Apple announced the M5 Ultra Mac Studio configuration on August 25, 2026, describing an option with as many as 36 CPU cores: 12 “super cores” and 24 performance cores. The company said systems would start arriving on September 22. A public Geekbench Browser entry dated September 15 purports to show a 36-core Apple M5 Ultra system identified as Mac17,15, running macOS 27.0 with 256GB of memory.
The listing reports 3,774 points in Geekbench 7 single-core testing and 52,516 points in multi-core testing. Those are attention-grabbing figures, particularly the multi-core result. But Windows workstation buyers, Mac professionals, and IT decision-makers should read the result with a clearer view of both what it demonstrates and what it cannot demonstrate.
What the early result actually shows
At face value, the uploaded result aligns with Apple’s published maximum CPU specification. Geekbench identifies two CPU clusters totaling 36 cores, arranged as 12 and 24 cores. Apple’s own M5 Ultra description likewise specifies 12 super cores and 24 performance cores in its top CPU configuration.
That correspondence makes the submission plausible as a test of a fully configured M5 Ultra CPU. It does not, however, independently establish that the system is a retail Mac Studio, a final shipping build, or representative of every M5 Ultra system customers will receive. The public record identifies the machine as Mac17,15, but the available Apple material does not map that identifier to a particular retail configuration.
There are further unknowns that matter in benchmarking:
- The upload is a single run rather than a collection of reproducible tests.
- Its uploader, system provenance, thermal conditions, background activity, and storage setup are not independently established.
- The record does not identify the GPU-core configuration, so it says nothing conclusive about whether the GPU was fully enabled.
- Memory capacity is listed as 256GB, but a capacity figure alone does not establish the full memory configuration or its impact on every workload.
The frequency field deserves similar care. The listing reportedly labels 4.61GHz as Base Frequency. That should not be rewritten as a measured peak or boost clock. A base-frequency label and an observed maximum operating frequency are different claims, especially on modern processors that dynamically manage clock speeds.
The Threadripper comparison is real—but narrow
The reason this score has generated attention is its position among high-end Geekbench 7 multi-core uploads. At the time reviewed, a Ryzen Threadripper PRO 9985WX result led the relevant comparison with 56,118 points. The reported M5 Ultra score of 52,516 is 3,602 points lower, a difference of 6.42%.
That is an unusually close single-benchmark comparison considering the CPUs involved. AMD specifies the Threadripper PRO 9985WX as a 64-core, 128-thread Zen 5 workstation processor with a 350W default TDP and boost speeds up to 5.4GHz. The cited 56,118-point result was produced on a Linux AVX2 system using Geekbench 7.0.0.
The comparison looks even more dramatic in the other direction. One Ryzen Threadripper PRO 9995WX submission scored 50,911, placing it 1,605 points, or 3.15%, behind the reported M5 Ultra result. AMD’s 9995WX is a 96-core, 192-thread Zen 5 processor also specified with a 350W default TDP and boost speeds up to 5.4GHz.
That does not mean a 36-core M5 Ultra is generally faster than AMD’s flagship 96-core workstation CPU. It means one M5 Ultra upload produced a higher Geekbench 7 multi-core score than one particular 9995WX upload. Those are materially different statements.
The 9995WX result used Ubuntu 24.04.4 Linux AVX2 and had approximately 1TB of memory. The M5 Ultra result ran macOS on Arm architecture with 256GB of memory. The 9985WX leaderboard-leading run also used Linux AVX2, with 251GB of memory. These are not controlled, configuration-matched comparisons.
Why cross-platform scores need context
Geekbench is designed to compare systems across operating systems and processor architectures. That is valuable: Windows, Linux, and macOS users need some way to place otherwise dissimilar hardware in a common performance framework. Geekbench 7 also redesigned its multi-core benchmark to better reflect application behavior.
Yet cross-platform comparability should not be mistaken for identical test conditions. A score still emerges from a specific combination of hardware, operating system, compiler and instruction paths, firmware, memory setup, cooling behavior, and background system state.
In this case, all of the highlighted submissions use Geekbench 7.0.0, which eliminates one major source of mismatch. But the operating environments are still different:
- The M5 Ultra upload is a macOS AArch64 result with 256GB of memory.
- The high-scoring 9985WX upload is a Linux AVX2 result with 251GB of memory.
- The cited 9995WX upload is a Linux AVX2 result with approximately 1TB of memory.
The distinction matters especially at the workstation level. A CPU benchmark is not a complete test of a workstation platform. It cannot establish leadership in GPU rendering, local AI inference or training, memory bandwidth, PCIe expansion, storage throughput, ECC implementation, virtualization, professional application certification, or server-oriented workloads.
That is highly relevant to Windows professionals. Threadripper PRO systems are often selected not solely for a CPU score, but for a wider platform proposition: very high core counts, large memory configurations, expansion capacity, and the ability to pair the CPU with discrete professional GPUs. The reported Geekbench result does not test those decisions. Nor does it test whether a Mac Studio’s integrated architecture is preferable for a particular creative pipeline.
Apple’s M3-to-M5 claim is a separate question
Apple says M5 Ultra can deliver up to 1.3 times higher multithreaded performance than M3 Ultra. The reported 52,516-point Geekbench result is directionally consistent with the idea of a substantial generation-over-generation gain. But it does not independently validate Apple’s 1.3x claim.
Apple’s published figure is an Apple-controlled performance claim, and the available material does not present it as a Geekbench 7 result. Different benchmarks emphasize different behaviors, so matching a broad marketing multiplier to a single public Geekbench score would be unsound.
The same restraint applies to claims involving server hardware. The early M5 Ultra listing does not establish that M3 Ultra—or M5 Ultra—beats numerous dual-socket EPYC systems. No reviewed leaderboard evidence supports that broader conclusion. Server and dual-socket comparisons require carefully selected workloads, matched software environments, and clear consideration of memory capacity, I/O, scalability, and reliability requirements.
What Windows workstation buyers should take from this
For Windows users considering a high-end desktop, the immediate takeaway is not that a Mac Studio replaces a Threadripper PRO workstation. It is that Apple’s top-tier silicon may be competitive in at least one modern, cross-platform CPU benchmark even when measured against systems with considerably more CPU cores.
That could matter for teams with mixed Windows and macOS fleets. A portable score like Geekbench can help identify broad CPU capability when evaluating machines for cross-platform development, photography, video work, audio production, or general compute-heavy productivity. It may also increase pressure on buyers to look beyond core counts alone when comparing systems.
But a purchasing decision still needs workload-specific testing. A Windows-based CAD, engineering, visualization, scientific, or AI workflow may depend on applications, plug-ins, discrete GPU support, drivers, peripheral compatibility, upgradeability, or expansion hardware that this CPU-only score does not address. Conversely, macOS-native applications may benefit from architectural and software characteristics that a generic cross-platform score cannot fully capture.
Organizations should also resist turning the 350W Threadripper PRO specification into an efficiency comparison. Apple has not supplied a matching power figure in the reviewed material, and Geekbench Browser results do not provide measured whole-system power consumption. Performance per watt, acoustics, sustained clocks, and thermal behavior all require dedicated, repeatable testing.
What to watch next
The most useful next evidence will be repeated M5 Ultra submissions from retail systems after customer availability begins. A range of results would reveal how much normal variation exists and whether the 52,516-point multi-core score is typical, unusually strong, or affected by an early system configuration.
Independent tests should also move beyond CPU scoring. Sustained rendering, video encoding, software compilation, memory-intensive work, GPU compute, AI workloads, storage tests, and application-specific benchmarks will answer questions that Geekbench cannot. For workstation customers, expansion and software compatibility will remain just as important as an aggregate CPU score.
For now, the reported 52,516-point result is best treated as an impressive early data point. It suggests that M5 Ultra could land in serious workstation CPU territory in Geekbench 7 multi-core performance. It does not yet prove broad leadership over Threadripper PRO, validate every Apple performance claim, or establish which platform is best for a real-world Windows or professional workflow.