), resides mostly around 1750

Furmark
Voltage range: 0.843-0.850v
Clock range under load: 1708-1850 (220mhz oc), mostly around 1770
At 75c the clock speed started slumping into the high 1600s (1678-1700). This happened after about 15min.
Half hour mark at 78c (flat on desk)
So not markedly different from the 5-pipe copper/alu heatsink. Finishing it might do something, but I'd doubt getting more than a degree or two... Probably not worth the effort.


That all said, congrats to the thread and getting the card properly modded in! We're late to the party, but the keg's not empty yet lol.

Also... It looks like a functional vBios has been discovered for the standard-sized card (thanks @A-Gits and @etienne51 for picking up on that)...
At least if MSI produces any more mutants, we might get one more gen out of these dinosaurs. I mean... at 5 years old they're technically timebombs now... we'll have to see... I mean... my M15x has got to be 7 and hasn't exploded yet...


Bonus: 3940xm OC Settings/Notes with the 3-pipe sink
(sitting flat, 8-thread 1024, stability verified on 1, 2, 4 threads after)

41/41/41/41 - voltage offset 0 - TSBench temp ceiling sitting flat: 88c - gets through about 45% of the test at 56w before the 55w long power limit catches up and forces 39x. Strangely, 40/40/40/40 is almost exactly the same even though it runs at about 54w until getting capped back to 39x at 45% or so.

43/43/43/43 - voltage offset 13 - temp ceiling 92c - gets through about 20% of the test at 64.5w before the 55w long power limit catches up. (Note that I was able to complete the test at a voltage offset of 5 but my compiler would crash during builds until I gave it more juice).

44/44/44/44 - voltage offset 18 - temp ceiling 92c - gets through about 20% of the test at 66.5w before the 55w long power limit catches up.

45/45/44/44 - voltage offset 25 - temp ceiling 92c - redundant. Same results as 44/44/44/44 on an 8-thread test. It'll single/dual thread like a ************ but modern games are pretty thready so this is less meaningful in 2018.

That said, I'll likely continue using 41x at no offset. I'll post a 45x 250/400 benchmark later though for ****s.

I'll also likely try to tinker with these voltage offsets a bit (particularly the 43x as there may have been something else at play there)

 

And here are a couple of burnouts (45/45/44/44, 250/400)

TimeSpy


Firestrike



Interestingly, with the beefy heatsinks, temperatures are no longer an issue. On the cooling pad, the CPU hit 90c and the GPU hit 64c... It's really just the limits of the parts themselves (can't change the short/long duration and limits on the 3940xm, can't convince the GPU to use "boost 3" under heavy loads).

Even more interesting is the draw at the wall: 225w during the firestrike combined test. The 980m (@1.06v) would take it to 235w (41x cpu no offset).

I'd like to catch the draw at the wall when the card goes into overdrive...

 

Ok, thanks for all these results and test data. It's very interesting for me to see that, in my current configuration, I get the exact same performance out of my 1070.

I finally got that in the mail today! The Dell 330W adapter.



Here is the latest benchmark I ran just now. I'm on 42/41/40/40 and 220/220. Remember I don't have the fastest CPU, for now it's a 3840QM. Upgrade probably coming soon, with a 3940XM.



Finally I'm able to reach a score between 13700 and 13800, with the same settings I had before. I couldn't go past 13200-13300 with the 240W adapter.
I can see the power consumption go up to 231W at the wall, and with the 240W adapter it was around 210-215W at best. So I guess my PSU was indeed limiting, it's not the case anymore. Great news!

Regarding the temperatures, I tried something yesterday.



This is something I built on my Prusa i3 MK2 3D printer. A couple feet to get a better airflow for both the CPU and GPU fans. It raises the back of the laptop by 2.5cm (1in). It fits tightly around the back rubber feet, and it stays there really well.
I don't mind the slight angle it gives to the laptop, and the gain in temperature is really huge. So I think I'll definately keep it that way.

Here is a comparison of the max temps reached, after 40m of Prime95 on the CPU, and later 40m of Furmark (1920x1080, Fullscreen, x8 MSAA) on the GPU.
- Laptop flat on the table: CPU 91°C / GPU 87°C
- Laptop with the feet: CPU 86°C / GPU 78°C

So, with the feet, I have -5°C on the CPU, and -9°C on the GPU. And that's with the most stressful stresstests I could run, I guess. Both the CPU and GPU should run a bit lower when gaming I suppose. And I still need to upgrade the CPU fan.
So I think I might be looking at something around -10°C on both the CPU and GPU with the feet. Great improvement I'd say.

As @Schurke mentionned.

Indeed, it's just that the airflow is very limited with the laptop sitting on a flat surface.

 

And another test I did, btw.

I swapped the heatsinks a few days ago, to compare the stock heatsinks of the M17x R4 with the copper heatsinks with 3 heatpipes (CPU) and 5 heatpipes (GPU).
I did these tests with the back of the laptop raised a bit (about 3cm), to make sure I would have no cooling issues.

CPU: Stock HS 87°C / Copper HS 84°C
GPU: Stock HS 89°C / Copper HS 74°C

So I had a -3°C improvement on the CPU and a huge -15°C on the GPU.

Now I'm starting to doubt these results, thinking maybe I had a thermal paste issue with the stock heatsink on the GPU, the difference seems really huge, while not much changed with the CPU.

 

Does the core make good contact with the heatsink? I saw you polished the center of the hs, maybe you scrubbed off too much?...

Gesendet von meinem SM-G935F mit Tapatalk

 

Yes, I had a close look at the heatsink, it seems to be perfectly flat. And the thermal paste spreads correctly everywhere, in what seems to be a really thin layer. When I remove the heatsink everything looks normal.

 

Defective card? Or your gpu fan is not running at its supposed max speed.
Imo its still too hot.

Gesendet von meinem SM-G935F mit Tapatalk

 

Oh! I don't know if you guys remember when I was talking about Black Ops 3 not running well at all, below 120fps when it was supposed to run way way better than that.

Now it works above 200fps, actually around 230fps... Wow.

I did two things, I reinstalled Windows, and I replaced the PSU. I don't know if it was only a driver issue, or the fact that the performance was limited by the power supply. But now everything works as expected for gaming!

There is still that issue with the temperature, that's a bit weird. Tomorrow I'll be at a friends, he has air conditionning. I'm curious to see how ambient temperature impacts the cooling here.

 

@etienne51 "Remember I don't have the fastest CPU, for now it's a 3840QM. Upgrade probably coming soon, with a 3940XM."

Can I buy your 3840qm when you do the upgrade for the 3940xm?

Enviado do meu Huawei Mate 10 Pro através do Tapatalk

 

@aconix Hmm... I'm not exactly sure when I'm going to go for the upgrade, maybe really soon, maybe a bit later. But that's indeed a possibility. I'll let you know.

 

No problem, I don't have urgency I think I will buy the 1070 on April

Enviado do meu Huawei Mate 10 Pro através do Tapatalk

 

Do you have a Link to the 3D file you used to print those?

I was hoping there would eventually be a 1080 Ti that would work by now. or at least a 1070 Ti.

 

11 series refresh could be as early as July, if the rumour mill is to be believed. I wouldn't hold your breath on a 1080 or a ti variant working in the m17x R4 however... the ~125w limit on the mxm port is a bit of a show-stopper.
I do recall someone testing a 1080 on the R4... I think they said it "worked" but quickly aged/degraded the motherboard... Anyone else remember that?

 

Sorry for the late reply. I attached my STL file.

 

perhaps solder on higher rated VRMs? or maybe a GTX 1170 if they don't kill of MXM completely for brands who do care like Clevo and Eurocom?