NVIDIA Announces GeForce RTX 30 Series for Laptops: Mobile Ampere Available Jan. 26th

[harlan4096]

As the cornerstone of their CES 2021 keynote, NVIDIA today has announced their GeForce RTX 30 series graphics adapters for laptops. Based on the same

Ampere architecture and the same GPUs as NVIDIA’s existing desktop GeForce cards, the RTX 30 series for laptops will bring Ampere’s features and much of its performance to laptops, just tuned to the sub-150 Watt power requirements that come with a laptop. Laptops will be getting versions of the RTX 3080, RTX 3070, and RTX 3060, with the first of these new generation of laptops set to become available later this month, on January 26th.

Though an obligatory step in the roll-out of a new GPU architecture, the release of laptop parts is an important one for NVIDIA and its OEM partners. Laptops represent an ever-growing share of the PC gaming market, and perhaps just as importantly for NVIDIA, it’s a market that, above entry-level discrete graphics, they have virtually complete control of. So a big part of NVIDIA’s fortunes hinge in part on getting laptop gamers to upgrade to new machines with the latest and greatest in NVIDIA graphics technology. In fact in some respects laptops are more important than desktops, as the low power needs for laptops means that these machines get some of the best chips NVIDIA bins – which in turn tends to be reflected in the prices of some of these laptops.

For their latest generation, while not being explicitly noted by NVIDIA, it looks like a change in nomenclature is at hand. In the past couple of generations the company has simply referred to their mobile parts with the same name as their desktop parts – e.g. GeForce RTX 2080. However for the RTX 30 series, we’re seeing parts listed as “laptop GPU”, e.g. “GeForce RTX 3080 Laptop GPU”. It’s a small change, but an important one: as we look at the specifications, it’s clear that NVIDIA isn’t always using the same GPU in a mobile part as they have its desktop counterpart, so any kind of true equivalency between laptop and desktop has gone out the window. In practice, laptop graphics adapters have always been their own product family, and now by returning to having distinct naming and specifications for their laptop adapters, NVIDIA has gone back to more clearly defining this structure.

GeForce RTX 3080 Laptop GPU

Diving into specifications and features, the new GeForce RTX 30 series adapters cover the RTX 3080, RTX 3070, and RTX 3060 families. And though NVIDIA’s own specifications don’t explicitly list what GPUs are driving these configurations, based on the number of CUDA cores and other factors, it looks like NVIDIA is using a mix of the existing GA104 GPU and what would be their new GA106 GPU to drive these parts. Notably, RTX 3080 in the desktop uses a cut-down GA102 part, not GA104, and with GA102 being unsuitable for anything short of a quite literal desktop replacement notebook, GA104 instead sits at the heart of the RTX 3080 Laptop GPU as well as all flavors of RTX 3070.

The leading member of their new laptop adapters, the RTX 3080 Laptop GPU – or what I'm calling 3080L for short – appears to be based on a fully-enabled GA104 GPU. All 48 SMs are enabled, as are all 8 memory channels, so 3080L stands to be GA104 in its top form. Clockspeeds for mobile parts are highly variable owing to differences in laptop cooling and TDP headroom, but officially NVIDIA is giving the boost clock for the 3080L a pretty wide berth, with a range of 1245Mhz to 1710MHz.

Meanwhile, for memory NVIDIA is using tried and true (and lower power) GDDR6. Unfortunately memory clocks are not disclosed – so it’s not immediately clear if this is going to be 12Gbps or 14Gbps memory – but for the 3080L the adapter will come with 8GB or 16GB of memory on a 256-bit bus. The inclusion of a 16GB option is a welcome change after years of 8GB being the largest laptop configuration, and it’s interesting to note that due to laptop space requirements, this strongly implies that NVIDIA is using 16Gb memory modules, which just recently hit the market, rather than their usual 8Gb modules (which need to be in a clamshell configuration to allow for 16GB of VRAM).

Finally, while power consumption/TDP is largely up to the OEMs, NVIDIA’s official guidance is that the 3080L is designed for 80W to 150W+ operation. Towards the lower-end will be Max-Q laptops and other thin and light designs, getting some of NVIDIA’s finest silicon. At the other end of the spectrum will be true DTR laptops, which can handle the 150W or more of power consumption from a graphics adapter with near-desktop performance.

GeForce RTX 3070 Laptop GPU

Following the RTX 3080 Laptop GPU is the RTX 3070 Laptop GPU (3070L). This appears to be a cut-down version of the GA104 GPU, with fewer SMs enabled and a narrower TDP range. Overall, 40 SMs are enabled here, with boost clocks ranging from 1290MHz to 1620MHz. Meanwhile memory is available in just an 8GB GDDR6 configuration, with the complete 256-bit memory bus enabled.

With an 80W to 125W TDP window, the biggest change here is likely to be felt at the high-end, where relative to the 3080L, the 3070L has both lower peak clockspeeds and fewer SMs. NVIDIA doesn’t offer a lot of performance guidance here – clearly the new laptop parts are going to be a good deal faster than the old ones, based on what we’ve seen in the desktop – but it’s worth noting that NVIDIA is pitching the 3070L as being well-suited for 1440p gaming. Which is the first time we’ve seen them seriously float the idea of 1440p with anything less than an 80-class laptop graphics adapter.

GeForce RTX 3060 Laptop GPU

Last but not least in the new mobile family is the GeForce RTX 3060 Laptop GPU (3060L). Like the desktop GeForce RTX 3060 also announced today, the specifications for the RTX 3060L point to it being based on a new GPU, which in NVIDIA’s usual naming schemes would be GA106. At this point we don’t have any detailed information about GA106 in particular, so we only know how it’s configured for the desktop and laptop 3060s respectively.

In any case, the 3060L comes with 30 SMs enabled, for a total of 3840 FP32 CUDA cores. With the usual disclaimer about TDPs applying here as well, the adapter’s boost clock is rated for anywhere between 1283MHz and 1703MHz. This would put overall performance, in terms of FP32 FLOPS, between 9.9 TFLOPS and 13.1 TFLOPS.

The memory subsystem on the 3060L has been cut down relative to the other laptop parts. Here there’s just a 192-bit memory bus, meaning that memory can only be installed in multiples of 6GB. In this case, NVIDIA has opted for 6GB for 3060L whereas desktop 3060 gets 12GB. 6GB shouldn’t be too much of a deal-breaker here, but it will stand out a bit when NVIDIA is doing more on the desktop. Otherwise, as we don’t have specific information on memory clockspeeds, it’s not clear right now if 3060L comes with 12Gbps or 14Gbps GDDR6 – though as the lowest-power SKU, the former is increasingly likely.

Speaking of power, NVIDIA’s official TDP guidance for 3060L is 60W to 115W. This puts the floor below both the 3070L and 3080L, and interestingly, also 5W below that of the RTX 2060 it replaces, which had a floor of 65W. Meanwhile 115W will be more likely to show up in DTR-like designs, which will be better able to reach the higher-end of the part’s rated boost clock.
...

Continue Reading