Apple’s next wave of desktop silicon just got its first independent scoreboard. Geekbench 7 listings reviewed by MacRumors show the M5 Ultra chip posting a Metal graphics score of 366,744, a 59% jump over the M3 Ultra’s average result of 230,420. The M6 chip, Apple’s mainstream successor to the M5, scored 93,217 on the same Metal test, a 34% gain over the M5’s average of 69,563.
The numbers matter because Apple has spent the past several chip generations leaning harder on its own GPU architecture rather than outside graphics IP, betting that Metal-optimized silicon can close the gap with dedicated cards from Nvidia and AMD. MacRumors also noted that the M5 Ultra’s OpenCL score lands roughly in line with Nvidia’s GeForce RTX 4080, a comparison that would have sounded far-fetched for Apple’s integrated graphics only a few years ago.
This isn’t a leak pulled from an anonymous forum post. It’s benchmark data drawn from Geekbench’s public results database and cross-checked by multiple outlets, including MacRumors, AppleInsider, and mobidevices.com. Here’s what the scores actually say, what they leave out, and how they stack up against the rest of the GPU market on September 20, 2026.
What the MacRumors Benchmark Data Actually Shows
The report centers on two separate chips that arrived in Apple’s Mac lineup this year: the M5 Ultra, a flagship part built for the highest-end Mac Studio configurations, and the M6, the volume chip that shows up in machines like the Mac mini. Both were benchmarked using Geekbench 7’s Metal graphics test, which measures how fast a GPU processes rendering and compute workloads through Apple’s native graphics API.
MacRumors compared the M5 Ultra’s listed score against the M3 Ultra’s average score across submitted results, and did the same for M6 against M5. That averaging matters. A single Geekbench listing can be skewed by thermal conditions, background processes, or a specific machine configuration, so comparing one fresh top score against a pool of historical averages tends to favor the newer chip. Even accounting for that, a 59% jump for M5 Ultra and a 34% jump for M6 are large enough margins that they’re unlikely to be noise alone.
Both chips are still working their way through the Geekbench Browser database, which is why the reported scores come from individual submitted results rather than a large, stable sample size. That’s normal for hardware in its first weeks of wide availability, and it’s also why outlets are treating these as early indicators rather than final performance verdicts.
Geekbench 7’s Metal Test, Explained
Geekbench 7 is a cross-platform benchmark suite built by Primate Labs, and its Metal test specifically measures GPU performance through Apple’s graphics and compute framework. It runs a mix of workloads, including image processing, ray tracing, and machine learning inference tasks, then rolls the results into a single composite score. Higher numbers mean faster GPU throughput on that workload mix.
Geekbench also runs an OpenCL test, a cross-vendor compute standard that lets you compare Apple silicon against GPUs from Nvidia, AMD, and Intel on comparable ground. That’s the test MacRumors used to put the M5 Ultra next to the RTX 4080, since Metal scores can’t be directly compared to a Windows machine’s Nvidia driver stack. The tradeoff is that OpenCL doesn’t always reflect how a chip performs in software actually optimized for Metal, so both numbers tell a partial story on their own.
None of this is exotic testing methodology. Geekbench’s database is public, and reviewers, Apple silicon trackers, and hardware forums use it as a common reference point precisely because results are submitted by real machines running real configurations rather than manufacturer-supplied lab numbers.
M5 Ultra vs M3 Ultra: Breaking Down the 59% Jump
The headline comparison is straightforward on paper. M5 Ultra’s Metal score of 366,744 beats the M3 Ultra average of 230,420 by 59%, according to MacRumors’ analysis of the Geekbench listings. That’s a bigger single-generation leap than most CPU or GPU makers manage in a normal product cycle, and it lands at a moment when Apple has publicly staked its Mac Studio pitch on GPU-heavy workloads like 3D rendering, video encoding, and local AI inference.
The table below lays out the primary MacRumors comparison alongside a second, earlier sighting reported by other outlets, so readers can see both data points side by side rather than treating a single score as the final word.
| Chip | Geekbench 7 Metal Score | Compared Against | Baseline Score | Change | Source |
|---|---|---|---|---|---|
| Apple M5 Ultra | 366,744 | M3 Ultra (average) | 230,420 | +59% | MacRumors |
| Apple M6 | 93,217 | M5 (average) | 69,563 | +34% | MacRumors |
| Apple M5 Ultra (early sample) | 360,019 | M3 Ultra, 512GB unified memory unit | 255,009 | +41% | Other benchmark trackers |
| Apple M6 (Mac18,5 build) | 97,137 | M5 (average) | 69,563 | +40% | Other benchmark trackers |
| Apple M5 Ultra OpenCL | Not disclosed | Nvidia GeForce RTX 4080 | Roughly in line | Qualitative | MacRumors |
A 59% gain is a real generational leap, not a rounding error. It also lines up with the general direction Apple has pushed since the original M1 Ultra: more GPU cores, more shared memory bandwidth, and tighter integration between the CPU, GPU, and Neural Engine on a single package. Reviewers have not yet published full third-party benchmark suites for the new Ultra chip, so treat this specific score as an early signal that the architecture is delivering meaningfully more graphics throughput, not as a substitute for hands-on testing across creative and gaming workloads.
M6 vs M5: A Smaller Gain That Still Matters
The M6’s 34% Metal score gain over the M5 average is less dramatic than the Ultra chip’s jump, but it’s arguably more important for Apple’s business. The M5 Ultra ships in a small number of high-end Mac Studio configurations bought mostly by professionals doing video, 3D, or AI work. The M6, by contrast, is the chip most Mac buyers will actually own, since it’s built for volume products like the Mac mini.
A separate Geekbench listing for a machine identified as Mac18,5, which other trackers tie to the M6 generation, recorded a Metal score of 97,137. That’s roughly 40% above the M5’s 69,563 average, a bit higher than the 34% figure MacRumors cited from its own comparison set. The gap between those two numbers is a useful reminder that early benchmark data from different submissions won’t always agree exactly, even when they’re measuring the same silicon.
For everyday users, a graphics score improvement in the 34% to 40% range translates into faster export times in Final Cut Pro, smoother performance in Metal-accelerated games, and more headroom for on-device AI features that Apple has been building into macOS. It’s a meaningful upgrade without being a generational reinvention, which tracks with how Apple has historically positioned its non-Ultra chips.
Where the Two Data Sets Diverge
It’s worth sitting with why MacRumors’ numbers (366,744 and 93,217) don’t perfectly match the secondary figures (360,019 and 97,137) that other trackers picked up. Geekbench listings are submitted by individual machines, not curated lab benches, so cooling, background load, and even the specific memory configuration of a Mac Studio or Mac mini unit can shift a score by a few percentage points in either direction. Neither number is wrong. They’re both snapshots of the same underlying chips captured at slightly different moments from slightly different hardware samples.
| Chip | MacRumors Reported Score | Secondary Tracker Score | Absolute Difference | Variance |
|---|---|---|---|---|
| Apple M5 Ultra | 366,744 | 360,019 | 6,725 | 1.8% |
| Apple M6 | 93,217 | 97,137 | 3,920 | 4.2% |
A variance under 5% between two independently sourced benchmark sightings is tight by Geekbench’s usual standards, which is part of why outlets treated both sets of scores as credible enough to report rather than dismissing either one as an outlier.
Why There Was No “M4 Ultra”
Sharp-eyed readers will notice the M5 Ultra comparison skips a generation, measuring against M3 Ultra rather than an M4 Ultra. Apple never shipped an Ultra variant of the M4 chip, a gap that tracks with reporting on the Mac Studio’s release cadence throughout the M-series era. The Ultra chip is built by fusing two Max-series dies together over a high-bandwidth interconnect, a manufacturing approach that doesn’t map cleanly onto every chip generation Apple ships in a given year.
That’s also why the M5 Ultra comparison in this report jumps straight to M3 Ultra as its baseline. It’s the most recent Ultra-class chip that actually exists, which makes it the only fair apples-to-apples comparison point MacRumors had available.
How M5 Ultra Stacks Up Against Discrete GPUs
The most eye-catching line in the MacRumors report isn’t a raw score, it’s the comparison to Nvidia. The outlet reports that the M5 Ultra’s OpenCL score comes in roughly in line with the GeForce RTX 4080, a card that’s been a mainstay of high-end gaming and creative rigs since its 2022 launch. MacRumors didn’t publish an exact point-by-point score for that comparison, so treat it as a qualitative read rather than a precise gap measurement.
Context matters here. The RTX 4080 is no longer Nvidia’s current flagship. Nvidia’s newer 50-series cards, including the RTX 5090, sit well above it in raw compute throughput, and pricing on those top-end cards has climbed sharply this year as AI-driven demand for GPU silicon has squeezed consumer supply. So an Apple chip landing “roughly in line” with a three-year-old high-end GPU is a genuinely strong result for an integrated, power-efficient design, even though it’s not a claim that Apple silicon has caught the very top of the discrete GPU market.
It’s also worth remembering that OpenCL and Metal measure different things, and that neither test captures gaming performance directly. A chip can post a strong compute score while still trailing in game-specific rendering pipelines, driver optimization, or ray tracing implementations, all of which matter more to someone buying a machine primarily to play games rather than render video or train models locally.
Metal vs OpenCL: Two Different Report Cards
Apple’s own software stack is built around Metal, so Metal scores are the more representative number for anyone who lives inside macOS apps like Final Cut Pro, Logic Pro, or Xcode’s Metal-based simulators. OpenCL exists mainly as a cross-platform yardstick, useful for comparing an Apple chip against a Windows machine’s Nvidia or AMD card, but less reflective of how that chip performs inside Apple’s native ecosystem.
That distinction explains why MacRumors leaned on Metal scores for the headline M5 Ultra and M6 comparisons, then switched to OpenCL specifically to make the Nvidia comparison legible. Readers evaluating whether to upgrade should weight the Metal numbers more heavily if their workload lives inside Apple’s software, and treat the OpenCL comparison as a rough sanity check against the wider GPU market rather than a definitive verdict.
What This Means for Mac Studio and Mac mini Buyers
For anyone weighing a Mac Studio purchase, the practical takeaway is that the M5 Ultra represents the largest single-generation graphics jump Apple’s Ultra lineup has posted in Geekbench testing to date, based on the scores currently available. That’s relevant for video editors, 3D artists, and developers running local machine learning workloads who lean on GPU compute daily.
Mac mini buyers looking at the M6 get a smaller but still tangible upgrade. A 34% to 40% Metal score gain won’t transform a machine used mostly for browsing and office work, but it will show up in anything that leans on the GPU, including Metal-accelerated photo editing, casual gaming, and the on-device AI features Apple has steadily expanded across recent macOS releases. Anyone still running an Intel-based Mac should also note that macOS’s newest release has already dropped support for Intel hardware, which makes the Apple silicon upgrade path less optional than it used to be for users who want continued OS updates.
Buyers comparing against Apple’s mobile silicon roadmap should also know the company’s A20 Pro chip, which powers recent iPhone models, has posted its own Geekbench gains this year, with a reported single-core score around 4,042 that puts it within reach of some Mac-class chips on CPU work, even though GPU compute still separates phone and desktop silicon by a wide margin.
The Business Case: Why Graphics Performance Matters to Apple Right Now
Apple’s Mac Studio and Mac mini lines don’t sell in anywhere near the volume of the iPhone, but they carry outsized importance for Apple’s pitch to professional and enterprise buyers. Creative studios, engineering teams, and AI developers evaluating local hardware for training or inference workloads care directly about GPU throughput, memory bandwidth, and how well a chip’s compute performance holds up against dedicated graphics cards.
A credible claim that the M5 Ultra’s compute performance lands near a card like the RTX 4080 gives Apple a talking point it hasn’t had in this category before: a workstation-class Mac that doesn’t require a separate GPU to compete on raw graphics compute. That matters more this year than in prior cycles, since GPU supply constraints and rising prices across the Nvidia and AMD lineups have made every alternative compute platform more attractive to buyers trying to control hardware budgets.
Historical Context: Apple Silicon’s GPU Trajectory
Apple’s shift to in-house silicon reshaped the Mac’s performance story starting with the original M1 chip, and the Ultra tier has been the clearest showcase of that strategy’s ceiling. Each Ultra chip is built by joining two Max dies over Apple’s UltraFusion interconnect, a design choice that lets Apple scale GPU core counts and memory bandwidth well beyond what a single die could support.
From Early Ultra Chips to M5 Ultra
Every Ultra-generation chip Apple has shipped has posted a graphics score improvement over its predecessor, and the M5 Ultra’s reported 59% jump over M3 Ultra continues that pattern at a scale that stands out even against Apple’s own track record. What makes this cycle notable isn’t just the percentage gain, it’s that Apple is now willing to invite direct comparisons against Nvidia’s GPU lineup rather than only benchmarking against its own prior chips, a sign of how much confidence the company has in its current GPU architecture.
Competitive Landscape: Apple Silicon vs the Discrete GPU Market
Nvidia and AMD still dominate the discrete GPU market, and nothing in this benchmark data changes that. What it does change is where the floor sits for integrated, power-efficient silicon. A chip that draws a fraction of the power of a desktop RTX card while landing “roughly in line” with it on a cross-platform compute test is a meaningfully different proposition than the integrated graphics conversation looked like even three years ago.
That said, gamers and GPU-bound professionals shouldn’t read this as Apple silicon replacing high-end discrete cards outright. Independent GPU benchmark trackers consistently show that raw compute parity on synthetic tests doesn’t always translate into equivalent real-world game performance, driver maturity, or software ecosystem support. Windows gaming, in particular, remains overwhelmingly built around Nvidia and AMD hardware, and that isn’t changing because of one Geekbench comparison.
How to Check These Scores Yourself
Anyone who wants to verify these numbers directly, or track new submissions as they land, can pull results straight from Geekbench’s public browser database. Geekbench also ships a command-line tool for users who want to benchmark their own machine and compare it against the listings referenced in this report.
# Example: running Geekbench 7's Metal benchmark from the command line
# (illustrative syntax; consult Geekbench's own documentation for exact flags)
./geekbench7 --compute metal
# Upload and compare your result against the public database
./geekbench7 --uploadRunning the benchmark locally won’t reproduce MacRumors’ exact averaging methodology, but it will show where your specific machine lands relative to the scores discussed in this report, which is the same raw data these comparisons are built from.
What to Watch Next
- More Geekbench submissions will firm up the averages. Early scores from a handful of machines tend to move as the sample size grows, so expect the M5 Ultra and M6 averages to shift somewhat over the coming weeks as more listings appear.
- Third-party review outlets will likely publish full benchmark suites. Geekbench is one data point among many, and independent reviewers typically follow up leaked or early scores with broader creative, gaming, and compute benchmark rounds.
- Expect continued comparisons against Nvidia’s current lineup. Now that Apple has invited an RTX 4080-class comparison, expect outlets to push further and test M5 Ultra against Nvidia’s newer 50-series cards directly.
- Pricing and GPU supply pressure will keep shaping the narrative. With discrete GPU prices climbing across the market this year, an Apple chip that competes on compute without a separate graphics card is likely to keep drawing attention from budget-conscious professional buyers.
- Apple’s AI feature roadmap will lean harder on these GPU gains. Faster on-device compute gives Apple more room to push local AI processing in macOS without relying on cloud inference, a direction the company has been steadily building toward.
These are analyst-style projections based on the current data and Apple’s recent product direction, not confirmed Apple roadmap details.
The Bottom Line
The scores MacRumors reported are early, drawn from a limited number of Geekbench submissions, and not a substitute for full third-party review benchmarks. But the direction is clear enough. M5 Ultra’s 59% Metal score gain over M3 Ultra, and M6’s 34% gain over M5, mark a real step up in Apple’s GPU performance, at a moment when the company has enough confidence in the result to invite a direct comparison against Nvidia’s GPU lineup. For buyers evaluating a new Mac Studio or Mac mini this year, that’s a data point worth paying attention to, even while the fuller performance picture is still being filled in.
Frequently Asked Questions
What is the M5 Ultra’s Geekbench 7 Metal score?
MacRumors reported a Metal score of 366,744 for the M5 Ultra, a 59% increase over the M3 Ultra’s average score of 230,420. A separate early sample reported by other trackers scored 360,019 against a specific M3 Ultra unit with 512GB of unified memory that scored 255,009.
How much faster is the M6 than the M5 in benchmarks?
The M6 scored 93,217 on Geekbench 7’s Metal test, a 34% gain over the M5’s average score of 69,563, according to MacRumors. A separate listing for a Mac18,5 device, tied to the M6 generation by other trackers, scored 97,137, roughly a 40% gain over the same M5 baseline.
Does the M5 Ultra really compete with an Nvidia RTX 4080?
MacRumors reported that the M5 Ultra’s OpenCL score lands roughly in line with Nvidia’s GeForce RTX 4080, without publishing an exact side-by-side figure. It’s a compute benchmark comparison rather than a gaming performance claim, so it shouldn’t be read as proof the two are interchangeable for every workload.
Why did Apple skip an M4 Ultra chip?
Apple never released an Ultra-tier version of the M4 chip, which is why the current comparisons measure M5 Ultra against M3 Ultra, the most recent Ultra-class chip Apple had previously shipped.
What benchmark tool was used for these scores?
All of the scores referenced in this report come from Geekbench 7, specifically its Metal graphics test for the Apple-to-Apple comparisons and its OpenCL test for the cross-vendor comparison against Nvidia hardware.
Are these Geekbench scores final and reliable?
They’re early results pulled from a limited number of public Geekbench submissions, not lab-controlled benchmark runs. Scores can shift as more machines are tested and added to the database, so treat the current numbers as an early indicator rather than a final verdict.
Which Mac models use the M5 Ultra and M6 chips?
Apple has used the M5 Ultra in its highest-end Mac Studio configurations, while the M6 has appeared in machines including the Mac mini, according to Apple’s own product rollout and prior reporting.
Should I upgrade my Mac based on these benchmark results alone?
A Geekbench score is a useful signal but not a complete picture. If your work leans heavily on GPU compute, the reported gains are meaningful, but it’s worth waiting for full third-party review benchmarks across your specific software before making a purchase decision based on one leaked test.
