How will Ampere scale on laptops?

Considering how well Ampere seems to undervolt, I'd expect that to happen. I believe the point it starts to get really inefficient is above 1850 Mhz, and it costs an extra 100 watts to push to 2000 Mhz (for the 3080). I'm fine taking a minor (10% or less) performance hit if it cuts the power draw by a third.

 

I expect we will see more power gimped 3070 or 3080 Max-P. Don't forget that Dell throw in a castrated 2080 Super Max-P into their soldered mess m17 R3. More ODM/OEMs will follow same paths and will do exactly the same.

Remember a bellwether is an individual who either leads or indicates trends;

 

Castrated as in its 150w as opposed to 200w?

 

When I go for Max-P I go for maxed out cards. Can't get that with castrated cards. On top you pay more for less.

Doesn't some of the custom 3080 desktop cards come with 380W TGP? 150W cards will be around 2.55 times less TGP. And not so sure the 20GB 3080 "if it happens" will be on 320W. All boils down to Big Navy.

 

If thermals permit i would shunt mod the M17R3. Why rely on the manufacturer when you can rely on yourself.

Thats all assuming i was daft enough to get an alienware in the first place.

 

The normal is that the manufacturers design cooling around the added TGP. Often with a 10-15% headroom on top for coming dust, weared out components etc (to reduce repairs cost afterwards).

150w+15% = 172W

200W+15% = 230W

And you have the unified heatsink. More heat from Gpu will affect Cpu and so on. Imagine the desktop graphics card had to share the cooling with the Cpu in a desktop. Nasty?

 

Its not that straightforward though, besides, alienware never had any qualms with their laptop cpus running at 99°C

Heatsinks aren't built based on the tdp requirement. They are built to the best the laptop chassis can fit then sometimes downgraded and downcosted (smaller less heatpipes [Asus Zephyrus], less heatpipes [MSI GE, Asus Scar]) based on the sku.

 

Based on the fact that TGP of the 3070 is around 220W and TGP of the 2080Ti is around 250W, and the performance of the two is about equal, it doesn't really seem like the 3070 is that much more power efficient at the same performance level. Are we really going to see Ampere perform that much better than Turing at the same TDP? Seems like it might be marginal improvement on efficiency but nothing staggering unless I'm completely misunderstanding something. Seems like around 13% power improvement for the same performance level.

The big draw of the Ampere desktop cards seems to be that they are WAY CHEAPER for the same or better performance level as Turing, it doesn't really look like the perf/watt is that much better.

EDIT: also the 2080Ti is powering 3GB more RAM.. that's even worse as an 8GB 2080Ti would shave a small amount off that 250W and probably not see any performance decrease if they could keep the memory bus the same width.

 

I’ve been saying this since the ampere presentation. All they did was feed the core more power compared to Turing. The efficiency gains are so small, which means laptops are going to see probably the worst “improvement” in a “generation”. Fermi to Kepler, Kepler to Maxwell, and Maxwell to Pascal were huge. Pascal to Turing was meh and now Turing to Ampere is even more meh when you look at performance vs power consumption

 

And that is why I have been say nvidia will NEED to bump up mobile tdps vs Turing to gain any performance.

GaN chargers are here so much more powerful adapters can be made without making them bigger. And pathetic laptop heatsinks need to go, they can be much much better than what we have now. We are far from any physics limitations, just limitations in the aspect of innovation.

 

Probably why they are chomping at the bit to go TSMC 7nm to squeak out a little more power efficiency.

 

More like +20% and will increase up to around 25% power efficiency depending on the resolution.

https://www.igorslab.de/nvidia-gefo...mpere-kann-auch-sparsam-und-knuffig-klein/12/

 

I wonder if that 2080Ti is boosting past "stock" TDP of 250W in 4K. I suppose at any rate it seems like based on the game performance as well the 3070 was winning most of those. It's actually a 20% decrease in power at 4K if you go the other direction instead of calculating increase. I wonder if the 2080Ti was boosting higher at 4K or using more of it's 11GB pool of memory causing higher memory usage power increments. GDDR6X is supposed to be more efficient than GDDR6.. with some perhaps accounting for frequency muddling that a bit..

At any rate, so.. TDP for TDP and frame for frame... we may be looking at roughly 30% or so more performance at the exact same TDP, which would be interesting, though not sure it would be enough for people to suddenly dump their Turing notebooks.

The decision is probably more clear on desktop where TDP doesn't mean as much if people already have the PSU overhead available. I guess if you also had to buy a new PSU it might make the value proposition a little less interesting.

 

Undervolted AMD RX 6800 with 183 W TGP offers clues about an upcoming RX 6800M laptop-grade GPU matching the desktop RTX 2080 Ti
http://forum.notebookreview.com/thr...00-gpus-on-oct-28.834078/page-8#post-11059699

From the link...

We know that Nvidia is already planning to release mobile versions for the RTX 3000 series, but we are not really sure if Team Green can provide decent performance gains over the 2080-series given the limitations imposed by the 8 nm fabrication process from Smasmung. AMD could have a big advantage here, but it also needs to be mindful of the pricing schemes.

Maybe we will see a huge failure both from Nvidia and AMD for coming gamingbooks Wouldn't it be nice with yesterday's desktop performance in tomorrows notebooks, LOOL

 

I'm breaking my usual cycle of upgrading every 2 years, because this time I'm keeping my current machine until PCIE 4.0 and 5nm GPUs are on the table.

See you in 2023.

 

Rumors....

NVIDIA GeForce RTX 3070 / 3080 Mobile for AMD Ryzen 5000H CPUs videocardz.de

3080 Mobile as castrated chips. Back to M branded chips (not equal the desktop chips as Maxwell-Pascal-Turing), but with the Max-Q stinker as an awful attachment in the M graphics lineup. Now with double M vs the old days. Nice, Nvidia sneaked in Max-Q to increase their already high profit. This to be able to help the notebook manufacturers create Apple designed gamingbooks.




Exclusive: NVIDIA GeForce RTX 3000 Mobility Series Lineup, Roadmap 1H 2021 And AMD CPU Support! wccftech.com

We have been told that the RTX 3080 and RTX 3070 mobility GPUs will be fast-tracked to land in January 2021, however, the GPUs will initially only be available in ASUS and MSI flavors. HP, Dell, and Lenovo have reportedly decided to wait until April to support these new GPUs as they want to pair it with launches that are happening then.

My source also mentioned that they expect the RTX 30 mobility series to have a very short life cycle (considering there are rumors of NVIDIA planning to shift to 7nm TSMC for its SUPER variants of the RTX 30 lineup).

 

Glad i didn't bother waiting before i dropped my money on a laptop with a 2080 super then. These look nothing like their desktop counterparts.

 

Oh boi, somehow i thought that this will be happening but never thought it would be THAT bad, thank god the 2080 in my book will be sufficient for a while, especially if DLSS get's (hopefully) more widespread support to squeeze some more performance out of it.

Gesendet von meinem GM1913 mit Tapatalk

 

This makes sense, new laptops seems to always hit in April-June en masse.

 

Those mother flippers! 2500 less CUDA cores, only 256-bit bus (vs 320 bit) and the slow old GDDR6 .... damn lairs that's a castrated desktop RTX 3070.... what a waste

 

Lol called it as soon as the September ampere presentation was given that you’ll be lucky to get marginally better than the 2080 Max Q. Ampere is way too power hungry to be called a generational leap in mobile.

 

Going by this:
https://www.reddit.com/r/nvidia/comments/jn878f/small_beginners_undervolting_guide_for_rtx_3070_fe/
This one says that its around 140W for 1900Mhz:
https://www.reddit.com/r/MSI_Gaming/comments/jkvdwd/msi_rtx_3070_ventus_2x_oc_undervolt_results_wow/
And 175-180W for 1800-1900Mhz:
https://www.reddit.com/r/nvidia/comments/jkdib3/quick_rtx_3070_fe_undervolting_results_85_pl187w/

180W after undervolt for around 1800Mhz, didn't really expect the GPU to be THAT power hungry, more or less it seems :/

So, considering the rumored 80W from those websites, what are those 3070 going to clock in, 1200Mhz, with maybe 1300Mhz boost, or is Nvidia going to use the good 7nm dies(not even sure if they make the GA104 in 7nm..) for laptops?..
That or saving pretty good dies, but even 140W for 1900Mhz, how will that end after being capped to 80W :/

 

Hint: the 3070M will perform like the 2080M and the 3080M will perform like the 3060. It’s just like the old days. Nvidia should return to its old naming convention for laptops because insinuating that these mobile chips are anywhere near their desktop counterparts is false advertising and terribly misleading

 

In which case, if they are using Ampere numbers, they need to use ones based on destkop models of same performance with the M suffix at the end to indicate each product level of performance.
Otherwise, yes, I agree they are being misleading and are doing false advertising (and yet, no one is doing anything to stop them).

 

These will be 8nm. 7nm (if it ever arrives) will be reserved for their super refresh.

 

There shouldn't be max-p variants for Ampere. As I understand max-p cards should have the same graphic processor as it's desktop counterpart (GA102 for the RTX 3080). max-q is the old "M" series?

 

Max-P and Max-Q use the same graphics processor as desktop, just with different power limits and different memory clocks.

 

I meant in the sense that if say mobile 3080 is on the level of desktop 3060, it should be designated as 3060M.

 

You'll need to wait for the normal 3080 for mobile and see how they behave.. MAX-Q 3080 is basically a 3070 or less..

 

If it has a max TDP of 150W it will perform worse than the new 3060 Ti which pulls around 200W. Probably the "full 200W" mobile RTX 3080 will perform similar to the desktop RTX 3060 Ti (that is a bit better than the desktop 2080 Super) and the 150W mobile RTX 3080 version will perform similar to the desktop RTX 3060 (similar to the desktop 2080 Super) and the max-q RTX 3080 will perform similar to the desktop RTX 3050 Ti (similar to the desktop 2070 Super)..

 

Benchmarks today show the 2080 super and 3060ti being roughly the same and trading slight blows. If the mobile 3080 is doing desktop 3060ti performance then I see little reason to upgrade.

 

I would bet it would perform better than the 3060.

If it does, sucks ass

 

Probably 16 GB of VRAM and better RT performance because the RTX 3080 mobile will have second gen of RT cores. Also more SM (CUDA cores) which means that the mobile RTX 3080 will scale much better than the RTX 2080 Super when overclocking.

 

There's also a chance it could just have more SM/RT cores just clocked lower which might put it over the edge. I wouldn't expect anywhere near desktop perf though looking at how much power the desktop cards are consuming.

 

I wouldn't be amazed if the RTX cards for Clevo's would push 250w to be honest since we have 2 bricks for the X170SM-A.

 

The 200W Mobile 3080 (not the crippled Max-Q variant) should easly beat 2080 Ti



https://www.kitguru.net/components/...nvidia-rtx-3060-ti-founders-edition-review/4/

 

250w Clevos be like

 

@joluke
Bro, and how did you get a stable 200mv undervolt on that CPU?! I can't for the life of me push past 120 on my 9900kf (Crash in Battlefield)

 

Then they would have to re-design the whole MB with better power delivery, heatsink + swap to 2x330w bricks The trend for gamingbooks go the opposite way, bruh. Thin, slimmer, and neat disgusting as Applebooks. See average of class Gaming.

Nope.

 

That sucks.

 

Unless they switch to amd cpus, which they totally should

 

I wish

 

How come you get that conclusion? The 3070 is 220W and the 3060 Ti is 200W. The mobile 3080 at tops is going to be 200W (or as low as 150W) so the performance should be on pair with the 3060 Ti. The mobile 3080 is going to have the same memory memory bandwidth (448 GB/s) as the desktop 3060 Ti , although the mobile 3080 is going to have ~20-30% more CUDA cores but at a slower clock so the core will perform the same as the 3060 Ti's core which is slower than the 2080 Ti. For video cards within the same generation watts = performance level. Unless the mobile cards had better binned chips, but that's unlikely

 

3080 mobile at 200W with better binned silicon. + Ampere is from before more power efficient than Turing (2080 Ti) silicon.

See the power consumption between 3070 vs3060 Ti. Opposite vs. specs.

https://www.guru3d.com/articles_pages/geforce_rtx_3060_ti_founder_edition_review,6.html

Edit.
200W Mobile 2080 Super scores around 10800 in Time Spy graphics. The 3060 Ti around 12000. I doubt that a major graphics upgrade only will provide around 12.5% increase in graphics performance Going from Turing to Ampere isn't the same as from 2080 to 2080 Super Mobile

 

Interesting chart, but seems weird to me. Here are a few from Igor's lab:







https://www.igorslab.de/en/nvidia-g...way-and-point-on-quiet-path-from-this-claw/9/

The biggest problem to me is the power limit of 200W. it wouldn't be the first time that Laptop users get so minimal performance improvement over generations. I remember, 780M ---> 880M, or 1080 ---> 2080 (150W)

Ampere is a huge jump in performance but also is a huge jump in power consumption. From 215W for the RTX 2080 to 320 W for the RTX 3080. The real jump in performance in laptops might come with the 7nm architecture in mid 2021. Probably with 7nm Nvidia would be able to get RTX 3070/RTX 2080 Ti performance at 200W

 

To be honest I'm really considering getting a RTX 2080 for my P775DM3-G and a 9900KF and just invest the rest on a desktop. If it's going to be such a crappy leap of performance (12%) it's not worth to give so much money for a new laptop. We'll see

 

3070 is capped. Maxed GA104 chip offers 6,144 Cuda Cores. Expect 3080 Mobile will get a maxed out 3070 chips and maybe brand it GA103 http://forum.notebookreview.com/thr...scale-on-laptops.834039/page-12#post-11060748

This is also from Igors lab. Undevolted 3070. Expect binned silicon for mobile graphics will run even more effecient... http://forum.notebookreview.com/thr...scale-on-laptops.834039/page-12#post-11054740


Ampere Architecture Whitepaper

 

If this is true I will think on investing in a new laptop because the jump from 1080 to 3080 mobile will be worth it imo

 

Laptop chips aren't binned. They seem to be more power efficient because unlike desktops, the tdp only covers the gpu + vram, not the whole gpu board fans and all like on desktops.