Nvidia's Exciting New 50 Series Laptop GPUs

Nvidia has announced only part of their standard range of dedicated mobile GPUs, the 5070 through 5090. However, this year they are bringing us one more mobile graphics card to bridge the gap between the 5070 and the 5080, the 5070Ti.

Last year the 4070 was only marginally better than the 4060, but there was a big performance increase and price increase to the 4080. So, we like the introduction of a 5070Ti. The motherboards that will be used by manufacturers will have two variants, one that supports the 5070 and yet to be announced 5060 and 5050 (which we all know are coming next), and then one for 5070Ti and above. This means it’s likely that we’ll see some models offering only the lower or higher range of cards.

DLSS 4

One of the biggest announcements with the 50 series cards is that they will now have access to DLSS 4, which enables Multi Frame Generation: a method of generating up to three additional frames per traditionally rendered frame. Something to keep in mind though, is that this isn’t available in most games yet, and you’re still only getting the responsiveness of actual rendered frames. Even if your FPS number is much higher, you’re going to be experiencing the same latency. On the other hand, one more benefit of DLSS 4 is that it improves single frame generation for both the 40 and 50 series cards, supposedly reducing VRAM usage. This is a big win if true. Clearly, we’ll have to independently test this to find out if this is truly free frames or if there is a cost to it.

Specs

As far as specs go, the 5090 will come with a TDP range of 95 to 150W. But, we’ve heard from manufacturers that they feed these GPUs up to 175W. We think the delta is the boost. They also have 24GB of GDDR7 memory, and over 10,000 CUDA cores. They are also claiming 1,824 TOPS, which is almost triple what the 4090 was capable of. The memory bus, or interface, width of all the cards appears mostly the same from last generation, except the 5080 did get a boost up to 256-bit this time around.

Nvidia Laptop 50 Series Specs

VS The 40 Series

Nvidia 40 Series Laptop Specs

So far, the main differences between the 40 series and 50 series seem to be a significant increase in memory and AI TOPS. However, just because there is such a big increase in TOPS doesn’t mean the standard rasterization games rely on will see anywhere near this increase. For example, we are only seeing an 8% increase in CUDA cores from the 4090 to the 5090. In fact, because they didn’t mention rasterization, it has us a little worried. When showing their desktop benchmarks, Nvidia showed less of an improvement in the games that didn’t support DLSS 4’s multi-frame generation or have DLSS at all. This aligns to our concerns that there won’t actually be a significant improvement of truly rendered frames when upgrading to the 50 series. Check out Jayz2Cents video for some great analysis on this, they estimate about a 25% increase rather than the double Nvidia was highlighting in those instances utilizing DLSS 4.

Difference in Performance Between Desktop 40 and 50 Series

When it comes to memory, we weren’t happy that the lowest mobile card announced, the 5070, is still coming with 8GB of memory at 128-bit. This means that I’d expect to see the later release of the 5060 or 5050 to have the same 8 and 6GB of memory respectively as the prior generation. The reason we are disappointed is that we do expect these GPUs to perform better and be more capable of handling higher resolution gaming. Higher resolution rendering requires more graphics memory.

VS The Desktop Cards

When comparing these to the desktop 50 series cards, the desktop GPUs have more memory. 32GB on the 5090 instead of 24GB, for example. You may also notice they did not list the memory bandwidth for the mobile GPUs. Josh has requested this from Nvidia.

Nvidia Desktop 50 Series Specs

Conclusion

As a whole, this Nvidia release seems to be the most geared towards AI and machine learning, which is not a shock when you look at how often that theme came up at CES this year. Hopefully for data scientists this should make model training more viable on laptop GPUs.

We are also very excited for the 50 series and DLSS 4, but as always, we want to help you see past the hype. You deserve to make an informed decision on what's right for you and your wallet. So we hope this helped, thanks for reading!