The new RISC-V CPU claims record performance per watt

Micro Magic Inc. is a small electronics design company based in Sunnyvale, California. Has developed a prototype CPU that is many times more efficient than the world’s leading competitors while maintaining reasonable source performance.

We first noticed claims by Micro Magic during the EE Times earlier this week Announced In the company’s new prototype CPU, it appears to be the fastest RISC-V CPU in the world. Micro Magic consultant Andy Huang said the CPU can produce 13,000 Gormarks (later more) at 5GHz and 1.1V, while releasing 11,000 Gormarks at 4.25GHz – the latter, using only 200mW. At 4.327GHz / 0.8V, and 5.19GHz / 1.1V the CPU proved to run on an Odroid board for the EE Times.

Later that week, Micro Magic Announced The same CPU can produce 8,000 Gormarks at 3GHz, while using only 69mW of power.

OK, but what is Gormark?

Part of the difficulty lies in figuring out what Micro Magic claims for its new CPU Gormark And how many of them are needed to build a faster CPU. This embedded microprocessor is a deliberately simplified CPU benchmarking tool released by the Benchmark Federation that should be platform-neutral and as simple to configure and use as possible. Gormark focuses only on the main pipeline functions of a CPU, including basic read / write, integer and control functions. This particularly avoids most of the effects of system differences on memory, I / O and more.

The Embedded Microprocessor Benchmark Federation (EMPC) is a group with broad industry representation: Intel, Texas Instruments, ARM, Realtech and Nokia are some of its notable and easily identifiable members.

Now that we understand all that, the next step in better evaluating the claims of micro magic is to run some of our own Gormark definitions. All we have to do here is clothe its GitHub Repository, Then a make Command – optionally, with arguments XCFLAGS="-DMULTITHREAD=8 -DUSE_FORK=1" If you want to test on multiple threads / cores at once.

I still have an Apple M1 Mac Mini, as well as the Raison 7 4700 U-powered Acer Swift 3, so they are my test systems for comparison. It is much easier to obtain source performance scores than it is to obtain truly comparable power measurements. On a Linux computer running Raison, I used the app turbostat Obtain both core and set power measurements when running tests.

I have no access to almost anything turbostat As for the Apple M1, I picked up the full-system power draw on the wall for that site, and spent most of the reading on the desktop idle from prolonged reading during testing. This is very bitter, and I warn readers not to rely too much on the performance of the M1 in these numbers compared to the Swift 3 – but Micro Magic is enough to get some perspective on its new RISC-V (pronounced “risk claim. Five”) CPU.

For trials!

The Micro Magic CPU, for now, is single-core and single-threaded — according to Huang, which can be built “easily” as a 25-core part. Micro Magic has provided statistics — in one case, a screen shot for performance at 3GHz, 4.25GHz and 5GHz. At 3GHz clocked at maximum power, the Micro Magic CPU scores 4700u or 1/4 Goremark of the Apple M1. At the maximum 5GHz clock, it manages one-third of their performance.

This suffices to inform us that the Micro Magic chip, the world-class competitor for traditional ARM and x86 CPUs in phone or laptop applications, is not in its current form – but it is much closer to them than previous RISC-V implementations. On a powerful 3GHz clock, the Micro Magic CPU is almost three times faster, for example, SiFive’s Freedom U540 CPU Single-threaded running. At 5GHz, it has more than four cores of SiFive.

With about a quarter of the performance of the world’s leading x86 and ARM mobile processors, the Micro Magic CPU is not yet high. But when we factor in power performance, things go crazy. While creating the above charts I gave every possible doubtful effect to my Risen and Apple processors. As for Apple, I only had access to the full-system power draw, so I deducted the “desktop passive” power draw from the “under test” power draw.

I tested Apple and AMD CPUs both single-threaded and multitread when checking power performance. Surprisingly, both parts produced more performance per watt when exercised with a work thread for every CPU thread available. None of these are as efficient as the command of Micro Magic.

At 4.25GHz, the Micro Magic Raison 4700U can accomplish less than a third of the workload. At 3GHz, that number drops to less than oneEighth Required power.

Which is better?

The Linux operating system already supports the RISC-V architecture – so headless or near-head controllers should provide decent performance paired with extreme power performance, while Micro Magic’s new CPU will most likely be there. Things get more complicated once we start talking about whole, consumer friendly systems. With the exception of hardware concepts such as GPU and LTE modem, creating a full Android phone based on a non-ARM architecture would be a huge effort.

If we take the numbers of micro magic for granted, it is necessary to point out that they are Already Beats the performance of some solid mobile phone CPUs. Even on its performance-first 3GHz clock, the Micro Magic CPU performed better than a Qualcomm Snapdragon 820.

If we use EMBC Published Single core score for Snapdragon 820 and Ananttek’s single core CPU power test As a result, We get about 16,000 Gormarks per watt. It triples the performance of the single-threaded Raison 4700u, and is slightly better than the Raison optimally operated with multitreaded workload.

In other words, the prototype CPU of Micro Magic is both significantly faster And A reasonably modern and yet highly efficient smartphone with more power capacity than the CPU.

Conclusion

All of this sounds very exciting – Micro Magic’s new prototype Linux offers solid smartphone-quality performance within a fraction of the power budget, using an instruction package that already runs by default. Even better, the company does not know itself.

Micro Magic was first established in 1995 and was purchased by Juniper Networks for 0 260 million. In 2004, it was reborn under its original name by the original founders, Mark Santoro and Le Tavro, who led the team that first worked on the sun and developed the 300 MHz Spark microprocessor.

Micro Magic wants to offer its new RISC-V design to customers using the IP license model. Simplicity of design – RISC-V requires one-tenth of the modern ARM configuration opcodes – further ease production concerns because RISC-V can create CPU designs The spacecraft is running, Sharing a scaled space with other designs.

That said, it will be a massive effort to take the entire smartphone ecosystem like commercial Android to the new framework. In addition to creating the operating system, drivers for all hardware, from the GPU to the WiFi to the LTE modem, and third-party application developers must reassemble their own applications for the new architecture. .

We still take the claims of Micro Magic at face value. We have seen a screen shot of the 8,200 Gormark score, and we have seen a 69 MW power reading, and it is not entirely clear that the power reading was representative of the full benchmark flow.

Still, this is an amazing development. Not only does the new design not only seem to perform better when it breaks performance records massively, it does so with a more conceptually open design than its competitors. Unlike RISC-V ISA x x86, ARM and MIPS are offered under open and non-royalty licenses.