Is the next step for RISC-V a data center or a cell phone SoC?

Whether it's the Internet of Things, automotive, or even aerospace, there is no shortage of RISC-V. However, there is still a big gap between RISC-V and the real mainstream vision. Take the continuing hot data center market. RISC-V in this area is still relatively quiet. But, in RISC-V's grand plan, data center products are coming out one by one and may even enter the ARM-dominated cell phone SoC market.

Intel's mysterious RISC-V products

Intel has been making many moves in RISC-V recently, whether joining the RISC-V International Foundation, providing design and manufacturing platforms for RISC-V, etc. RISC-V seems quite promising in the eyes of Intel, the x86 giant.

WikiChip's David Schor revealed that Intel and SiFive collaborated on the design of Horse Creek, a platform based on the Intel 4 process and the SiFive RISC-V core P550, which has recently been streamed to the lab, and the initial test results are "very good. " Although it is not clear what Intel's definition of Horse Creek is, the P550 is a high-performance core, based on an advanced process like Intel 4, and also uses Intel's inherited Creek naming, so this product, even if it is not a data center product, is certainly not unrelated to high performance.

Let's look at the press release of this product when it was first released in June 2021. The product will be available in a wide range of sizes and sizes. The product will be officially released this year, depending on Intel's subsequent development progress.

RISC-V is about to make an impact in the data center

ARM recently commented on RISC-V at a foreign media conference in Neoverse. They believe that RISC-V does not currently show strong competitiveness in the data center space and cannot threaten ARM even in the short term. Although RISC-V stands on the shoulders of giants, it still lacks the accumulation of architecture, standards, and software in the ecology, and there is still a lot to learn from ARM in the governance model.

It is important to know that ARM has only entered the data center market after years of sedimentation, and its market share is still small compared to x86, about 5%, according to the latest statistics. But even so, ARM does not regard RISC-V as a competitor enough to see how weak RISC-V is in this area.

RISC-V's current ecology in the data center is indeed weak, as a competitor to ARM, in addition to several server CPUs released, has also begun to enter the emerging field of DPU, but also with almost all the top cloud service vendors reached a cooperation, it can be said that ARM in the server has been in a full sprint state. On the contrary, RISC-V side, data center level products mainly stay in the high-performance core stage. Most of the rest are AI chips, such as hardware gas pedal products. Many have not been mass production, let alone commercial.

But we have seen RISC-V from the embedded core on the development of speed, and this speed is about to sweep the data center market. For example, for the general-purpose server CPU, there are two companies to calculate the energy technology and micro core news, and the mass production is imminent, directly against ARM's Neoverse product line, and even after the product route has been planned.

In fact, since the birth of the RISC-V not long ago, this progress is already very good, at least in the field of AI chips. RISC-V is already known to everyone, more Esperanto and Tenstorrent, such excellent startups in the launch of high-performance AI products, the home of these chips is also the data center. In manufacturing, TSMC, Intel IFS, and Lattice continue to have RISC-V high-performance chips in the pipeline. Intel has also partnered with the Barcelona Supercomputing Center to develop RISC-V high-performance processors that will push supercomputing to the Zettascale level.

Does Apple want to switch to RISC-V?

RISC-V does not need to be a leader in the core computing field from the beginning but can be done from low-power MCU to cell phone SoC, from AI chip to server CPU., such design ideas have long been taken into account, and there is already a part of the chip on the market began to use ARM + RISC-V dual architecture, first take the RISC-V core as a co-processor.

But this is not enough. RISC-V needs to be like a smartphone SoC, home computer CPU, or GPU, such as the real volume of the core of the explosive products so that when people hear the word "RISC-V," they will not feel strange. In light of recent news, Apple will likely be the first to take the first step in RISC-V mobile SoCs, but the result may not be what we think, but rather, as mentioned above, penetration from the "edge."

SemiAnalysis' Dylan Patel has confirmed that Apple is actively converting some ARM cores into RISC-V cores for future Apple Silicon chips. Combined with last year's news that Apple was recruiting RISC-V engineers, this news is not less credible. In last year's public recruiting information, Apple mentioned in the job requirements that it has knowledge of ARM's NEON and uses RISC-V to build embedded subsystems for Apple's multi-device ecosystem.

NEON is ARM's SIMD instruction set extension, similar to X86's AVX512, SSE, etc. The 128-bit wide NEON instruction set is used in Apple's A-series ARM CPUs. So why did Apple switch the core of these subsystems to RISC-V?

Here's a quote from SiFive's description of RISC-V vector extensions: RISC-V vector extensions are a powerful and efficient solution in terms of code size, performance, and area, making up for the lack of efficiency of packed-SIMD and GPUs on large processor data sets, and ARM's NEON is exactly that for packed-SIMD.

Another feature of RISC-V vector extensions is that they are vector length independent, and code written for their RISC-V vector extension-compatible processors can run on other RISC-V vector-compatible processors, reducing the difficulty of having to rewrite code after hardware updates. For Apple, this provides the convenience of new system compatibility and, more importantly, allows new software features to be implemented in iOS, WatchOS, and tvOS.

For Apple, which has just fully switched to ARM, it seems a bit unrealistic to throw themselves into the RISC-V camp now, considering how much they've already invested in microarchitecture, toolchains, and systems. But Apple might not mind changing if it brings huge advantages to some subsystems regarding power consumption and programming. After all, they have already taken some action.

Take Apple's current A15 SoC, for example; many of these subsystems also use the ARM core, such as ISP, display engine, and video codec. But in these gas pedal execution units, what is used is Apple's own AMX instruction set.

Write at the end

Regardless of how these ISAs evaluate each other, having such a diverse ecosystem is good for the industry, with more choices and opportunities for innovation. Although the current RISC-V in the data center is still lagging behind a lot, and there is no big breakthrough product launch, the cultivation process can give the newborn architecture a solid foundation. The gap is just the catalyst for developing the entire ISA ecosystem.