Nvidia says RTX Spark games like an RTX 5070. I checked under emulation

Nvidia's pitch for the RTX Spark is a bold one. At Computex, the company told press its Arm laptop chip games at around RTX 5070 level, with a 10 to 15% hit from emulation at most. That's a big promise, even for a chip Nvidia says will reinvent the PC for Windows on Arm, because almost every Steam game is built for x86 and has to pass through Microsoft's Prism translation layer first.

I've already run Cyberpunk at 1440p on the DGX Spark, and found out that it passes most anti-cheat tests. So I lined it up against a desktop RTX 5070, ran five games on five engines, and checked Nvidia's math. The number Nvidia quotes is real. You just have to turn on a lot of frame generation to see it.

How I tested Nvidia's RTX 5070 claim on the RTX Spark

Both chips have the same CUDA core count, but very different power budgets

My RTX Spark laptop is the 64GB N1X model, and the opponent is my desktop with a Ryzen 7 9850X3D, 96GB of RAM, and an RTX 5070. A desktop card might seem like an odd pick for a laptop chip. But both GPUs have 6,144 CUDA cores, and that's how our other RTX Spark coverage lines the two up, so it's the fairest silicon match I have.

Sure, it's not a fair fight on power or memory bandwidth, as the table shows. But matching core counts is the cleanest way to see what the silicon itself can do.

Every game ran two profiles. The GPU profile is 1440p at maximum settings with full ray tracing and no upscaling or frame generation, which piles the work onto the graphics card. The CPU profile is 1080p at low settings, which takes the GPU out of the way and shows how well the processor, and Prism, keep up.

Five games, five engines, and one benchmark that refused to run

I picked games with built-in benchmarks on different engines, because Prism doesn't treat every engine the same: Black Myth: Wukong (Unreal Engine 5), Shadow of the Tomb Raider (Foundation), Horizon Zero Dawn Remastered (Decima), Cyberpunk 2077 (REDengine), and Wolfenstein: Youngblood (id Tech on Vulkan). Wukong ran in the full game, not the standalone tool.

Metro Exodus was meant to be on this list. Its benchmark tool is supposed to run a scripted scene, but it dropped me at the main menu every time. Horizon took its place. Every number comes from each game's results screen, except Youngblood, which I calculated from its frame-time log.

With the GPU doing the work, the RTX Spark is about half a desktop RTX 5070

Four of the five games land between 51% and 62% of the desktop's frame rate

When the graphics card is the bottleneck, the RTX Spark is remarkably consistent, holding 51% to 62% of the desktop's frame rate. Horizon is the odd one out at 26%, and it barely moves at 1080p low, which says something other than the GPU is holding it back.

The hardware gives some clues. TechPowerUp lists a 2,346MHz turbo clock for the RTX Spark's GB20B GPU, though laptop makers can tune that, and mine sat around 1,850MHz for most of my runs. It spiked to 2,099MHz mid-run, and hit 2,300MHz the moment a run ended, so it clearly has more to give when power and heat allow.

Locking the desktop to RTX Spark clocks still left a 25% gap

My first idea was to cap the desktop card at the Spark's 80W. That went nowhere fast, because nvidia-smi reports a 175W minimum power limit for the desktop RTX 5070. So I locked its clocks to match the Spark instead, at 1,850 from the sustained boost clock that my unit managed.

The first locked run was useless, because Shadow of the Tomb Raider had quietly reset itself to its lowest preset. Once I had it back on Highest with ray-traced shadows on Ultra, the desktop dropped from 128fps at stock clocks to 91fps locked. The RTX Spark managed 68fps, so it's still at 75% of a desktop card running at the same speed.

Against laptop RTX 5070s, the gap shrinks to 16 to 20%

Nvidia's claim is about the laptop RTX 5070, though. My Cyberpunk 2077 run at PC Gamer's settings, 1080p with RT Ultra, averaged 35.97fps, which lines up with PC Gamer's own 36fps result for the RTX Spark. That puts the Spark at 80 to 84% of the RTX 5070 laptops PC Gamer tested. Unfortunately, I didn't have my own 5070 mobile laptop handy to grab some of my own numbers on this occasion.

That's not far outside Nvidia's 10 to 15% figure. With the GPU as the limit, the RTX Spark really is within shouting distance of a laptop RTX 5070.

CPU-limited games are where the RTX Spark falls behind

At 1080p low, the Spark keeps between 19% and 53% of the desktop's frame rate

Take the GPU out of the equation, and the consistency disappears. The GPU results sat in a tight band, but these swing from 19% to 53% depending almost entirely on the engine.

To be clear, this isn't a pure measure of Prism's cost. The desktop's Ryzen 7 9850X3D is one of the fastest gaming CPUs around, so part of the gap is emulation and part is simply a faster processor. Separating the two needs different tests, which I'm saving for a follow-up.

Horizon Zero Dawn Remastered is the worst case, and Youngblood is the best

Horizon is the game that worries me most for anyone buying an RTX Spark for open-world games. It only went from 29fps to 38fps when I dropped from 1440p max to 1080p low, while the desktop hit 196fps. Lighter settings barely helped, so the GPU isn't what's holding the Spark back here.

Youngblood is the opposite. It's the only game where the RTX Spark held a bigger share of the desktop's performance at 1080p low than at 1440p max. Whatever id Tech is doing, it survives Prism far better than the rest of the lineup.

Frame generation is how Nvidia gets to its number

The RTX Spark only clears 100fps in Cyberpunk 2077 with 4x frame generation

Nvidia promises over 100fps at 1440p in AAA games with ray tracing, DLSS, and Reflex, so I tested that in Cyberpunk 2077. I ran it at 1440p with ray tracing on Ultra and DLSS Quality, first with 2x frame generation on both machines. The RTX Spark averaged 63fps, while the desktop RTX 5070 averaged 127fps, so it's back to about half the desktop's frame rate.

Switching the Spark to 4x frame generation pushed it to 110fps, clearing Nvidia's mark. So the number is real. It just isn't comparing like with like, since the desktop got to 127fps with half as many generated frames.

And 110fps at 4x doesn't feel like 110fps. The Nvidia overlay showed 35 to 40ms of latency, and the low dipped to 99fps. Reflex was locked on with frame generation enabled on both machines, and the Spark doesn't let you turn on Reflex Boost at all. You don't get anything for free, even fake frames.

The RTX Spark games like a laptop RTX 5070, as long as the GPU is the limit

So, does the RTX Spark game like an RTX 5070? Against a laptop RTX 5070, it's close. Against a desktop card, it's about half when the GPU does the work and far less when the CPU does, and Nvidia's 100fps promise needed 4x frame generation in Cyberpunk.

If you mostly play GPU-heavy, single-player games at 1440p with DLSS and frame generation on, the RTX Spark is a surprisingly capable gaming laptop for an Arm chip. If your library leans on CPU-heavy open-world games like Horizon, temper your expectations. And if you play online, check the anti-cheat results before you buy, because some of the biggest multiplayer games won't launch at all. If it's local AI you're after instead, the RTX Spark is a developer's dream, but AMD's Ryzen AI Max+ is what most people actually need.