Apple M1 Ultra (48-GPU) Benchmark, Test and specs
Last updated:
The Apple M1 Ultra (48-GPU) has 20 CPU cores and can process 20 threads at the same time. The processor was presented in Q1/2022 and is based on the 1. Generation of the Apple M series series. In the Geekbench 5 benchmark, the Apple M1 Ultra (48-GPU) achieved a result of 1,793 points (single-core) or 24,055 points (multi-core).
| Name: | Apple M1 Ultra (48-GPU) |
|---|---|
| Family: | Apple M series |
| CPU group: | Apple M1 |
| Architecture: | M1 |
| Segment: | Mobile |
| Generation: | 1 |
| Predecessor: | -- |
| Successor: | Apple M2 Ultra (60-GPU) |
CPU Cores and Base Frequency
The Apple M1 Ultra (48-GPU) has 20 cores. The clock frequency of the Apple M1 Ultra (48-GPU) is 3.20 GHz. An initial performance assessment can be made using the number of CPU cores.
| CPU Cores / Threads: | 20 / 20 |
|---|---|
| Core architecture: | hybrid (big.LITTLE) |
| A-Core: | 16x Firestorm |
| B-Core: | 4x Icestorm |
| Hyperthreading / SMT: | No |
|---|---|
| Overclocking: | No |
| A-Core Frequency: | 3.20 GHz |
| B-Core Frequency: | 2.06 GHz |
Artificial Intelligence and Machine Learning
Processors with the support of artificial intelligence (AI) and machine learning (ML) can process many calculations, especially audio, image and video processing, much faster than classic processors. Algorithms for ML improve their performance the more data they have collected via software. ML tasks can be processed up to 10,000 times faster than with a classic processor.
| AI hardware: | Apple Neural Engine |
|---|---|
| AI specifications: | 32 Neural cores @ 22 TOPS |
Internal Graphics
The Apple M1 Ultra (48-GPU) has an integrated graphics that the system can use to efficiently play back videos. The Apple M1 Ultra (48-GPU) has the Apple M1 Ultra (48 Core) installed, which has 768 streaming multiprocessors (6144 shaders).
| GPU name: | Apple M1 Ultra (48 Core) |
|---|---|
| GPU frequency: | 1.30 GHz |
| GPU (Turbo): | No turbo |
| Compute units: | 768 |
| Shader: | 6144 |
| Max. GPU Memory: | 64 GB |
| Max. displays: | 5 |
|---|---|
| Generation: | 1 |
| Direct X: | -- |
| Technology: | 5 nm |
| Release date: | Q1/2022 |
Hardware codec support
Processors with integrated graphics can process video codecs faster. Support for modern codecs can significantly increase system efficiency during video playback.
| h265 / HEVC (8 bit): | Decode / Encode |
|---|---|
| h265 / HEVC (10 bit): | Decode / Encode |
| h264: | Decode / Encode |
| VP8: | Decode |
| VP9: | Decode / Encode |
| AV1: | No |
|---|---|
| AVC: | Decode |
| VC-1: | Decode |
| JPEG: | Decode / Encode |
Memory & PCIeThe Apple M1 Ultra (48-GPU) supports a maximum of 128 GB memory. Depending on the mainboard, the processor can use a maximum of 8 (Octa Channel) memory channels. This results in a maximum bandwidth of the main memory of 819.2 GB/s. |
|
| Memory type: | Memory bandwidth: |
|---|---|
| LPDDR5-6400 | 819.2 GB/s |
| Max. Memory: | 128 GB |
| Memory channels: | 8 (Octa Channel) |
| ECC: | No |
| PCIe: | 4.0 |
| PCIe Bandwidth: | -- |
Thermal ManagementThe Apple M1 Ultra (48-GPU) has a TDP of 60 W. Based on the TDP, the system manufacturer can and must adapt the cooling solution to the processor. |
|
|---|---|
| TDP (PL1 / PBP): | 60 W |
| TDP (PL2): | -- |
| TDP up: | 120 W |
| TDP down: | -- |
| Tjunction max.: | -- |
Technical details
Modern production reduces the waste heat of a processor and increases its efficiency. The Apple M1 Ultra (48-GPU) is made in 5 nm and has 52.00 MB cache.
| Technology: | 5 nm |
|---|---|
| Chip design: | Chiplet |
| Socket: | -- |
| L2-Cache: | 52.00 MB |
| L3-Cache: | -- |
| AES-NI: | Yes |
| Operating systems: | macOS |
| Virtualization: | Apple Virtualization Framework |
|---|---|
| Instruction set (ISA): | ARMv8-A64 (64 bit) |
| ISA extensions: | Rosetta 2 x86-Emulation |
| Release date: | Q1/2022 |
| Release price: | -- |
| Part Number: | -- |
| Documents: | -- |
20C 20T @ 3.20 GHz
buy now on Amazon and save!
Benchmark results
The benchmark results for the Apple M1 Ultra (48-GPU) have been carefully checked by us. We only publish benchmark results that have been created by us or that have been submitted by a visitor and then checked by a team member. All results are based on and fullfill our benchmark guidelines.
Screenshots:
Screenshots:
Cinebench 2024 (Single-Core)
The Cinebench 2024 benchmark is based on the Redshift rendering engine, which is also used in Maxon's 3D program Cinema 4D. The benchmark runs are each 10 minutes long to test whether the processor is limited by its heat generation.
|
AMD Ryzen 5 7600X
6C 12T @ 5.30 GHz |
||
|
Intel Core i5-13600K
14C 20T @ 5.10 GHz |
||
|
|
Intel Core i5-13600KF
14C 20T @ 5.10 GHz |
||
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
AMD Ryzen 9 7945HX
16C 32T @ 5.40 GHz |
||
|
AMD Ryzen 7 7700
8C 16T @ 5.30 GHz |
||
|
|
AMD Ryzen 9 7900
12C 24T @ 5.40 GHz |
||
Cinebench 2024 (Multi-Core)
The Multi-Core test of the Cinebench 2024 benchmark uses all cpu cores to render using the Redshift rendering engine, which is also used in Maxons Cinema 4D. The benchmark run is 10 minutes long to test whether the processor is limited by its heat generation.
|
Intel Core i9-13900HX
24C 32T @ 5.40 GHz |
||
|
|
AMD Ryzen 9 7945HX
16C 32T @ 5.40 GHz |
||
|
|
AMD Ryzen 9 7900
12C 24T @ 5.40 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Intel Core i9-13900
24C 32T @ 5.60 GHz |
||
|
|
AMD Ryzen 9 7900X
12C 24T @ 5.60 GHz |
||
Cinebench R23 (Single-Core)
Cinebench R23 is the successor of Cinebench R20 and is also based on the Cinema 4 Suite. Cinema 4 is a worldwide used software to create 3D forms. The single-core test only uses one CPU core, the amount of cores or hyperthreading ability doesn't count.
|
AMD Ryzen 7 PRO 5750G
8C 16T @ 4.60 GHz |
||
|
Intel Core i7-1185G7
4C 8T @ 4.80 GHz |
||
|
|
Intel Core i9-11900T
8C 16T @ 4.90 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Pro (10-CPU 14-GPU)
10C 10T @ 3.20 GHz |
||
|
|
Apple M1 Max (24-GPU)
10C 10T @ 3.20 GHz |
||
Cinebench R23 (Multi-Core)
Cinebench R23 is the successor of Cinebench R20 and is also based on the Cinema 4 Suite. Cinema 4 is a worldwide used software to create 3D forms. The multi-core test involves all CPU cores and taks a big advantage of hyperthreading.|
|
Intel Core i5-13600K
14C 20T @ 5.10 GHz |
||
|
|
Intel Core i5-13600KF
14C 20T @ 5.10 GHz |
||
|
|
Intel Core i9-10980XE
18C 36T @ 4.30 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
20C 20T @ 3.20 GHz |
||
|
|
AMD Ryzen 9 3950X
16C 32T @ 4.00 GHz |
||
|
|
Intel Core i7-12700KF
12C 20T @ 4.70 GHz |
||
Geekbench 5, 64bit (Single-Core)
Geekbench 5 is a cross plattform benchmark that heavily uses the systems memory. A fast memory will push the result a lot. The single-core test only uses one CPU core, the amount of cores or hyperthreading ability doesn't count.|
|
Intel Core i5-13600T
14C 20T @ 4.80 GHz |
||
|
|
Intel Core i7-11700K
8C 16T @ 5.00 GHz |
||
|
|
Intel Core i7-11700KF
8C 16T @ 5.00 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Intel Core i7-12800HX
16C 24T @ 4.80 GHz |
||
|
|
Intel Core i7-12850HX
16C 24T @ 4.80 GHz |
||
Geekbench 5, 64bit (Multi-Core)
Geekbench 5 is a cross plattform benchmark that heavily uses the systems memory. A fast memory will push the result a lot. The multi-core test involves all CPU cores and taks a big advantage of hyperthreading.
|
AMD EPYC 7532
32C 64T @ 2.90 GHz |
||
|
|
AMD EPYC 7413
24C 48T @ 3.40 GHz |
||
|
|
AMD Ryzen 9 7950X
16C 32T @ 5.20 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Intel Core i9-13900KF
24C 32T @ 5.50 GHz |
||
|
|
Intel Core i9-13900K
24C 32T @ 5.50 GHz |
||
Geekbench 6 (Single-Core)
Geekbench 6 is a benchmark for modern computers, notebooks and smartphones. What is new is an optimized utilization of newer CPU architectures, e.g. based on the big.LITTLE concept and combining CPU cores of different sizes. The single-core benchmark only evaluates the performance of the fastest CPU core, the number of CPU cores in a processor is irrelevant here.|
|
Intel Core i9-12900HX
16C 24T @ 5.00 GHz |
||
|
|
Intel Core i5-13490F
10C 16T @ 4.80 GHzNot verified |
||
|
|
Intel Core i5-13600HX
14C 20T @ 4.80 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Pro (10-CPU 14-GPU)
10C 10T @ 3.20 GHz |
||
|
|
Apple M1 Max (24-GPU)
10C 10T @ 3.20 GHz |
||
Geekbench 6 (Multi-Core)
Geekbench 6 is a benchmark for modern computers, notebooks and smartphones. What is new is an optimized utilization of newer CPU architectures, e.g. based on the big.LITTLE concept and combining CPU cores of different sizes. The multi-core benchmark evaluates the performance of all of the processor's CPU cores. Virtual thread improvements such as AMD SMT or Intel's Hyper-Threading have a positive impact on the benchmark result.
|
Intel Xeon w9-3495X
56C 112T @ 2.90 GHz |
||
|
|
Intel Xeon w7-2495X
24C 48T @ 3.30 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
20C 20T @ 3.20 GHz |
||
|
|
AMD EPYC 9554
64C 128T @ 3.75 GHz |
||
|
|
AMD EPYC 9654P
96C 192T @ 3.55 GHz |
||
|
|
AMD EPYC 9654
96C 192T @ 3.55 GHz |
||
iGPU - FP32 Performance (Single-precision GFLOPS)
The theoretical computing performance of the internal graphics unit of the processor with simple accuracy (32 bit) in GFLOPS. GFLOPS indicates how many billion floating point operations the iGPU can perform per second.|
|
Apple M2 Ultra (76-GPU)
Apple M2 Ultra (76 Core) @ 1.40 GHz |
||
|
|
Apple M2 Ultra (60-GPU)
Apple M2 Ultra (60 Core) @ 1.40 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
Apple M1 Ultra (64 Core) @ 1.30 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
Apple M1 Ultra (48 Core) @ 1.30 GHz |
||
|
|
Apple M2 Max (38-GPU)
Apple M2 Max (38 Core) @ 1.40 GHz |
||
|
|
Microsoft XBox Series X
AMD Custom Radeon Graphics (XBox Series X) @ 1.83 GHz |
||
|
|
Apple M2 Max (30-GPU)
Apple M2 Max (30 Core) @ 1.40 GHz |
||
Blender 3.1 Benchmark
In the Blender Benchmark 3.1, the scenes "monster", "junkshop" and "classroom" are rendered and the time required by the system is measured. In our benchmark we test the CPU and not the graphics card. Blender 3.1 was presented as a standalone version in March 2022.
|
AMD Ryzen Threadripper 2970WX
24C 48T @ 3.00 GHz |
||
|
|
Intel Core i9-12900KS
16C 24T @ 5.20 GHz |
||
|
|
AMD Ryzen 9 3950X
16C 32T @ 4.00 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Intel Core i9-12900KF
16C 24T @ 5.00 GHz |
||
|
|
Intel Core i9-12900K
16C 24T @ 5.00 GHz |
||
Performance for Artificial Intelligence (AI) and Machine Learning (ML)
Processors with the support of artificial intelligence (AI) and machine learning (ML) can process many calculations, especially audio, image and video processing, much faster than classic processors. The performance is given in the number (trillions) of arithmetic operations per second (TOPS).|
|
Apple M2 Ultra (60-GPU)
24C 24T @ 3.50 GHz |
||
|
|
Apple M2 Ultra (76-GPU)
24C 24T @ 3.50 GHz |
||
|
Qualcomm Snapdragon 888
8C 8T @ 2.84 GHz |
||
|
|
Apple M1 Ultra (48-GPU)
20C 20T @ 3.20 GHz |
||
|
|
Apple M1 Ultra (64-GPU)
20C 20T @ 3.20 GHz |
||
|
Apple A16 Bionic
6C 6T @ 3.46 GHz |
||
|
|
Apple M2 Max (30-GPU)
12C 12T @ 3.50 GHz |
||
Benchmarks
Cinebench 2024 (SC)
177 entries
177 entries
Cinebench 2024 (MC)
178 entries
178 entries
Cinebench R23 (SC)
527 entries
527 entries
Cinebench R23 (MC)
504 entries
504 entries
Geekbench 5 (SC)
2,399 entries
2,399 entries
Geekbench 5 (MC)
2,372 entries
2,372 entries
Geekbench 6 (SC)
1,648 entries
1,648 entries
Geekbench 6 (MC)
1,599 entries
1,599 entries
FP32 SP (iGPU)
1,925 entries
1,925 entries
Blender 3.1 Benchmark
212 entries
212 entries
AI / ML Performance
50 entries
50 entries
News and articles for the Apple M1 Ultra (48-GPU)
Apple M2 Pro with more CPU cores and production in 3 nm ?
Posted by Stefan on 2022-06-28
After Apple surprisingly presented the normal Apple M2 processor as the first successor to the Apple M1 already in June 2022, the Apple M2 Pro and the Apple M2 Max may follow in autumn. The difference in performance compared to the basic version could be even greater this time and range from 25 to 40 percent.
In this article we describe how Apple could realize this and why the price should also go up.
In this article we describe how Apple could realize this and why the price should also go up.
Apple M2 vs Apple M1 - What are the differences?
Posted by Stefan on 2022-06-09
During WWDC 2022 on June 6th, 2022 Apple surprisingly presented its new Apple M2 processor. This will initially be installed in a completely revised Apple MacBook Air with 13.6 inches, a even thinner form factor and new colors. Apple also updated the already known (old) Apple MacBook Pro 13.3 with touch bar.
Many leakers had counted on Apple not presenting its new second-generation M processor until the fall. But things turned out differently. And there is very likely a reason for that: the improvements in the Apple M2 are limited compared to the predecessor.
Many leakers had counted on Apple not presenting its new second-generation M processor until the fall. But things turned out differently. And there is very likely a reason for that: the improvements in the Apple M2 are limited compared to the predecessor.
Description of the processor
The Apple M1 Ultra has a total of 20 CPU cores, which with a clock frequency of 3.2 GHz (performance core) resp. 2.06 GHz (efficiency core) at the same level as already with the normal Apple M1 resp. the larger expansion stages Apple M1 Pro and Apple M1 Max lie.The Apple M1 Ultra with 48-GPU cores has the same number of CPU cores as the Apple M1 Ultra with 64 GPU cores. The only difference between the two Apple M1 Ultra processors is the internal graphics unit (iGPU).
The Apple M1 Ultra is currently the largest and fastest Apple processor. It combines two individual Apple M1 Max processor chips via an interface with a bandwidth of 2.5 TB / s. A similar technology has long been used by AMD in its AMD Ryzen processors.
In addition to the CPU cores, the two graphics units (24 GPU cores each) are synchronized via the interfaces of the two processors in the Apple M1 Ultra. The interface itself, according to Apple, consists of 1000 separate lines, which make such a large bandwidth possible at all.
The storage bandwidth of the LPDDR5-6400 memory is close to 820 GB / s, which is approx. 10x as fast as current desktop processors from AMD and Intel. Since Apple does not put any external graphics device in its new products, the memory does not need to be so fast to slow down the GPUs of the Apple M1 Ultra. Processor and graphics unit divide dynamically up to 128 GB large memory.
The Apple M1 Ultra with 48 GPU cores achieves an FP32 computing power (simple accuracy) of 15.9 TFLOP / s. This is 25% less than the larger expansion stage of the Apple M1 Ultra with 64-GPU cores. This results in an FP32 performance of 21.2 TFLOP / s. Nevertheless, even the 48-core variant of the GPU is clearly superior to the iGPU of the competition and approx. 4x as fast as the iGPU of the AMD Ryzen 9 6980HX. Even in direct comparison with a dedicated graphics card from AMD or NVIDIA, the new Apple M1 Ultra looks good and is at about the level of an AMD Radeon RX 6800 (16.2 TFLOP / s) respectively. an NVIDIA RTX 3060 Ti (16.5 TFLOP / s).
Popular comparisons
