Apple's M2 Ultra SoC is being benchmarked right and left and we can compare those numbers to AMD, Intel, and NVIDIA chips to see if the new SoC really delivers the claimed performance figures.
Apple M2 Ultra Is Slow Than Last-Year CPUs From Intel & AMD, 50% Slower Than NVIDIA RTX 4080
The Apple M2 Ultra SoC is the flagship design that the company has on offer. It features a total of 134 Billion transistors, 24 CPU cores running in a 16 (Performance) and 8 (Efficiency) configuration, up to 76 GPU cores, & 192 GB of memory that is unified and accessible across the whole chip at a maximum bandwidth of 800 GB/s. There's also the new Neural Engine which has been upgraded with 32 cores and the company is claiming a 20% speedup on the CPU & 30% speedup on the GPU side.
The performance benchmarks of the Apple M2 Ultra SoC were leaked within Geekbench 6 and the older Geekbench 5. We get a taste of both the CPU and GPU performance since compute results have also been leaked. So let's start with the CPU side, we first have the clock speeds which range from sub-3 GHz up to 3.7 GHz. We obviously can't say for sure which are the Performance and Efficiency cores since the benchmark doesn't separate the cores based on their architecture or states.
The CPU managed to post a single-core score of up to 2809 points in single-core and 21,531 points in the multi-core tests. For comparison, the Intel Core i9-13900KS scores 3083 points while AMD's Ryzen 9 7950X scores 2875 points. In multi-threaded benchmarks, the same chips score 21665 and 19342 points, respectively. So as you can see, the workstation-grade Apple M2 Ultra SoC isn't faster than the mainstream CPU offerings from Intel and AMD.
Apple M2 Ultra SoC Geekbench 6 (Single-Threaded)
Apple M2 Ultra SoC Geekbench 6 (Multi-Threaded)
If you compare the chip to something like an AMD Threadripper and Intel Xeon W chip, then those would absolutely crush the M2 Ultra in the multi-threaded tests but the single-threaded lead will be on Apple's M2 Ultra since AMD & Intel Pro CPUs are designed for multi-threaded heavy workloads and that's where they extend their multi-core leadership to.
Apple M2 Ultra SoC Geekbench 5 (Single-Threaded)
Apple M2 Ultra SoC Geekbench 5 (Multi-Threaded)
Moving on to the Compute benchmarks, here's where I saw comparisons being made against the NVIDIA RTX 4080 and AMD GPUs within the Metal API which is predominantly optimized for Apple SoC's. NVIDIA and AMD GPUs aren't known for their Metal specific optimizations since the market share of those GPUs running Apple OS is very small and non-existent. As such, in OpenCL, the M2 Ultra SoC ends up 50% slower than NVIDIA's RTX 4080 and that's not even the flagship GPU.
Apple M2 Ultra SoC Geekbench 5 (OpenCL Compute)
In the more popular OpenCL API, the Apple M2 Ultra only ends up close to the NVIDIA GeForce RTX 4060 Ti which is a mainstream graphics card and yes, this is the full 76 GPU core configuration of the M2 Ultra that we are comparing.
Once again, these benchmarks prove that it's wiser to take the numbers and performance presented by Apple (and all other companies for that matter) with a grain of salt and look at independent results to see if the numbers hold up to their claims. We've only used mainstream parts for comparison in these benchmarks and using workstation-oriented solutions will tilt the results more in favor of Intel, NVIDIA, and AMD who have much more powerful CPU and GPU designs being offered to Pro and Workstation users out there.
Furthermore, Apple's M2 Ultra SOC-based Mac Pro offerings will come at a base price of $7000 US and that's not even using the full SoC specs or the 192 GB unified memory.
from Wccftech https://ift.tt/u1ZK79r
0 Comments