M3 Ultra · 512GB · 4TB SSD · Dedicated remote rental
Waiting for M5 Ultra? Try M3 Ultra 512 GB.
Evaluate your models and workflow before buying, or while waiting for your M5. Dedicated remote access; nothing is shipped.
3 days · $299 · one-time payment
This is remote access to an M3 Ultra hosted in Kobe, Japan. Nothing is shipped to you. Connect over Tailscale using SSH or VNC; access depends on machine availability and connection readiness.
You receive a clean macOS environment and configure your own models and software. Environment-setup support is not included.
Large-model examples measured on this M3 Ultra
The existing benchmark report contains these July 2026 measurements on the rental machine. They help you choose an evaluation workload; they do not describe M5 Ultra performance.
| Model / runtime | Peak memory | Generation |
|---|---|---|
| DeepSeek R1-05284bit MLX | 380.7 GB | 20.3 tok/s |
| Kimi K2.7 CodeUD-Q2_K_XL GGUF / llama.cpp | 339.8 GB | 24.6 tok/s |
| Qwen3.5 397B-A17B4bit MLX / mlx-vlm | 229.7 GB | 38.1 tok/s |
Single-run snapshots from July 19–22, 2026; 300 generated tokens with model-specific prompts of 2,048–2,872 tokens. Runtimes and methods differ by row. Longer contexts, different artifacts or software can change memory use and speed. See the full report for conditions and limitations.
See measured M3 Ultra benchmarksM5 Ultra 512 GB shipping: a dated Apple snapshot
Apple’s selected M5 Ultra 512 GB / 1 TB configuration showed 13–15 weeks to dispatch when checked on October 31, 2026. This is a dated observation, not today’s promised lead time.
This is an estimate of time until dispatch, not an arrival date at your address. No destination postcode was entered. Your order’s estimate may differ.
View the five Apple stores, configurations and dated observations
Selected configuration: M5 Ultra, 36-core CPU, 80-core GPU, 512 GB unified memory, 1 TB SSD, no added pro apps or trade-in. The United States and Japan were also checked with a 4 TB SSD and displayed the same 13–15-week dispatch estimate.
| Apple store | Observed time to dispatch | Selected configuration / source |
|---|---|---|
| United States | 13–15 weeks | Apple · 512GB / 1TB Apple · 512GB / 4TB |
| Japan | 13–15 weeks | Apple · 512GB / 1TB Apple · 512GB / 4TB |
| United Kingdom | 13–15 weeks | Apple · 512GB / 1TB |
| Germany | 13–15 weeks | Apple · 512GB / 1TB |
| Australia | 13–15 weeks | Apple · 512GB / 1TB |
Use the evaluation before buying or while waiting
Before placing an order
Check whether your chosen model, quantization and context length fit on this 512 GB M3 Ultra. Compare outputs on your own tasks and identify the software you will need.
After placing an order
Start model evaluation, prepare scripts and validate your macOS workflow during the wait. Keep the results and environment versions so you can repeat the tests on your M5 Ultra.
What this M3 Ultra evaluation can tell you
- Whether your chosen model and context length load and run on this M3 Ultra configuration, and how much memory the run uses.
- How different model artifacts and quantizations answer your representative tasks.
- Whether your MLX, llama.cpp or other macOS-compatible tools work together in your intended workflow.
These are observations on M3 Ultra. M5 Ultra has different CPU, GPU and memory-bandwidth characteristics. Its generation speed, prompt-processing speed, memory use and power consumption require measurements on M5 hardware. The same advertised 512 GB capacity does not guarantee identical usable memory or results.
Prepare a useful short evaluation
Choose representative tasks
Bring a small set of prompts, inputs and expected outcomes from the work you actually plan to do. Decide which models, quantizations and context lengths to compare.
Record the environment
Save the model artifact and revision, runtime version, macOS version and settings. Allow time to download models and configure your software.
Measure the workload
Record model load success, peak memory, prompt-processing time, generation time and output quality for the same inputs. Leave room for macOS, the runtime and the KV cache: the full 512 GB is not available for model weights. Test your intended context length rather than relying on a short prompt.
Try the complete workflow
Run your scripts and integrations, including remote development over your own network. Check that the tools and output formats suit your task.
Keep results for the next machine
Export the scripts, configuration and measurements you need before the rental ends. Repeat performance and memory tests on M5 hardware when it arrives.
Take a useful evaluation record with you
Download the free Markdown worksheet before your rental. Record your own measurements, save the results before access ends, then repeat the same workload on M5 hardware. No registration is needed to download the worksheet.
What you will record
- Model artifact, revision, quantization, runtime and macOS versions
- Context length, prompt and generation tokens, load success and errors
- Peak memory with the measurement tool and units, prompt-processing time and generation rate
- Your quality criterion, observed results and the same workload measured again on M5
The worksheet is free. Mac access uses the published paid rental plans.
Questions before you rent
Is the rental machine an M5 Ultra?
The rental machine is a Mac Studio M3 Ultra with 512 GB unified memory and a 4 TB SSD, hosted in Kobe, Japan. It is dedicated to one renter at a time and accessed remotely. Nothing is shipped.
Can I rent before ordering an M5 Ultra?
Yes. You can evaluate your models, quantization choices and macOS workflow on M3 Ultra before making a hardware purchase, or use the rental while waiting for an existing order.
Can M3 Ultra measurements predict M5 Ultra performance?
Results describe the tested M3 Ultra configuration. Differences in CPU, GPU, memory bandwidth and runtime behavior mean M5 Ultra speed, usable memory and power consumption need testing on M5 hardware.
Can I start using the Mac immediately?
Access depends on current machine availability, payment, preparation, allocation and connection readiness. Review the current availability and start conditions in the pricing and rental confirmation pages.
Are the models and development environment already configured?
The rental provides a clean macOS environment. You install and configure your own models, runtimes and development tools. Environment-setup support is not included.
How long does an M5 Ultra 512 GB take to ship?
On October 31, 2026, the selected M5 Ultra 36-core CPU, 80-core GPU, 512 GB memory and 1 TB SSD configuration displayed 13–15 weeks to dispatch in Apple’s US, Japanese, UK, German and Australian stores. This dated estimate is not an arrival date; check Apple for your current configuration and destination.
Test your workload on a dedicated 512 GB Mac
Choose a published rental term for your evaluation. The rental duration is fixed by the plan you purchase; it does not track an Apple order or automatically continue until your M5 arrives.
3 days · $299 · one-time payment