News

Apple’s New M1 Ultra SoC Claims to Pack a Punch in Performance

March 10, 2022 by Jake Hertz

Keeping its in-house chip momentum rolling, Apple unveils its newest member of the M1 family, the M1 Ultra. Let's dive into this new system-on-chip (SoC) and see what it brings to the table.

When Apple first switched to in-house silicon for the Mac, many were left wondering what the company’s motives were. 

Did Apple want uniformity across its platform? Was it end-to-end control? Or did Apple believe in itself to design better silicon than what existed on the market?

 

Apple’s line of Mac silicon.

Apple’s line of Mac silicon. Image used courtesy of Apple

 

Since then, Apple has released a slew of impressive silicon offerings, and the more they release, the harder it is to deny that Apple may very well be able to out-design its competitors. 

This week, Apple announced what it believes is its most impressive silicon offering yet, the M1 Ultra, an SoC that claims unprecedented performance levels.

This article will look at what’s known about the chip, how well it performs, and the technology that makes it possible.

 

Apple's M1 Ultra: What’s Inside?

From an extremely high level, Apple’s new M1 Ultra is a seemingly simple piece of technology. 

The new SoC essentially consists of three major components: two M1 Max dies connected by Apple’s proprietary and newly announced UltraFusion Technology.

By combining two M1 Max dies together on one package, the new SoC reaches a total transistor count of 114 billion, a number which Apple claims is the most ever in a personal computer chip. 

The M1 Ultra features a 20-core central processing unit (CPU), consisting of 16 performance cores supported by 192 KB of instruction cache, 128 KB of data cache, and 48 MB total L2 cache. Additionally, it has four high-efficiency cores, supported by 128 KB of instruction cache, 64 KB of data cache, 8 MB of total L2 cache.

 

The M1 Ultra consists of two M1 Max dies (top and bottom) connected by Apple’s UltraFusion technology (middle).

The M1 Ultra consists of two M1 Max dies (top and bottom) connected by Apple’s UltraFusion technology (middle). Image used courtesy of Apple

 

On top of those specs, they are accompanied by a 64-core GPU capable of up to 196,608 concurrent threads and 21 teraflops of performance and a 32-core neural engine capable of 22 TOPS. 

Overall the SoC can be configured with up to 128 GB of unified memory, almost 3x more than the most powerful graphics processing unit (GPU) offerings on the market. 

While this is impressive, the real talking point of the M1 Ultra is the UltraFusion Technology. 

By using a silicon interposer to connect the two M1 dies via over 10,000 signals, UltraFusion can provide 2.5 TB/s communication between dies. 

Compared to offerings like NVLink from NVIDIA or Infinity Fabric from AMD, UltraFusion offers 4x more bandwidth. Even better, with UltraFusion, the software can recognize the two GPUs as a single GPU, meaning easier programming and significantly increased performance for applications running on the M1 Ultra.

 

M1 Ultra's Performance Specs

Beyond its innovative new technology and extremely high transistor count, the M1 Ultra claims some unprecedented performance specs and power consumption results.

In terms of multi-threaded performance, the M1 Ultra claims to deliver 90% higher performance than the fastest competing 16-core PC chip in the same power envelope. 

The M1 Ultra also claims it can reach its peak performance using 100 fewer watts than its competitors.

 

The M1 Ultra matches competitor performance while consuming 200 W less power.

The M1 Ultra matches competitor performance while consuming 200 W less power. Image used courtesy of Apple

 

Even more impressive, compared to the fastest GPU on the market, NVIDIA’s RTX 3090, the M1 Ultra’s GPU states it can achieve faster performance while requiring 200 W less power. 

Finally, the M1 Ultra is said to be the only chip on the market capable of playing back up to 18 streams of 8K ProRes 422 video.

 

Apple Silicon's Industry Impact

In a little less than two years, Apple has gone from no in-house silicon for its PCs to having arguably the most powerful family of PC silicon on the market. 

This shift from Apple, and even future companies moving to in-house chip fabrication, may strike fear in competitors such as NVIDIA and AMD, whose best-performing devices were seemingly blown out of the water by the M1 Ultra.

Fortunately for these competitors, the M1 family is not for sale and can only be utilized within Apple products. 

So, while Apple may claim the spot of best PC silicon, its negative impact on companies like AMD and NVIDIA will likely not extend beyond the subsequent increase in Mac sales. 

However, the pressure is now on for competitors to figure out how Apple could achieve such performance and efficiency, likely meaning an attempt to replicate UltraFusion to enable high throughput, low latency die-die connectivity.

From the consumer perspective, this should hopefully lead to overall better market offerings in years to come as the rest of the industry plays catchup.