Nvidia GeForce RTX 2070 SUPER and RTX 2080 SUPER

It depends a lot on the game. In RTS games higher than 60 is nice, but not that big of a deal. Like you say in FPS games though it's a big deal. Fast motion really makes higher than 60 Hz shine.

With that said I prefer more pixels. I'd take 4k 60 over 1080p 144 any day, but yes I do notice a big difference between 1080p 60 and 144.

If only 4k 120 were easier to come by. My cpu can't handle 120 fps consistently though.

I still remember as a little kid how glorious 640x480 Windows games looked vs 320x240 res DOS versions of the same game.

 

For competitive FPS, given the same skill level, the difference between 60FPS and 144+FPS is night and day.

Don't hurl personal insults at people that like to upgrade hardware, we have enough immature hyperbolic garbage being slung around in a pc-masterrace-ish stance already.. (crapbook/cancer/nvidiot/amdumb/microsloth/microshaft/etc..etc..)

 

oh it was just in general and not directed at anyone more or less the scene....Not everyone has the same vision I personally wear glasses and have poor vision and cannot see the difference between 1080p and 720p.....not everyones the same but point taken.


edit but if you have a killer system yes I stand with my opinion.... it's dumb to upgrade ....money could be spent elsewhere

 

144Hz works great if you can push 144+ fps.

 

How is this even a thing? My P775DM3-G with a GTX 1080 scores about 20.5k in Firestrike (graphics score) - about 10% difference. 2080 Max-Q? More like Max-Scam imo.
I get it, it is supposed to be thin and light, but still...

 

You're preaching about how a luxury you don't get to experience on a daily basis isn't worth it. How would you know?

You honestly sound like your trying to convince yourself, not us. It's best to buy what you can afford for the level of gaming you can accept, and let others do the same.

To flip the mirror, the GTX 1050 has negative longevity as a purchase in 2020. But it is what you bought and I will leave you to it as your decision.

 

Sorry that I have to vent my frustration right now, but I'm really upset that I don't see these being matched with mobile Ryzen 4k. My GT72 is shot for games, but I'll play my PS4 to the PS5's launch before I'll buy an Intel mobile CPU.

Has anyone heard anything I missed?

 

frank azor said No high end GPUs with Mobile Ryzen till 2021 or later. The best mainstream you can do right now is 4900H and RTX 2060. Sager has a model with the 4900h and 2070(non super)

 

Link? All I see is desktop CPUs.

https://xoticpc.com/collections/cus...AMD+Ryzen&rb_snize_facet1=NVIDIA+RTX+2070+8GB

It's probably as good as the prices a 4900H would have, but it's newness makes me want to hang back and see if there is issues. Also, bringing desktop CPUs into the mix brings Intel as well ...and I have to refigure the spec/price math.

Do you think the mobile 2070 will last until mobile Hopper [since buying now means skipping mobile Ampere] with decent performance on nextgen* ports? [*figure mid 2022]

 

Many sources online, here is one of them:

https://pokde.net/gadget/laptop/amd-ryzen-4000-laptops-not-rtx-2070-or-higher/

 

I would say it depends on your resolution, frame rate, and graphical settings expectations. FWIW, the PS5 and Xbox Series X have broadly equivalent GPU horsepower to the desktop 2070 Super and 2080 respectively, while the mobile 2070 performs closest to the desktop 2060.

Also I doubt Hopper will launch for mobile in 2022. Mobile Turing Super just came out, and mobile Ampere will probably release sometime in 2021. The most recent rumors put desktop Ampere at 2H 2020/Q1 2021, and mobile parts typically trail desktop by a few months. So a 2022 release for mobile Hopper will only be a year after mobile Ampere, which is way too short. I’d expect mobile Hopper to be no sooner than 2023.

 

I really, really wish there were more powerful GPU's coupled with Ryzen CPU's. My laptop has a GTX 1660 TI and a i7-9750H, it's fine, but when I saw the 4900H benchmarks I regretted buying it, and started looking for a laptop with the Ryzen CPUs. I found some that seem decent, but end of the day I'm not spending $1,400 to replace a laptop that was $1,000. That's a marginal upgrade, if I am going to replace my current laptop it's going to be with something bleeding edge that is a substantial upgrade.

 

So that was a typo on your part?

 

No, mainstream laptops won’t see better than a 2060 with Ryzen. Only boutique brands like sager will go one step higher and pair the 4900HS with a 2070.

@Meaker@Sager can you confirm this?

 

No, that's not sort of thing I can confirm.

 

AFAIK Asus has an exculsivity deal for 6 months to use the better binned 4800hs/4900hs in their laptops. We will probably see some more laptops with 4900hs and 2070s/2080s later this year from other manufacturers.

 

Maxq + shunt mod is where its at. I'm getting 26k graphics in firestrike and 9.9k in timespy.

Will be pushing higher when the 330w adapter arrives (courtesy of gentechpc)

 

I think I will also shunt mod my 2070, I'm a bit bummed that its so limited at only 115W, just got the laptop up and running, but at 115W I'm looking at a clock of around 1700Mhz at 825-850mV, can't go above that due to being so power limited.

 

At 850mv im doing 1860mhz with a +140mv overclock.

Have you tried overclocking yet? I can run +140core and +800 memory. With a non max-q card you should be able to do +1000-1200mhz on the memory

 

Just made 2 runs of superposition after doing a clean install and installing all the drivers.
Tried all stock and the clocks would be around 1600-1650Mhz, then I just gave it +100Mhz and enabled the OSD on Afterburner, not even touched memory.
Second run then ran at 1700-1750Mhz and I just opened the curve editor and 1750Mhz puts it around 825-850mV, I think it might be able to run 1800Mhz around that voltage.
It should be non max-Q since its a "mxm" card in a GT75 chassis.

Initial score was 4700, then with +100Mhz it went to 49xx 1080p Extreme.

So tempted to just shunt mod and hope that I have a decent core that will hold 1900+Mhz.

 

Is this a trick when you feed lower power readings to controller so that it gives higher values than the maximum TDP? Meaning you can get closer to "full" power but at the cost of potentially very high temperatures.
I did a bit of research on the internet, but it left me a bit confused whether I understood correctly.

 

The GPU uses one/two/three resistors,usually called shunt resistors to measure current, and it also measures voltage, so the GPU can calculate how much power it is utilizing, given that VBIOS mods are harder and harder to do, lowering the value of those shunt resistors is the easiest and fastest way to have more TDP, because with the lower resistor values the same current will give a lower reading of the current that what the GPU thinks its using and so end in effect with a higher TDP.

 

What is the highest stable overclock you can manage? Run unigene heaven and increase the overclock until it crashes, do the same for the memory overclock.

It messes up the shunt resistors so the gpu thinks its using less power than it actually is, effectively removing the power limit.

 

I know, but I will only have spare time next weekend, but I sure will thinker it it. I will then report back with some data.

 

The VRMs and power delivery components run quite hotter with a shunt mod. I had to completely replace themalpads with higher performance ones, to keep components from cooking underneath the heatsink. As well as add additional pads on some components that weren't cooled before.

Just make sure things are properly cooled and check up on them after a while. As I've seen cards with discolored PCB near the VRMs due to the above-mentioned.

 

I'm almost OCD with temperatures, and going for a shunt mod I wont go past the limits of the VRM's, at most 150Watts, the GPU has 5 for the Vcore and the other two that have dual MOSFETs seem to be for VRAM rail.
So far the stock 2070 at 115 isn't going above 62ºC under load, and thats with stock thermal paste.

 

When you do the shunt mod the fast and dirty way, simply shorting them, the card will report a very low TDP. Mine reports around 20W when pushed in TimeSpy.
So in my case the VRMs and power delivery had to be babied.

 

Mine is reporting 10w at full load. and yes, i've upgraded the thermal pads on mine as well as placing additional pads to let some of the heat from the mosfets be dissipated by the case.

It also helps that the GS75 has a dedicated heatpipe for cooling the mosfets

 

Better be safe than sorry, doing these precautions will also help with the longevity for the rest of the components. Especially when you can't simply replace the GPU itself and will be forced to buy an entirely new motherboard.

 

At most I'm aiming for a final 150Watts, because I finally got some time, and it is indeed pretty power starved, scores are middle of the road, nothing crazy, tried going with +100 on core and memory, always starts at 1880-1900Mhz then drops to the 1700Mhz mark, with drops to 1680Mhz, will need to play with the curve editor to find max stable clocks in the 825mV.
Temps are always in the 65-70ºC, still on factory paste.

 

Did you lock the voltages by hitting L?

 

I'm guessing those of you with shunt mods are seeing "pwr" as the throttle reason in GPU-z, despite reported power being no where near the power limit.

I get the same on a modded 2060. I get power related throttling despite only 35W of the 90W TDP being reported (usually 130W actual, but it does sometimes drift higher). I think I figured out why. The VRM responsible for the core voltage reports its current readings to the GPU core, which bypasses shunt mods. Going around 50% higher than default TDP starts tripping the core power limit.

It's very easy to mod the reported current. I plan on trying it tomorrow and will make a guide if it works. Clevo uses all the same parts as the desktop Founders boards so this has a pretty big affected user base.

Also for those worried about blowing out FETs, the FETs used on the clevo 20 series all have on die temp sensors. I have 2/3 of the 2070's power circuit (will have a full 2080 circuit soon) and thermal shutdown gets tripped at about 140W unless I keep the card under 80C. A 2070 should be fine up to around 200W.

 

Only seen VREL, when card is set for 1.063v, with all 3 shunts shorted on this 1080N.

 

Yeah 10 series doesn't have this issue. Only 20.

For VREL you just need to keep your card under 50C. Good luck with that pulling ~230W with power limit removed.

 

A 400W vapor chamber helps

 

In my case I get the pwr throttle reason when I run a cpu and gpu stress test at the same time. That's because the total system power draw is exceeding the 230w my adapter can supply.

I'll be recieving my 330w adapter in a week which will hopefully allow me to hit desktop 2080 performance.

 

Thats interesting to know.
Is it as simple as changing registers with something akin to RWeverything?
But the shunts are still measuring something, so in the end we might need to shunt mod+soft mod to get the best results.
Really interested in knowing a bit more about your findings.

 

I'll be testing that in a week or so once i get my 330w adapter.

I have a 190w 2080 vbios which i can flash on top of having the shunt mod.

Hopefully i dont blow up my laptop

 

As long as the card doesn't drop too much clock due to temps and boost 4.0, you should see some nice gains.

 

Shunt mods should be unnecessary, but they improve efficiency, particularly on 20 series where the shunts are in series. There is one 5mOhm shunt for total card power, then a second for just core power, but this 2nd shunt seems to not be used in any calculation. The pair of resistors that then scale the shunt reading give much more adjustment range than the shunts.

As for the VRM current reporting mod, I haven't quite gotten to it yet. I'm still waiting for FETs from China to show up to double up my core FETs (Clevo MXM 2060 has 2 core phases with doubled FETs and inductors. I moved 2 FETs to the unused phase pads and activated the phases) and until then I only have 4 total, which are giving me issues with throwing thermal trip errors already, and a VRM current reporting mod will make that worse. 20 series cards don't have Schottkey diodes on their power phases, so I am adding them from dead cards. I am also connecting my phase inductor pairs together so one FET on a phase can drive both. I also deleted the seemingly unused core current shunt. These 3 mods plus another harder to describe one should save me 2-3W on power delivery, and push thermal shutdowns to a bit higher current and core temp.

 

I've already gained 25% more performance.

Went from 7.9k timespy graphics to 9.9k

 

My 2070 is from MSI, so it has 5 phases plus two doubled phases that seem to be for the memory rail, so it should be able to withstand running at 150W versus the stock 115W.
I'm not going above that at least initially, in part because the 1080 uses an extra power connector, I will need to see if the power connector is powered even when it as the 2070 installed, because MSI might be doing something to only power it up in presence of a 2080.

I will need to take my laptop apart, but I need to use it a bit more, given that its brand new and thermal performance is decent I will refrain from tearing into it right away, or the missus might kill me.
I will also inspect what kind of hardware is used to make the rails, its seems to be some sort of DrMOS, and there is indeed what seems to be a 0805 resistor next to them, need to find what chips are being used.

 

Ok I confirmed that the core current shunt can indeed be fully bypassed without clocks locking. My added diodes and inductor pairing does work, BUT I messed up core solder connections and now the card crashes with memory artifacts on OS load. It worked briefly to at least confirm no lockout. Solder damage probably happened from adding caps behind the core (many of the caps behind the core are actually for memory), and heat gun melting core solder with a slight pcb warp. I can probably fix this. Undo bend. Hit perimeter of core with heat gun.

UPDATE:
OK turns out the issue might have only been liquid metal wicked under the core into the solder. There may be no damage at all.

UPDATE 2:
Card fixed. It did need a bit of reflow. New mods get me +205 core instead of +190. Almost 21k FS. Should be over 22k tomorrow with current reporting mod.
https://www.3dmark.com/3dm/46695145

 

You mean Time Spy, right?

Bummer Hope you can fix it.

 

Yup my bad

Firestrike graphics went from 21k to 26k

 

For you shunt modders, I emailed Monolithic Power and they actually responded to me with the datasheet for the MP86941 power FET that Clevo RTX boards use. Some info:

Current rating is 30A. The tiny chips burn 5W at 30A. The temp limit is 160C.

The 30A limit seems to correspond exactly to a 120A current limit on the core that I am observing. This would mean that 2070's have a 180A limit, and 2080's a 240A limit. I would think with shunt modding that 2070's would very rarely hit that limit, and 2080's could probably only hit it when raytracing.

So yeah I can raise my current limit, but since I am already getting emergency thermal shutdowns at 160C at max current rating, I probably should not until I get more FETs.

 

Those are some weak sauce mosfets. Is clevo fond of cutting costs?

Even my GS75 uses 5 50A NCP 303150 for the core and it a max-q.

 

In general the clevo 2080 is substantially inferior to the 1080 board in design. There was really no need to go for such a small mosfet, but it's the highest amp FET for 19.5V supplies offered by Monolithic Power. Most of their FETs are for desktops and are only rated for 16V. MP does make the VRM for these cards, so I guess clevo tried to keep it simple by also sticking to their FETs. Your FETs are similar is performance to the old TI 87350d. Idk why your current rating is higher than it. It has the same losses and package size.

 

Probably has a similar design but with higher tolerances for higher current draws.

 

At 50A they are dissipating 14Watts, so, with 5 of them at full tilt its 70W of losses, they can do 50A, doesn't mean they can do 50A sustained, NEVER trust the first page of a component datasheet, its the marketing page..