Discussion in 'Sager and Clevo' started by anassa, Jul 19, 2020.
You can make the chip fit your power target.
Looks like we have the same setup!
You are getting better of everything though, I also used the Digital Foundry optimized settings and have roughly the same fps as you.
My setup is:
GPU: 1830~1835Mhz @ .912V +450 on VRAM
CPU: All core @ 4.6Ghz -0.1201V on Core and -0.1357V on Cache
RAM - just whatever the XTU auto settings are
If I run auto fans and play the GPU and CPU will get up to the 90s, if I do max fans it will vary from ~65 to ~75 depending on ambient. But without full fans the CPU/GPU just get too hot.
From what I have read the Pascal mobile GPUs seemed closest to the desktop variants, like my 1080 runs the same as a desktop 1080 @ 1830MHz, but it seems like the Ampere series are going to opposite way, and there might be SUPER variants again. I am thankful for the pixel density of 15" +1080p and Gsync, it looks good and works well so far.
If it fits (aka MXM?), cooler works, and is accepted by the bios, a 3080 (12GB or more than 8GB VRAM) would be nice, but as we all know, sure we can try to get it to fit our power target, but we will always push that as far as possible! Multiple profiles for winter, summer, when A/C is on, etc hahaha. It is kind of nice though to have something that will just run cool and you can work on overclocking to the max without worrying too much about temps or feeling like you are crippling your GPU to allow it to run. Anyways, I look forward to what the 30XX series mobility will bring! (Personally I still wish AMD would bring something relevant to the table, like the 7970m back in the day)
2021 will be interesting for mobility cards.
^^ just realized I had already mentioned my specs previously - but it is a bit more accurate as I checked the specs again.
Maybe I need to re-paste and check contact, because those temps seem high. Maybe even check contact and pad placement.
I did a lot of work on my thermals, including using a taller IHS, replacing thermal pads with K5 Pro, liquid metal on all surfaces, and getting the 17-blade fans from the 9th gen refresh (higher static pressure and less annoying pitch than original 13-blade fans).
I'm using the built-in dynamic resolution scaler to achieve a locked 60 FPS in Cyberpunk now. I set Max Frame Rate in Nvidia Control Panel to 60 FPS and in-game like this:
Combined with the AMD CAS sharpening, when the resolution drops I barely even notice it. It's still generally between 900p and 1080p (verified by comparing with static resolution scale set to 85% or 900p) when it drops, and the slightly softer image is actually a little easier on the eyes because the SSR in this game makes reflective surfaces really grainy.
Another benefit to using the 60 FPS cap with an undervolted GPU is that the CPU and GPU are below 70C the whole time and the fans barely spin up.
I also did a little work as you saw from the first post for thermals lol - delid and LM for 8086k, adjusting thermal pads and paperclip mod + P775 heatsinks, but I am interested in the 17 blade fan, where did you get it from? Did you notice a difference?
I feel hopeful with the potential mobility 3080 specs:
If it is true: GA104, 6144 Cores (more than desktop 3070), 16 GB Memory, 1700Mhz, 150w? Not bad! Sure it isn't a real 3080, but it is stronger than a 3070, so that isn't bad for a 15" laptop. Especially if it really is only 150w, then I won't have to worry about heat as much and can have fun with it. Now the price....
Good idea on the fps cap and dynamic resolution scaler . . even though, I tend to really dislike them. Maybe if I just set 85 - 100? With 60FPS cap? That could be helpful.
I think I am going to re-paste and recheck pads this week and first see how that works.
Got the fans from here: https://www.aliexpress.com/item/400...earchweb0_0,searchweb201602_,searchweb201603_
The biggest difference I noticed is that they sound much less annoying than the original fans under load. The pitch is more pleasant to my ears.
That benchmark is OpenCL, which is a compute benchmark that uses very little power on the GPU. That's why the 80W 3080 Max-Q in that Zephyrus is able to boost high and score like a 2080 Ti / 3070. In actual gaming/benchmark 3D graphics loads it'll perform far worse due to being extremely power limited and having super low boost clocks. At 150W I'd expect the 3080M to be like the desktop 3060 Ti, and at 200W (if there is one) to be like the 3070.
Setting a minimum bound of 85% works too, and should be the same, since I think I have yet to notice the dynamic resolution scaler drop below 900p. I used 65% (720p) just in case there are more demanding sections later in the game.
980 desktop on mobile was pretty close I guess too. But TDPs were different.
Ah k, you just used that link, I was hoping for something within U.S. Shipping will take a while, oh well!
I figure the core count would be correct and more than the desktop 3070, so we can figure out our own Mhz that will work with our cooling limitations. The one advantage mobile parts seem to have had over their desktop counterparts has been more VRAM, like the 980m had 8GB while the desktop 980 only had 4GB, and now it seems like we will have 16GB, so that is a plus! If I can get the 3080 mobile at 1600Mhz ish, I don't think that would be bad at all! It would at least be a worthwhile upgrade from the 1080.
Ah, that is correct! I forgot about the 980. Good point!
So I did some power draw testing for the 1080 gtx, using the CMD prompt:
nvidia-smi.exe -i 0 --loop-ms=1000 --format=csv,noheader --query-gpu=power.draw
Not exactly sure where it pulls the numbers and how accurate it is, but it is still helpful!
First is 1830Mhz @ 0.925V
TimeSpy Graphics Test 1: 145 ~152Watt
TimeSpy Graphics Test 2: 150 ~ 175Watt
Cyberpunk 2077 (in the city just standing): 120~133Watt
This voltage requires me to use full fans for the voltage to not get to 90 or so degrees.
Second was 1502Mhz @ 0.762V
TimeSpy Graphics Test 1: 98~104Watt
TimeSpy Graphics Test 2: 104~112Watt
Didn't do Cyberpunk 2077
This voltage requires me to just use the auto fans and I don't have to worry about it.
I re-pasted and was going to try the RTX heatsink, but after looking at the difference between the P775 fin size and the RTX fin size, it seemed pointless to try the RTX heatsink. Looking at
yrekabakery temps with the RTX heatsink, it kind of sucks that the obviously larger P775 heatsinks don't seem to make that much of a difference. Maybe I should just throw on the RTX heatsink for giggles and see if it magical.
Maybe the 17 blade fan will make the difference? Kind of sucks that it looks like the heatsink setup has a cap around 120ish or so Watt before full fans are needed - but then it holds steady up till 150ish Watt before the cooling system can't keep up and it just continues to get hotter.
Quite a big difference from the thermals I'm getting on my 1080. Like I said before, I'm topping out at about 78C on max fans with a +170/+400 OC. These are temps and clocks near the end of a Time Spy stress test (20 minutes long):
It varies between 1950MHz and 2000MHz on the core as it's bouncing off the power limit, but it does stay at 190W the entire time.
Most important thing is making sure that the GPU core makes good contact with the heatsink. When I pressed the MXM module to the heatsink outside of the laptop while shining a flashlight, there was a .5-1mm gap between the heatsink and die caused by too-thick stock thermal pads. That gap shrinks if you use the paper clip mod, but it's still not ideal. When I took the thermal pads off and tested it again, there was no gap whatsoever and the core was making perfect contact with the heatsink. That's why I decided not to bother with thermal pads altogether and switched to K5 Pro, which conforms to any size gap ≤3mm and doesn't cause fitment issues since it is highly compressible, being basically a gummy paste. The perfect core contact also allowed me to get the full benefit of using liquid metal on the GPU, lowering temps further.
One thing to note, if you do go this route of bringing the heatsink and core closer together, don't do the paper clip mod because the contact is already perfect without it. Otherwise, the increased pressure could crack the die.
I don't think it's worth switching to the RTX unified heatsink. The most important thing is GPU core contact as stated above, and all the heatsinks are equally affected by this problem with the stock thermal pads. You've already modified your chassis to accommodate the P775 heatsinks, so might as well stick with them and fix the contact instead. But if you do use the RTX heatsink, make sure you use spring-loaded screws for the GPU, as that heatsink doesn't use retention arms on the GPU, otherwise the heatsink pressure won't be adequate.
When clocking there can be variations inbetween cards too.
Separate names with a comma.