Leaked pic shows NVIDIA GeForce RTX 3070 laptop GPU, which could release soon but will probably not be as fast as the desktop counterpart

[Redaktion]

The laptop-grade RTX 3070 GPU is apparently getting 8 GB of 12 Gbps GDDR6 VRAM, but it is very likely that the lower TGP of around 120 W will limit this chip's performance, so it will not really match the performance of a desktop-grade RTX 2080 Ti. Regardless of these limitations, the RTX 3070 mobility GPU should still be a decent upgrade over the current RTX 2080 Super.


https://www.notebookcheck.net/Leaked-pic-shows-NVIDIA-GeForce-RTX-3070-laptop-GPU-which-could-release-soon-but-will-probably-not-be-as-fast-as-the-desktop-counterpart.496441.0.html

[Pixie]

I expect this generation to be the biggest difference between mobile and desktop-grade GPUs since the Maxwell days, before they really shipped the desktop-class chips in laptops. Even with a GA102 Die in the 3080, the massive cutback in TDP will absolutely flatten performance comparisons in everything but monstrous desktop-replacements - we're going to be looking at a Comet-Lake class cooling nightmare if we want anything close to full performance. It's going to sting especially hard price-wise when they start asking for GA102 or GA104 prices in laptops that offer half the performance of the desktop counterparts.

[toven]

Couldn't thay just keep that 5888 Cuda cores san those rarely used tensor and ray cores then cut the chip price to half while running its maximum potential at 120w?

[_MT_]

Couldn't thay just keep that 5888 Cuda cores san those rarely used tensor and ray cores then cut the chip price to half while running its maximum potential at 120w?

That would require a new die. Which is not going to happen. And then it wouldn't be RTX. If a game is not using those cores, they shouldn't be consuming power. Well, it depends on how good their power management is. If they keep them powered on, some current is going to leak through even though they do nothing.

[_MT_]

I expect this generation to be the biggest difference between mobile and desktop-grade GPUs since the Maxwell days, before they really shipped the desktop-class chips in laptops. Even with a GA102 Die in the 3080, the massive cutback in TDP will absolutely flatten performance comparisons in everything but monstrous desktop-replacements - we're going to be looking at a Comet-Lake class cooling nightmare if we want anything close to full performance. It's going to sting especially hard price-wise when they start asking for GA102 or GA104 prices in laptops that offer half the performance of the desktop counterparts.

Well, it's up to manufacturers to consider using, for example, a desktop 3060 instead of a mobile 3080. And pricing is in Nvidia's hands. It's clear laptops won't be able to match desktops. Those chips are just too power hungry. And it's a question just how much faster are they going to be at more moderate power levels. Nvidia could definitely launch a, say, 150 W 3080. Whether it would make sense is just a question of price.

Another question is how much it matters. Do you really need to game at "4k" in a laptop? So what if you'll have to wait a little longer for that to be feasible. Laptop CPUs are also a lot weaker than desktop/ workstation processors. Laptops can't even hope to push the same power. We'll just have to wait and see what kind of performance mobile 3060 and 3070 deliver and at what cost and power compared to 2000 series chips.

[Aymeric Denis Orella]

I wouldn't be so negative, I expect this mobile gen to bring back performance closer to their desktop counterparts (Pascal) and do much better than the lackluster Turing mobile.
I'm aiming at a big desktop replacement anyway, so I expect some serious cooling system.

[Spunjji]

Laptop CPUs are also a lot weaker than desktop/ workstation processors. Laptops can't even hope to push the same power.

Not entirely true. The difference between Renoir at 45W and the 3000 series at 65W is almost non-existent, and in games, there's no real benefit to the processors that draw more power.

I'm interested to see if Nvidia can show some of that claimed 50% improvement in performance per watt when they back Ampere away from the bleeding edge - if so, the 3070 could be quite good. An incremental improvement (~20%) over the 2080 Super would be a massive disappointment; unless they drop prices, and this is Nvidia we're talking about, so that's unlikely.