Geekbench 7: Apple M5 Ultra GPU Performance Nears RTX 4080
Early benchmark results for two unreleased Apple Silicon chips—the M5 Ultra and M6—have appeared in the Geekbench 7 database.
The preliminary results are notable for different reasons.
The M5 Ultra has surpassed 52,000 points in multi-core performance, while its Geekbench 7 OpenCL score has reached a level close to the GeForce RTX 4080.
The M6, meanwhile, has produced a 3,999-point single-core score, currently placing it at the top of the Geekbench 7 single-core leaderboard.
However, these numbers should be treated as early indicators rather than final retail performance. The results available as of September 20, 2026, reportedly come from only 7 to 28 pre-release samples, depending on the chip and benchmark category.
Synthetic benchmark results can also vary significantly with firmware, thermal conditions, power limits, and system configuration.
🚀 M5 Ultra Breaks 52,000 in Multi-Core #
The full-scale M5 Ultra configuration in the leaked benchmark data features a 36-core CPU.
It achieved a Geekbench 7 multi-core score of:
52,353 points
That currently places it at the top of the public Geekbench 7 multi-core results cited in the dataset.
For comparison, the M3 Ultra recorded approximately 39,039 points, meaning the preliminary M5 Ultra result represents roughly a 34% increase.
Its single-core score was:
3,754 points
That is also a very strong result, although the unreleased M6 currently holds the higher single-core score.
M6 targets single-core performance #
The M6 configuration listed in the database features a 12-core CPU.
Its single-core result reached:
3,999 points
That puts the M6 above the M5 Ultra in single-core performance and currently gives it the highest single-core result in the referenced Geekbench 7 leaderboard.
The preliminary numbers also suggest a meaningful generational improvement.
| CPU | CPU Configuration | Single-Core | Multi-Core |
|---|---|---|---|
| Apple M6 | 12 cores | 3,999 | 22,142 |
| Apple M5 Ultra | 36 cores | 3,754 | 52,353 |
| Apple M3 Ultra | — | — | 39,039 |
According to the cited comparisons, the M6 delivers roughly a 10% single-core improvement over its same-tier predecessor.
Its reported single-core lead over AMD’s Ryzen 9 9950X3D2 has also increased from approximately 15% in the previous comparison to around 26%.
Multi-core performance tells a very different story.
With a score of 22,142, the M6 sits below the multi-core results of chips such as the M4 Pro and M3 Max in the referenced dataset.
That is not particularly surprising given the much larger CPU configuration of the M5 Ultra.
🎨 M5 Ultra GPU Results Show a Large Jump #
The M5 Ultra’s GPU configuration is listed with 80 GPU cores.
In Geekbench 7’s native Metal benchmark, it achieved:
351,568 points
For comparison:
- M3 Ultra: 249,564
- M5 Ultra: 351,568
- M5 Max: 238,929
The preliminary M5 Ultra result therefore represents approximately a 41% improvement over the M3 Ultra in this Metal benchmark.
The M6, despite its newer architecture, is positioned at a very different performance level because the tested configuration contains only a 12-core GPU.
Its Metal score reached:
90,736 points
That places it below the M4 Pro and M3 Max results referenced in the same comparison.
OpenCL puts the M5 Ultra close to RTX 4080 #
Direct Metal comparisons with NVIDIA GPUs are impossible because Metal is Apple’s proprietary graphics and compute API.
For a cross-platform comparison, Geekbench’s OpenCL test provides a more useful reference point.
The M5 Ultra achieved:
216,190 points
The cited NVIDIA results are:
| GPU | Geekbench 7 OpenCL Score | Difference vs. M5 Ultra |
|---|---|---|
| Apple M5 Ultra | 216,190 | — |
| GeForce RTX 4080 | 219,100 | ~1% higher |
| GeForce RTX 4090 | 274,656 | ~27% higher |
| GeForce RTX 5090 | 354,548 | ~64% higher |
On this particular benchmark, the M5 Ultra is therefore only about 1% behind the RTX 4080.
The gap becomes considerably larger against higher-end NVIDIA GPUs.
The RTX 4090 leads the M5 Ultra by roughly 27%, while the RTX 5090 is approximately 64% ahead according to these Geekbench 7 OpenCL results.
OpenCL is not the whole story #
There is an important caveat here.
Apple treats OpenCL as a legacy API and prioritizes Metal for modern graphics and compute workloads.
Consequently, an OpenCL score should not be interpreted as a direct measurement of real-world Mac GPU performance.
The situation is especially complicated for gaming.
Game performance depends on the graphics API, driver implementation, game engine, shader compilation, optimization, memory behavior, and many other factors.
A GPU that produces a similar OpenCL score to another GPU does not necessarily deliver similar frame rates in actual games.
For that reason, the M5 Ultra’s OpenCL result is best interpreted as a benchmark comparison, not as proof that it performs like an RTX 4080 across real-world workloads.
🧪 Geekbench 7 Changes How These Numbers Should Be Read #
Another important detail is that these results come from Geekbench 7, not Geekbench 6.
The two versions should not be treated as directly interchangeable.
Primate Labs released Geekbench 7 in July 2026 with changes to both CPU and GPU testing methodology.
CPU workload changes #
Geekbench 7 redesigned its multi-core testing to more closely reflect applications that genuinely benefit from multi-threading.
Instead of simply maximizing thread utilization, the updated methodology attempts to distinguish workloads according to how multi-threading is used in practical software.
That means historical comparisons should be made carefully.
A Geekbench 7 score should not simply be converted into an equivalent Geekbench 6 score using a fixed multiplier.
More modern GPU workloads #
Geekbench 7 also adds GPU workloads aimed at areas such as:
- Machine learning
- Content creation
- Modern GPU compute
- Contemporary graphics workloads
These changes are intended to make the GPU benchmark more representative of the broader workloads modern GPUs are expected to handle.
Expanded API support #
Geekbench 7 also expands its graphics and compute API coverage, including support for:
- Metal
- OpenCL
- Vulkan
- NVIDIA CUDA
This broader API coverage makes cross-platform testing more flexible, although API-specific optimization still means that results need to be interpreted within their respective contexts.
🔍 What the Early Results Actually Tell Us #
Taken together, the preliminary data paints an interesting picture of Apple’s next-generation desktop-class silicon.
The M5 Ultra is the obvious multi-core and GPU-focused part in these results.
Its 36-core CPU reaches more than 52,000 Geekbench 7 multi-core points, while its 80-core GPU produces a 351,568 Metal score.
Its OpenCL score of 216,190 is also close to the cited RTX 4080 result.
The M6 appears to pursue a different balance.
Its 12-core CPU reaches nearly 4,000 points in single-core performance, while its multi-core result is considerably lower than the M5 Ultra because it targets a much smaller CPU configuration.
Its 12-core GPU also produces a much lower Metal score than the M5 Ultra.
This makes direct “M5 Ultra vs. M6” comparisons somewhat misleading.
They are not simply two generations of the same chip occupying identical performance classes. The Ultra designation represents a substantially larger configuration.
⚠️ Retail Hardware Could Tell a Different Story #
The biggest limitation of these results is that the chips have not yet reached retail validation.
The available Geekbench 7 entries represent only a small number of pre-release systems, with the cited dataset containing approximately 7 to 28 samples depending on the chip and test.
Pre-production hardware can differ from final products in several ways.
Firmware may still be changing, power limits may not be finalized, and thermal behavior can vary depending on the test system.
More importantly, synthetic benchmarks cannot fully measure the experience of professional applications.
For the M5 Ultra, real-world testing should eventually answer questions such as:
- How sustained is the 52,000+ multi-core performance under long workloads?
- How does the 80-core GPU perform in professional rendering applications?
- How effective is unified memory for large creative and AI workloads?
- How does Metal performance compare with Windows/Linux GPU ecosystems?
- How well do professional applications take advantage of the additional CPU and GPU resources?
- How does gaming performance compare once actual software optimization is available?
These questions cannot be answered by a single Geekbench score.
📊 A Promising Early Look at Apple’s Next Silicon #
The preliminary Geekbench 7 results nevertheless provide an interesting snapshot of Apple’s next-generation Mac processors.
The M5 Ultra’s 52,353 multi-core score represents a substantial step over the cited M3 Ultra result, while its 351,568 Metal score shows another large generational increase in the native Apple graphics benchmark.
Its 216,190 OpenCL score also places it remarkably close to the cited RTX 4080 result, although the comparison should be treated cautiously because OpenCL is no longer Apple’s primary graphics API.
Meanwhile, the M6’s 3,999 single-core score currently stands out as the most impressive CPU figure in the dataset, suggesting that Apple’s next-generation architecture may continue to prioritize strong per-core performance alongside larger multi-core configurations.
For now, however, these numbers remain preliminary.
The real verdict will come when retail hardware becomes available and independent testing can measure sustained performance, thermals, power consumption, application workloads, and gaming performance.
Until then, Geekbench 7 offers an intriguing early indication of where Apple’s M5 Ultra and M6 are heading—but not yet the complete picture.