NVIDIA's upcoming RTX Spark platform has appeared in new Geekbench 7 results, giving an early look at the CPU performance of its ARM based SoC ahead of launch.
The stronger 20 core version reached a multi core score of 23,126, putting it about 11% behind Apple's 14 core M4 Max result of 25,957. Single core performance remains a larger weakness, with the RTX Spark scoring 2,570 compared with 3,404 for the M4 Max.
An 18 core RTX Spark configuration also appeared in the benchmark database. It scored 2,541 in single core and 21,776 in multi core performance.
These are leaked pre release results, so final retail systems could perform differently once hardware, firmware and drivers are finished.
RTX Spark Geekbench 7 results
| Processor | CPU cores | Single core | Multi core |
|---|---|---|---|
| RTX Spark | 20 | 2,570 | 23,126 |
| RTX Spark | 18 | 2,541 | 21,776 |
| Apple M4 Max | 14 | 3,404 | 25,957 |
Both RTX Spark test systems reportedly included 64 GB of unified memory and were running under a Balanced power profile rather than a maximum performance setting.
The 20 core model was only about 1.1% faster than the 18 core version in single core performance, but its larger CPU configuration produced a more useful 6.2% advantage in the multi core test.
Compared with Apple's M4 Max, the difference is much wider when only one CPU core is being tested. The 20 core RTX Spark result is approximately 24.5% lower.
That suggests NVIDIA's first major Windows ARM PC platform may be more competitive in heavily threaded workloads than in tasks that depend strongly on individual core performance.
Blackwell graphics could be the more important part
CPU benchmarks tell only part of the RTX Spark story.
NVIDIA developed the platform in collaboration with MediaTek, but the integrated Blackwell graphics hardware could be its main advantage over other ARM based Windows systems.
Current driver information points to different GPU configurations across the two versions. The 18 core model is expected to include 5,120 CUDA cores, while the 20 core version may increase that figure to 6,144 CUDA cores.
That 6,144 core count happens to match some existing desktop GeForce products, but it would be misleading to assume identical performance. Power limits, memory bandwidth, clock speeds and architecture configuration can produce very different results even when CUDA core counts are similar.
Real gaming and GPU compute benchmarks will therefore matter much more than specifications alone.
Early benchmarks already show improvement
Earlier engineering samples associated with NVIDIA's ARM PC effort produced weaker benchmark results, including performance that struggled against older Apple silicon.
The latest numbers suggest the platform has improved as development moves closer to release.
That could come from higher clock speeds, newer firmware, improved Windows ARM support or broader optimization. Without final retail hardware, it is difficult to determine how much performance is still available.
The Balanced power mode used during these tests is another important detail. A higher performance profile could potentially raise scores, although it would likely increase power consumption as well.
RTX Spark is expected to enter the PC market with an unusual combination of a high core count ARM CPU, unified memory and Blackwell graphics.
Its leaked CPU results suggest NVIDIA may not immediately challenge Apple's strongest chips in single core performance, but the 20 core model is already much closer in multi core workloads. The bigger question is whether its GPU performance, efficiency and Windows software compatibility can make the complete platform competitive when finished systems arrive.
Discussion (0)
Be the first to comment.