Apple’s M6 chip just posted a Geekbench 7 Metal score that puts its integrated graphics within shouting distance of a Nvidia GeForce RTX 4080, according to benchmark data reported by MacRumors on September 20, 2026. The M6 hit a Metal score of 93,217, a 34% jump over the M5’s average of 69,563, while a separate M6-based Mac (identified in Geekbench as Mac18,5) posted 97,137 in its own Metal run. Paired with a companion story about the new M5 Ultra, which scored 366,744 in Metal testing (a 59% gain over the M3 Ultra’s 230,420 average), the numbers mark one of the larger generational graphics jumps Apple Silicon has produced since the M1 launched in 2020. The full story sits inside a broader run of hardware news this week, but this particular jump stands out because of who it’s being compared against.
The headline isn’t just the raw score. MacRumors also reports that the M5 Ultra’s OpenCL result in Geekbench lands roughly in line with Nvidia’s GeForce RTX 4080, a desktop discrete GPU that shipped years ago but still anchors a lot of buying decisions for creators and gamers. That comparison, even a rough one, is a bigger deal than the percentage gains on their own. Apple has spent five generations trying to convince professional users that a laptop or a compact desktop can replace a graphics card. This is the first time the company’s own hardware has landed close enough to a mainstream discrete GPU number that the comparison holds up in a benchmark chart instead of a marketing slide.
The Geekbench 7 Metal Numbers Behind the Story
Geekbench 7’s Metal test measures GPU-bound compute and rendering tasks under Apple’s native graphics API, so the scores aren’t directly transferable to games or to Windows-based benchmarks. That caveat matters, but it doesn’t erase the trend. MacRumors’ reporting lists four separate data points worth tracking: the M6’s 93,217 Metal score against the M5 average of 69,563, a second M6 listing (Mac18,5) at 97,137, the M5 Ultra’s 366,744 against the M3 Ultra’s 230,420 average, and an additional early M5 Ultra sample at 360,019 compared with a 512GB Mac Studio running M3 Ultra at 255,009. Every one of those pairings points the same direction: Apple’s newest silicon is pulling further ahead of its own prior generation than it has in past cycles.
What’s notable is how consistent the jump is across chip tiers. A 34% gain on the standard M6 and a 59% gain on the flagship M5 Ultra aren’t identical numbers, but they tell a similar story: Apple’s GPU architecture, not just core count, improved meaningfully this generation. Geekbench listings can shift as more samples get submitted, and Apple has not published its own official benchmark claims for either chip as of this writing, so treat the specific figures as early but directionally solid.
M6 vs M5: A 34% Jump in One Generation
The M6’s 34% Metal improvement over the M5 average is the more modest of the two headline gains, but it’s arguably the more important one for everyday buyers. The M5 Ultra ships in high-end Mac Studio configurations that most people never touch. The base M6 lands in the machines Apple actually sells in volume: MacBook Pro, Mac mini, and eventually the entry Mac Studio. A generational GPU gain in that tier moves the needle for video editors, indie game developers, and anyone running local AI inference through Apple’s MLX framework.
For context on how fast Apple’s mobile-to-desktop silicon strategy has been iterating lately, the A20 Pro’s 7-core GPU and 50% memory bandwidth increase earlier this year followed the same playbook: smaller architectural tweaks compounding into a double-digit percentage gain rather than one dramatic redesign. The M6’s 34% Metal jump fits that same incremental-but-real pattern, and it lines up with the A20 Pro’s own Geekbench 7 leak hitting 4,042, which pushed Apple’s mobile chip closer to M5 Max territory months before the M6 desktop numbers surfaced.
M5 Ultra vs M3 Ultra: The Bigger Story at 59%
The M5 Ultra’s 59% Metal gain over the M3 Ultra average is the larger number in this data set, and it comes from a chip aimed squarely at professional workloads: 3D rendering, color grading, and machine learning training on Apple hardware instead of a rack of Nvidia cards. Apple builds Ultra-tier chips by fusing two of its larger dies together over a high-bandwidth interconnect, and a 59% Metal score gain suggests that approach is still scaling well even as die complexity grows.
| Chip | Geekbench 7 Metal Score | Change vs Prior Chip | Reported By |
|---|---|---|---|
| Apple M6 (Mac18,5 sample) | 97,137 | — | MacRumors |
| Apple M6 (reported average) | 93,217 | +34% vs M5 average | MacRumors |
| Apple M5 (average) | 69,563 | Baseline | MacRumors |
| Apple M5 Ultra | 366,744 | +59% vs M3 Ultra average | MacRumors |
| Apple M5 Ultra (early sample) | 360,019 | ~+41% vs M3 Ultra 512GB sample | Independent Geekbench listing |
| Apple M3 Ultra (average) | 230,420 | Baseline | MacRumors |
| Apple M3 Ultra (512GB Mac Studio sample) | 255,009 | Baseline | Independent Geekbench listing |
Two different M5 Ultra samples and two different M3 Ultra baselines produce two slightly different percentage gains (59% and roughly 41%), which is normal for crowd-submitted benchmark databases where sample size and system configuration vary. Both point to a real, large jump rather than a rounding artifact.
How M6’s OpenCL Score Stacks Up Against RTX 4080
MacRumors’ reporting frames the M5 Ultra’s OpenCL result as roughly comparable to Nvidia’s GeForce RTX 4080, without publishing an exact side-by-side OpenCL figure for the RTX 4080 in that comparison. That’s a meaningfully different claim than saying Apple’s chip beats or matches the RTX 4080 across every workload. OpenCL is one compute API among several, Nvidia’s card also runs CUDA-accelerated tasks that Apple silicon can’t touch natively, and gaming performance depends on driver-level optimization that Apple’s Metal ecosystem still lags on for AAA titles.
Still, the comparison is a useful marker. The RTX 4080 remains a card enthusiasts recognize, and pricing on Nvidia’s newer RTX 5090 has roughly doubled toward $5,000 amid AI-driven demand, pushing more buyers to weigh older or alternative options. A Mac chip that trades blows with a 4080-class GPU in at least one compute benchmark gives Apple a talking point it hasn’t had in years: that its integrated graphics can substitute for a discrete card in specific professional workflows, not just casual ones.
Competitive Landscape: Apple Silicon vs Nvidia, AMD and Intel Graphics
Apple doesn’t sell a standalone GPU, so it isn’t competing with Nvidia or AMD in the traditional sense. What it competes for is the buyer’s decision to build a workstation around a Mac instead of a Windows tower with a discrete card slotted in. That distinction shapes how to read these numbers.
| Platform | Type | 2026 Competitive Position | Maker |
|---|---|---|---|
| Apple M6 | Integrated SoC GPU | Metal score up 34% gen-over-gen; targets mainstream Mac buyers | Apple |
| Apple M5 Ultra | Integrated SoC GPU | OpenCL score reported near RTX 4080 class | Apple |
| Nvidia GeForce RTX 4080 | Discrete desktop GPU | Reference point for Apple’s own comparison | Nvidia |
| Nvidia GeForce RTX 5090 | Discrete desktop GPU | Street price near $5,000 amid AI demand | Nvidia |
| AMD Radeon (current desktop lineup) | Discrete desktop GPU | Competes with Nvidia, not directly benchmarked here | AMD |
Nvidia and AMD’s Radeon lineups still dominate anywhere gaming performance or CUDA-specific machine learning tooling matters. What Apple’s M6 and M5 Ultra numbers chip away at is the narrow but growing segment of professional users who don’t need CUDA and don’t play the latest AAA titles: video editors on Final Cut Pro, 3D artists using Metal-optimized renderers, and developers running smaller AI models locally through MLX instead of a cloud GPU instance.
Why This Matters Beyond Gaming Benchmarks
Geekbench’s Metal test doesn’t measure frame rates in a shooter, and nobody buys a Mac Studio to play the newest competitive multiplayer release. The audience that cares about these numbers is smaller and more specific: studios doing color work, engineers running simulations, and machine learning practitioners who want unified memory instead of shuffling data between system RAM and a GPU’s dedicated VRAM.
Unified memory is the actual differentiator here, and it’s easy to lose in a scores comparison. A Mac Studio with M5 Ultra can address a shared memory pool far larger than what fits on a single consumer GPU card, which matters more for loading big AI models than raw compute throughput does. That’s part of why Apple keeps pushing memory capacity as hard as GPU core count generation after generation.
Metal vs CUDA: The Software Layer Apple Still Has to Win
Hardware numbers only matter if software uses them well, and this is where Apple’s story gets more complicated. CUDA has a decade-plus head start as the default framework for machine learning research, and most published models, training scripts, and inference libraries assume an Nvidia card is on the other end. Apple’s answer is MLX and Metal Performance Shaders, both of which have improved, but neither has the ecosystem depth CUDA carries.
There’s also a workaround culture growing around this gap. Developers have already gotten Nvidia’s own upscaling tech, DLSS 5 running on Apple Silicon through a Metal workaround, even at low frame rates, just to prove the concept is possible. That kind of cross-platform tinkering is a signal that raw GPU compute is no longer the ceiling on what Apple hardware could theoretically do. Software compatibility is the remaining wall.
Historical Context: Apple’s GPU Climb Since the M1 Era
Apple’s Apple Silicon transition started in 2020 with the M1, a chip built primarily to prove that ARM-based Macs could match Intel on performance per watt. GPU performance was a secondary pitch in that first generation. By the time Apple introduced its first Ultra-tier chip, the pitch had shifted toward creative professionals who needed workstation-class graphics without a discrete card. Every generation since has pushed that positioning further, adding more GPU cores, more memory bandwidth, and now, with the M6 and M5 Ultra, benchmark scores that invite direct comparison to mid-range and upper-mid-range discrete GPUs instead of just other laptops.
What makes this cycle different is the pairing of two generational leaps landing in the same news cycle. A 34% gain on the mainstream chip and a 59% gain on the flagship Ultra chip, reported together, reads less like routine iteration and more like Apple deciding graphics performance needed a bigger push this generation than in the recent past.
Market Impact for Mac Studio and Mac Mini Buyers
For anyone shopping for a new Mac Studio or Mac mini in the coming months, the practical takeaway is straightforward: wait for M6-generation hardware if the GPU is the deciding factor. A 34% Metal score improvement is large enough to show up in real rendering and export times, not just in a benchmark app. For buyers already running M3 or M4-generation Macs, the upgrade case gets stronger the more GPU-bound the workload is, and weaker for anyone doing mostly CPU-bound office or development work.
Anyone weighing a fresh macOS install alongside new hardware should also factor in macOS 27’s arrival and the end of Intel Mac support, since a GPU upgrade and an OS upgrade are now arriving on roughly the same timeline. Pricing is the open question Apple hasn’t addressed publicly. Historically, Ultra-tier Mac Studio configurations carry a steep premium over the standard tier, and nothing in the current reporting suggests that changes with the M5 Ultra. Buyers deciding between a maxed-out Mac Studio and a Windows workstation with a Nvidia card will still need to weigh that price gap against the software ecosystem they’re already invested in, whether that’s Final Cut Pro and Logic Pro or a CUDA-based pipeline that only runs well on Nvidia hardware.
What Creative and AI Workloads Stand to Gain
Three groups benefit most directly from this generation’s GPU gains. Video editors working in 4K and 8K timelines get faster export and effects rendering proportional to the Metal score increase. 3D artists using Metal-native renderers see similar gains in viewport performance and final-frame render times. And developers running local large language models through MLX get more headroom before needing to fall back on a cloud GPU instance, particularly on Ultra-tier machines where the unified memory pool can hold larger models entirely on-device.
None of that changes gaming, where macOS still has a thinner library than Windows and where most AAA studios optimize for Nvidia and AMD hardware first. Apple’s GPU story in 2026 remains a professional-workload story, not a gaming one, and the M6 and M5 Ultra numbers reinforce that rather than change it.
Risks and Caveats: Synthetic Scores vs Real-World Performance
Geekbench scores are useful directional signals, not guarantees of real-world performance. Crowd-submitted results can include misconfigured systems, thermal throttling differences between laptop and desktop chassis, and small sample sizes that skew an average. The two different M5 Ultra figures cited in current reporting (366,744 and 360,019) against two different M3 Ultra baselines (230,420 and 255,009) are a good example: both point the same direction, but the exact percentage gain depends on which sample set gets used.
Apple has also not published its own official benchmark claims for either chip as of this writing, and neither MacRumors nor other outlets have disclosed exact system configurations, cooling setups, or software versions used for every listed score. Treat the percentages as a strong early signal of a real architectural improvement, not as a certified spec sheet.
What Comes Next: 5 Predictions for Apple’s GPU Roadmap
- Apple will lean harder on MLX and Metal Performance Shaders marketing once M6 Pro and M6 Max variants ship, using the M6’s Metal gains as the entry point for a bigger local-AI pitch.
- Expect an M6 Ultra to widen the OpenCL gap with Nvidia’s RTX 4080-class cards further, since Ultra-tier chips have delivered the larger percentage gains in each of the last several cycles.
- Nvidia and AMD are unlikely to change desktop GPU pricing or roadmaps in response, since Apple still isn’t selling a competing discrete card and the overlap in buyers remains limited to a narrow professional segment.
- Mac Studio and Mac mini refresh cycles built around M6-generation chips will lean on video and 3D creative benchmarks in marketing more than gaming comparisons, continuing Apple’s existing positioning.
- Third-party developers will keep building Metal workarounds for Nvidia-first tools like DLSS, signaling that the software gap, not the hardware gap, is now the bigger obstacle to Apple Silicon competing more broadly in GPU-bound work.
Apple has not confirmed pricing, release timing for additional M6 variants, or an official benchmark comparison against any Nvidia or AMD product, so all five points above are analysis based on the pattern of the last several chip generations rather than confirmed plans. Readers should treat them as informed forecasting, not Apple announcements.
Frequently Asked Questions
What is the M6’s Geekbench 7 Metal score?
MacRumors reports an M6 Metal score of 93,217, a 34% increase over the M5’s average score of 69,563. A separate M6-based Mac listing (Mac18,5) posted 97,137 in its own Metal test.
How does the M5 Ultra compare to the M3 Ultra in Geekbench?
The M5 Ultra scored 366,744 in Geekbench 7 Metal testing, a 59% gain over the M3 Ultra’s average score of 230,420, according to MacRumors. A separate early sample put the M5 Ultra at 360,019 against a 512GB Mac Studio running M3 Ultra at 255,009, roughly a 41% gain in that specific pairing.
Does Apple’s M6 actually beat the Nvidia RTX 4080?
No direct M6-to-RTX 4080 comparison has been published. MacRumors reports that the M5 Ultra’s OpenCL score in Geekbench lands roughly in line with the RTX 4080, without an exact matching figure for the Nvidia card in that comparison. Treat it as a rough parity claim in one compute benchmark, not a broad performance win.
Which benchmark suite produced these scores?
All of the scores discussed come from Geekbench 7’s Metal graphics test, which measures GPU compute and rendering performance under Apple’s native Metal API rather than under DirectX, Vulkan, or CUDA.
Should I buy an M5 or M3 Ultra Mac now, or wait for M6-generation hardware?
If GPU-bound work like video export, 3D rendering, or local AI inference is the priority, the reported 34% and 59% Metal score gains suggest waiting for M6-generation machines is worth it. For CPU-bound or general productivity use, the gap matters far less.
Does this benchmark data apply to gaming performance on Mac?
Not directly. Geekbench 7 Metal measures compute and rendering workloads rather than game frame rates, and macOS still has a smaller AAA game library than Windows. These scores say more about creative and AI workloads than about gaming.
Has Apple officially confirmed these benchmark numbers?
No. The figures come from Geekbench’s crowd-submitted database as reported by MacRumors and other outlets. Apple has not published its own official Metal benchmark comparisons for the M6 or M5 Ultra as of this writing.
What is unified memory, and why does it matter for these chips?
Unified memory lets the CPU and GPU on Apple Silicon share a single memory pool instead of relying on separate system RAM and dedicated GPU VRAM. On Ultra-tier chips like the M5 Ultra, that pool can be large enough to hold bigger AI models entirely on-device, which matters more for some workloads than raw GPU compute throughput does.
