邁向「青春期」的RISC-V

很長一段時間以來,微處理器市場一直非常同質化。目前有兩個陣營占主導地位。一個圍繞x86架構,另一個遵循Arm。一直有其他選擇,但多年來沒有人認真地與這兩者中的任何一個相媲美。直到最近,都是這樣。現在有第三種選擇變得越來越可行:它是 RISC-V。

時長: 26:12 | 發佈者: EE Times Taiwan | 發佈時間: 2021-12-30

邁向「青春期」的RISC-V

BRIAN SANTO: I’m Brian Santo, EE Times editor-in-chief. You’re listening to EE Times On Air, and this is the Weekly Briefing for the week ending December 17th.

BRIAN SANTO:我是 EE Times 主編 Brian Santo,你正在收聽的是 EE Times On Air。

The microprocessor market has been remarkably homogeneous for a very long time. Two camps currently dominate. One surrounds the x86 architecture, the other follows Arm. There have always been other alternatives, but none seriously rivaled either of those two for years. Until recently, that is. Now there is a third option that is becoming increasingly viable: it is RISC-V.

很長一段時間以來,微處理器市場一直非常同質化。目前有兩個陣營占主導地位。一個圍繞x86架構,另一個遵循Arm。一直有其他選擇,但多年來沒有人認真地與這兩者中的任何一個相媲美。直到最近,都是這樣。現在有第三種選擇變得越來越可行:它是 RISC-V。

So-called “open” technologies have revolutionized one electronics segment after another, starting with software and permeating from there. Processors have so far been immune to the trend, but perhaps for not much longer. RISC-V has an open component to it, and that’s part of its appeal. The technology is gradually generating more and more enthusiasm.

所謂的「開放」技術已經徹底改變了一個又一個電子領域,從軟體開始並從那裡滲透。到目前為止,處理器已經不受這種趨勢的影響,但可能不會太久。RISC-V 有一個開放的元件,這也是其吸引力的一部分。這項技術正逐漸產生越來越多的熱情。

The most recent culmination of that was the 2021 RISC-V Summit, held in San Francisco last week. Our guest today is Tirias Research principal analyst Kevin Krewell, who attended the summit.

最近的高潮是在舊金山舉行的 2021 年 RISC-V 峰會。我們今天的嘉賓是出席峰會的 Tirias Research 首席分析師 Kevin Krewell。

The x86 architecture originated at Intel in the early 1970s. By the turn of the century, it had become the most prevalent processor type. Arm has been around since the mid-‘80s, back when the company was still known as Acorn, but it didn’t start to become as pervasive as it is now until the smartphone era.

x86 架構起源於 1970 年代初期的英特爾。到世紀之交,它已成為最流行的處理器類型。Arm 從 80 年代中期就已經存在,當時該公司還被稱為 Acorn,但直到智慧手機時代它才開始像現在這樣普遍。

For some time now, engineers and product designers have had a choice between those two different processor types. Intel and Arm also have adopted distinct business models, though in both cases, Intel and Arm maintain some control of the processor technologies they’ve devised.

一段時間以來,工程師和產品設計師不得不在這兩種不同的處理器類型之間做出選擇。英特爾和 Arm 也採用了不同的商業模式,儘管在這兩種情況下,英特爾和 Arm 都對他們設計的處理器技術保持了一定的控制權。

Those involved in developing RISC-V intend to compete with a third business model: the open systems model. Now, it’s not entirely open. The part of RISC-V that is open is its instruction set architecture, or ISA. Because it’s open, though, because developing RISC-V is a collaborative endeavor, it promises to be the least expensive of the three, and if it’s not that much less expensive, it may offer more flexibility to product designers. Technologically, RISC-V processors are already competitive. There are RISC-V processors currently available that provide admirable performance. Organizations around the world are helping to refine the technology and specific extensions for specific applications, or sets of applications, as well as programming tools and other support technologies. They are developing an ecosystem.

參與開發 RISC-V 的人打算與第三種商業模式競爭:開放系統模式。現在,它還沒有完全開放。RISC-V 開放的部分是它的指令集架構,或 ISA。不過,因為它是開放的,因為開發 RISC-V 是一項協作努力,所以它有望成為三者中最便宜的,如果它不是那麼便宜,它可能會為產品設計師提供更大的靈活性。從技術上講,RISC-V 處理器已經具備競爭力。目前可用的 RISC-V 處理器可提供令人讚歎的性能。世界各地的組織都在?明改進特定應用程式或應用程式集的技術和特定擴展,以及程式設計工具和其他支援技術。他們正在開發一個生態系統。

It all goes to show that RISC-V is gaining momentum, a point that its backers aimed to reinforce at the RISC-V Summit held in the first week of December. Our old pal Kevin Krewell was at the Summit. Krewell is a principal analyst at Tirias Research. We invited him back to the podcast to talk about what he heard there.

這一切都顯示RISC-V正在獲得動力,其支持者的目標是在 12 月第一周舉行的 RISC-V 峰會上加強這一點。我們的老朋友 Kevin Krewell 在峰會上。Krewell 是 Tirias Research 的首席分析師。我們邀請他來到節目中談談他在峰會上的見聞。

Imagination Technology came out with their Catapult CPUs. What was the big deal there? And then maybe we can put it in context.

Imagination Technology 推出了他們的 Catapult CPU。這件事的重要性何在?或許我們可以來理解其脈絡。

KEVIN KREWELL: Okay. So the interesting part of the context, one thing is: Imagination Technology is known as a provider of GPU IP. Graphics processing, and they were the lead GPU in all the Apple stuff until Apple said, Oh, that’s okay. We can make our own now. And so it hurt the company for a while, but they also at one point in time bought MIPS, then had to spin maps off.

KEVIN KREWELL:好的。所以這件事有趣的部分是:Imagination Technology 是 GPU IP 的提供者。繪圖處理,它們是所有 Apple 產品中的領先 GPU,直到 Apple 說,喔,沒關係,我們現在可以自己來。這對該公司造成了一段時間的傷害,但他們也曾在某個時間點收購了 MIPS,又不得不放棄。

BRIAN SANTO: Finances?

BRIAN SANTO:因為財務?

KEVIN KREWELL: Finances. I think they’re in better financial footing now. They’ve sort of rebooted themselves. You can call it Imagination Technology 2.0 or maybe 3.0. But they’re now getting into the CPU business. And this is one of the advantages of the RISC-V instruction set is that it’s an open instruction set. It’s not a licensable, open licensed cores. I mean, there’s no RTL involved. You have an instruction set, and you have to build your own cores. So in a sense it’s not open source, because there’s no source, there’s just instructions.

KEVIN KREWELL:是財務問題。但我認為他們現在的財務狀況更好。他們已經重新啟動了自己。您可以將其稱為 Imagination Technology 2.0 或 3.0。他們現在要進軍 CPU 業務,這是 RISC-V 指令集的優勢之一,它是一個開放的指令集。它不是可授權的、開放的授權內核。我的意思是,不涉及 RTL。你有一個指令集,你必須構建自己的核心。所以從某種意義上說,它不是開源的,因為沒有原始程式碼,只有說明。

BRIAN SANTO: It’s just open.

BRIAN SANTO:它剛剛開放。

KEVIN KREWELL: It’s an open standard open, is really more what it is. And so Imagination decided that they needed a balance of both CPU and GPU. And so they’re getting into the best way they can do it, considering that it would be difficult for them to license Arm because they’ve always been the anti-Arm guys, the opposite of Arm.

KEVIN KREWELL:這是一個開放標準的開放,實際上更是如此。因此,Imagination 決定他們需要在CPU和GPU之間取得平衡--考量到他們很難取得Arm的授權,因為他們一直是與Arm對立的陣營,他們正在盡其所能地做到這一點。

BRIAN SANTO: They’ve deliberately aggravated Arm on a couple of occasions.

BRIAN SANTO:他們有幾次故意激怒 Arm。

KEVIN KREWELL: Absolutely. And so therefore, RISC-V is the perfect way for them to get into the CPU business and further aggravation of the Arm ecosystem or Arm itself. But it also puts them in an unusual position where they’re actually going to compete with their former partners, MIPS. Because MIPS also announced earlier this year that their future architecture is not going to be MIPS, it’s going to be RISC-V as well. So it’s another little company joining the RISC-V bandwagon.

KEVIN KREWELL:當然。因此,RISC-V是他們進入CPU業務並進一步挑釁Arm生態系統或Arm本身的完美方式。但這也讓他們處於一個不同尋常的位置,他們實際上將與他們的前合作夥伴MIPS競爭。因為MIPS在今年稍早時還宣佈,他們未來的架構不會是MIPS,也將是RISC-V。所以這是另一家搭上RISC-V熱潮的小公司。

And so they’re up to 2.4k members plus at the RISC-V Foundation. So they’re building a nice ecosystem around number of members. So this was actually a big coming out party at the RISC-V Summit, because you had companies like Imagination, you had SiFive. The founders of RISC-V are the elite executives, and SiFive. And they came up with their new core, the 650, which is an enterprise class core, very competitive with the A78 and 7 series of Cortex cores. But because it’s a simpler architecture, it’s a little smaller core, they can do things in a little less energy. So I think they have some real potential here to disrupt Arm as an application processor, not just as a deeply embedded core.

在RISC-V 基金會中擁有多達 2,400家成員,他們正圍繞著這些成員建立一個很好的生態系統。所以這實際上是RISC-V峰會上的一個盛大亮相派對,因為有像Imagination這樣的公司,有SiFive。RISC-V的創始成員都是精英高層,還有SiFive。他們提出了他們的新核心650,這是一個企業級核心,可與 Cortex A78和 7系列核心競爭。但是因為它是一個更簡單的架構,它的核心更小,能耗更小,所以我認為他們有一些真正的潛力可成為顛覆Arm的應用處理器,不僅是作為一個深度嵌入式的核心。

So the important things that are happening with the Summit included, they ratified 16 specifications. One of the things that’s been sort of dragging on has been a lot of these subcommittees doing extensions like hypervisor extensions and vector extensions, they’re finally getting all these things signed off. And therefore, we now kinda like stabilized the ecosystem with known, fixed release specification extensions to the basic architecture. And that helps move the whole ecosystem forward because now they have something to design to that is universally accepted by the RISC-V community.

因此,峰會發生的重要事情包括,他們批准了 16 項規格。一直拖延的事情之一是很多子委員會都在進行擴展,例如管理程式擴展和向量擴展,他們終於讓所有這些事情都得到了批准。因此,我們現在有點像通過對基本架構的已知的、固定的發佈規範擴展來穩定生態系統。這有助於推動整個生態系統向前發展,因為現在他們可以設計一些被RISC-V社群普遍接受的東西。

BRIAN SANTO: Right. We were talking earlier about a roadmap for RISC-V. As you get more extensions, you’re able to target specific platforms and specific implementations of those platforms. Is that right?

BRIAN SANTO:是的。我們之前討論過RISC-V的發展藍圖,隨著獲得更多擴展,業界可以針對特定平台和這些平台的特定實現。對吧?

KEVIN KREWELL: That is correct. You did listen well. The two key initiatives here are profiles and platforms. One of the sort of balancing acts in RISC-V is it’s an open ISA. So you can extend it, you can add new instructions, there are multiple ways of implementing vectors that they finally came to a compromise solution on. But there were some companies that started doing vectors before the standard was set. So they have their own specific implementation of vector.

KEVIN KREWELL:沒錯。你聽得很認真,這裡的兩個關鍵舉措是描述檔和平台。RISC-V的一種平衡行為是開放的ISA,所以你可以擴展它,你可以添加新的指令,有多種實現向量的方法,他們最終找到了一個折衷的解決方案。但是有些公司在標準制定之前就開始做向量,所以他們有自己特定的向量實現。

So there’s a concern that if things got too out of hand, that the ecosystem would fracture, going in a thousand different directions. To help prevent that, what the RISC-V International Foundation came up with is a series of profiles, and the profiles would then specify for a particular market segment, These are the recommended extensions that should be in a core, and try to keep the profile somewhat standardized so that there are just a few paths in the forest, not a thousand paths in the forest. And to keep the amount of permutations from becoming unmanageable for software developers.

所以有一種憂慮是,如果事情變得太失控,生態系統就會破裂,朝著一千個不同的方向發展。為了幫助避免這種情況,RISC-V國際基金會提出了一系列設定檔,然後設定檔將指明特定的細分市場,這些是應該在核心中的推薦擴展,並儘量保持描述檔的標準化,因此森林中只有幾條路徑,而不是上千條路徑。這也能防止軟體開發人員無法管理排列組合的數量。

So if we start focusing on different profiles for things like deeply embedded or servers, or application processors, you can sort of figure out what’s the best implementation, what extensions are needed for those segments, and focus on making that a little more standardized and have a profile for that.

因此,如果我們開始關注諸如深度嵌入式或伺服器或應用處理器之類的不同描述檔,您可以確定什麼是最佳實現,這些部分需要哪些擴展,並專注於使其更加標準化並具有個人色彩。

In addition, the ultimate goal is to have almost kind of like software that could boot on any platform. And that’s where the platforms come in. So there could be a standardized implementation of the hardware around the RISC-V CPU core. So that software could be more turnkey, and you could just drop in and run. And that’s similar to what has happened in the Arm ecosystem, where they standardized their server platform so that the software could just run on the Arm cores without having to have each individual implementation of Arm be different and have to be tuned for that specific implementation.

此外,最終目標是擁有幾乎可以在任何平台上啟動的類似軟體。這就是平台的用武之地。因此,圍繞RISC-V CPU核心可能會有硬體的標準化實現。因此,該軟體可以更加完整,你可以直接使用並運行。這與Arm生態系統中發生的情況類似,在那裡他們標準化了他們的伺服器平台,這樣軟體就可以在Arm核心上運行,而不必讓 Arm 的每個單獨實現都不同,並且必須針對該特定實現進行調整。

So some level of standardization is occurring in the RISC-V community. And that will keep things from going off the rails in the future.

所以RISC-V 社群正在發生某種程度的標準化,這可以防止事情的發展在未來脫離軌道。

BRIAN SANTO: Yeah, so if they’re going to try to compete with Arm, at least in some instances, it behooves them to follow the model it sounds like. They looked at Arm’s success and learned their lesson.

BRIAN SANTO:對啊…所以如果他們要嘗試與 Arm 競爭,至少在某些情況下,他們應該遵循一些模式。他們看見 Arm 的成功並吸取了教訓。

KEVIN KREWELL: Yeah, basically pretty much. I would say they’re looking at how Arm succeeded in building a stronger ecosystem. And they’re trying to emulate that to some extent. There’s still room for innovation, and there’s room for unique implementations. It’s just that in the broadest sense, if you want the ecosystem to thrive and you want enough development tools and standardized tools, and maybe someday even an Android port for RISC-V, there needs to be some level of standardization among all the different iterations that you can make.

KEVIN KREWELL:是的,基本上差不多。我會說他們正在研究Arm如何成功建立一個更強大的生態系統,他們正試圖在某種程度上效仿這一點。仍有創新空間,也有獨特實施的空間。只是從最廣泛的意義上說,如果你想要生態系統蓬勃發展,並且想要足夠的開發工具和標準化工具,甚至有一天是Android的RISC-V埠,那麼所有你可以做的不同演進之間需要有一定程度的標準化。

BRIAN SANTO: Let’s go to the conceptual thing we were gonna hold off on. We had been talking in Preview Conversation 1.0 about who needs RISC-V and why are there so many really brilliant developers moving into this? And I thought you had a really interesting observation. The success of Arm has apparently given everyone permission to do something other than another x86 clone.

BRIAN SANTO:讓我們來談談我們要推遲的概念性問題。我們在Preview Conversation 1.0中一直在討論誰需要RISC-V,為什麼有這麼多真正優秀的開發人員進入這個領域?我認為你有一個非常有趣的觀察。Arm 的成功顯然讓每個人都可以做一些不同於另一個複製 x86的事情。

KEVIN KREWELL: My specific point is that two of the startups — Esperanto and Ventana — in the RISC-V are founded by x86 CPU designers. Dave Ditzel did TransMeta, and Greg Favor worked on the K6 at AMD and then later at another startup. But you don’t hear about new x86 startups anymore. In fact, Intel just bought really what was the third x86 vendor Via, brought the Centaur team into Intel. So there really isn’t a third x86 vendor anymore. There’s just the two — AMD and Intel.

KEVIN KREWELL:我的具體觀點是,RISC-V領域的兩家新創公司——Esperanto 和 Ventana——是由x86 CPU設計師創立的。Dave Ditzel 負責 TransMeta,Greg Favor在AMD負責K6,後來又在另一家新創公司工作。但是您再也聽不到有關新的x86新創公司的消息了。事實上,Intel 才剛買下的是真正第三家x86供應商 Via,將Centaur團隊帶入 Intel。所以真的沒有第三家 x86 供應商了,只有兩家——AMD和Intel。

And nobody else is going after that market anymore because they don’t need x86. You can do stuff with Arm. And now Arm realizes that. And also I think Nvidia realizes it as well. That’s why Nvidia wants to buy Arm. That really is the alternative architecture now. But not everybody wants to spend millions of dollars to get an architecture license when the RISC-V ISA is robust and available basically for free. You can roll your own core.

沒有其他人會再追隨這個市場,因為他們不需要x86。你可以用 Arm 做一些事情。現在 Arm 意識到了這一點。而且我認為Nvidia也意識到了這一點。這就是Nvidia想要收購 Arm 的原因。這確實是現在的替代架構。但是,當 RISC-V ISA 功能強大且基本上免費提供時,並不是每個人都願意花費數百萬美元來獲得架構許可,你可以搞自己的核心。

Right now the ecosystem is not there yet at RISC-V. I think the big focus in 2022 is going to be building up the software ecosystem. Now that they’ve wrapped up all the specifications, they can start rolling out more software to get the software community going. But Arm is already there. Arm’s got Graviton cores, processors at AWS and there’s just tons of activity right now in rolling your own Arm cores, including Qualcomm and obviously, Apple is one of the leading cases for roll your own CPUs using Arm licenses.

目前,RISC-V 還沒有生態系統,我認為 2022 年的重點將是建立軟體生態系統。現在他們已經完成了所有的規範,他們可以開始推出更多的軟體來推動軟體社群的發展。但是Arm已經在那裡了。Arm在 AWS上擁有Graviton核心和處理器,並且現在有大量活動在推動自己的Arm核心,包括Qualcomm,顯然,Apple 是使用 Arm 授權搞自家CPU的領先案例之一。

What Apple is doing is eliminating x86 and replacing it with Arm. And I think that’s what people recognize. If you want to roll your own core, you can go to Arm; you don’t have to struggle with trying to roll a new x86 processor when it’s a very difficult instruction set to emulate. There all kinds of quirks, whereas Arm is more straightforward, and you can go to Arm and say, Can I have a core please? And they’ll say, Here you are! And you get started.

Apple 正在做的是消滅x86 並用 Arm 取而代之,我認為這就是人們所認識到的。如果你想推動自己的核心,你可以去Arm;如果那是一個非常難以硬體模擬的指令集,你不用為了嘗試開發新的 x86 處理器痛苦掙扎。有各種花樣可以搞,其中Arm更簡單明瞭,你能直接去找Arm說,給我一個核心好嗎?他們會回答,給你!然後你就可以開始了。

And so it’s a lot easier. You don’t have to fight with Intel and AMD about patents and all that.

這要容易得多,你不必與AMD和 AMD為了專利等問題展開鬥爭。

BRIAN SANTO: When the guys who are interested in RISC-V show up, they can say, See? Arm did it. It’s okay. You can go to an alternative architecture. If we’re a cheaper alternative, and we’ve developed our ecosystem, hooray for you. Right?

BRIAN SANTO:當對RISC-V感興趣的人出現時,他們可以說,看吧…Arm已經做到了,這是沒問題的,你可以改用一個替代架構;我們不僅是更便宜的,並且我們已經發展出了我們的生態系統…對你來說很棒吧?

KEVIN KREWELL: Right. So RISC-V now becomes that new disruptor. Arm disrupted x86, x86 disrupted the original RISC vendors like MIPS for servers, and alpha and PA-RISC and all those early RISC, Spark even.

KEVIN KREWELL:對。所以RISC-V現在成為了那個新的顛覆者。Arm顛覆了 x86,x86顛覆了最初的RISC供應商,例如伺服器的MIPS、alpha和PA-RISC以及所有早期的RISC,甚至是Spark。

So x86 disrupted them coming in from the bottom. Arm is disrupting x86 coming in from the bottom. Now RISC-V wants to do the same thing coming in from underneath and using economies to disrupt Arm as well. So it’s the cycle of life here.

所以x86從底層顛覆了它們,Arm也正從底層擾亂x86。現在RISC-V想要做同樣的事情,從底層進來,並利用經濟性來顛覆Arm。所以這裡是生命的輪迴。

BRIAN SANTO: The phrase I like is: Nibbled to death by ducks.

BRIAN SANTO:我很喜歡的一個句子是:被小鴨子啃死。

KEVIN KREWELL: Okay, you can use that phrase.

KEVIN KREWELL:不錯,你可以用這個句子。

BRIAN SANTO: Okay. Maybe that’s just me. Cool. All right. So the next question is, this last RISC event, did it give people viewing RISC-V encouragement that the ecosystem is cohering and is on its way?

BRIAN SANTO:好吧,也許這只是我個人這麼想。那麼下一個問題是,最近的這一場RISC-V活動是否鼓舞了關注RISC-V的人們,讓他們看見生態系統正在凝聚並有所進展?

KEVIN KREWELL: You know, this is a judgment call. And I would say yes. It was a hybrid event, so it’s kind of hard to tell that from a crowd point of view, because it was still a light crowd at the actual event itself. But there was a light crowd at DAC and a light crowd at Semicon West. So I couldn’t necessarily get a full feel for the just the crowd excitement. A lot of people were watching online.

KEVIN KREWELL:你知道,這是一個主觀判斷,而我會說是的。這是一場虛實整合的活動,所以從人群的角度很難判斷,因為目前實體活動的現場人數仍然不多,像是DAC和Semicon West的人潮都不多,我不一定能完全感受到人群中的興奮,很多人都是在線上參與。

But from my understanding of how this is developing, I think the work they’re doing on profiles and platforms is exactly what they needed to do to rein in some of the chaos. And this focus now on performance. There really is no limit on where RISC-V can go. They’ve got the instructions they need for vectors. Security is actually one of the areas they’re spending a lot of time on right now. I think everything that they’re doing in their community, it can make it a very robust architecture as it develops.

但根據我對發展情況的理解,我認為他們在描述檔和平台上所做的工作,正是他們為控制某些混亂所需要做的。而現在的重點是性能。RISC-V的發展方向確實沒有限制。他們已經獲得了向量所需的指令,安全性實際上是他們現在花費大量時間的領域之一。我認為他們在社群中所做的一切都可以使其成為一個非常強大的架構。

You’ve got a lot of smart people involved in the architecture. As one of the SiFive guys said, there are no limits to what you can do with RISC-V. And I think they’re pretty much correct. There really is no limit.

有很多聰明的人參與該架構。正如其中一位SiFive 的人所說,能用RISC-V做的事情沒有極限,我認為他們是非常正確的,真的沒有極限。

Performance comes somewhat from the instructions, but a lot of performance comes from micro architecture. And you can apply micro architecture to any instruction set. So in a sense, ISA becomes a little bit irrelevant to performance. It’s not completely because there are some things that the ISA does help with. But I see no practical limits to what you can do with RISC-V. And considering people want an alternative to both x86 and Arm, it’s the only real alternative today.

性能多少來自指令,但很多性能來自微架構。你可以將微架構應用於任何指令集,因此從某種意義上說,ISA變得與性能無關。這並不完全是因為ISA確實對某些事情有所幫助,但我認為可以用 RISC-V做的事情沒有實際限制。考量到人們想要x86和Arm的替代品,它是如今唯一真正的替代品。

And therefore, you’ve got smart people investing money, hundreds of millions of dollars being invested into the ecosystem when you look at all the startups. I don’t think anybody’s really aggregated all the startups involved in RISC-V, but they’re doing everything from chiplets at Ventana to giant AI inference chips at Esperanto. You’ve got small cores, you got big cores, SiFive is putting out these high performance cores. I think you’ve got a very diverse ecosystem now.

因此你看所有那些新創公司,都有聰明人投資;有數億美元被投資到生態系統中。我認為沒有人真正彙整那些參與RISC-V的所有新創公司,他們正在做各種東西,從 Ventana 的小晶片到Esperanto開發的大型 AI 推理晶片,有小核心,也有大核心。SiFive 正在推出這些高性能核心,我認為現在擁有一個非常多樣化的生態系統。

One other thing I would point out is, it is an open ISA, not an open source. So you still have to either buy cores from one of the vendors, but you have multiple vendors to buy it from. There’s Andes, there’s a bunch of small guys, and SiFive is the big guy. But when you license Arm, there’s only one place you can go to get Arm cores: Arm. Or you roll your own. And you have to get an architectural license to roll your own.

我要指出的另一件事是,它是一個開放性ISA,並非開放源碼;所以你還是得向某一家供應商購買核心,但你有多家供應商可以選擇。有晶心等等一些小公司,SiFive則算是一家大公司。不過你若是要取得Arm授權,只有唯一的一個地方能取得Arm核心:Arm。或者你也可以自己搞,但你還是得取得架構授權才能做你自己的東西。

So that somewhat limits who can roll their own. It’s got to be the high-end guys. With RISC-V, you can roll your own, you don’t have to pay for a license, or you can go to multiple vendors to get IP cores. So you’ve got a lot more choice.

所以這在某種程度上限制了誰可以做自己的東西,一定還是高階領域的廠商。而使用RISC-V,你可以自己做自己的,不需要支付授權費用;或者你可以從多家供應取得IP核心,你有更多的選擇。

BRIAN SANTO: My understanding is that a lot of the companies that are providing RISC-V cores are providing CPU cores. GPU can be important. My understanding is that there is no one providing a RISC-V GPU.

BRIAN SANTO:我的理解是,許多提供RISC-V核心的公司也都在提供CPU核心。GPU可能很重要,但我的理解是沒有人提供RISC-V GPU。

KEVIN KREWELL: There have been some people fooling around with it. I’ve seen some GPU cores based off of lots of small vector CPUs put together. I haven’t seen one mainstream yet, or break the kind of obscure.

KEVIN KREWELL:有一些人在搞這個。我有看到一些以小型向量CPU為基礎的GPU核心。我還沒有看到任何一個主流方案,或者是浮上檯面。

BRIAN SANTO: Is that a big deal to have one?

BRIAN SANTO:如果有了一個會是大事嗎?

KEVIN KREWELL: It could be. I think there’s a lot of opportunity there. I mean, you still have in terms of IP cores, there’s only two main IP licensees of GPUs. And that’s Arm and Imagination. There are some small vendors actually still there. But yeah, it could be. I think there’s an opportunity there. But so far, nobody’s really dug into it. Right now they’re focused on the CPUs. Actually, I think there could be opportunities for open source, or open instruction sets for AI acceleration as well.

KEVIN KREWELL:可能會是,我認為該領域有很多機會。我的意思是,在 IP核心方面,目前仍然只有兩家主要的GPU IP 授權業者,就是Arm和Imagination。有一些小供應商實際上仍然存在,但,是的,那會是件大事,我認為那裡有機會。但到目前為止,沒有人真正深入該領域。現在他們專注於 CPU。實際上,我認為可能有開源的機會,或者也有用於 AI 加速的開放指令集。

Here’s one of the underlying assumptions about the RISC-V and the founders of RISC-V: They think that one of the advantages of RISC-V is that no matter what you’re trying to do — CPU, GPU, even DSP or AI accelerators — you can start with an RISC-V ISA core, and then build on that. So the vision of RISC-V is that you can build an SOC, eventually, that every aspect of SOC starts with a RISC-V core and extends on that. So that the software tools across the entire System On Chip, the SOC, is based on one instruction set.

這是關於 RISC-V 和 RISC-V 創始者的基本假設之一:他們認為RISC-V的優勢之一是無論你嘗試做什麼——CPU、GPU,甚至DSP或AI加速器——都可以從RISC-V ISA核心開始,然後在此基礎上進行構建。所以RISC-V的願景是你可以構建一個 SoC,最終,SoC 的每個方面都從一個 RISC-V 核心開始並在此基礎上進行擴展。這樣整個單晶片系統的軟體工具都能以單一指令集為基礎。

BRIAN SANTO: Wow. It would be quite an advantage.

BRIAN SANTO:哇…那會是一個很大的優勢。

KEVIN KREWELL: Yeah. So that would simplify a lot of software development. Now, each functionality like a DSP functionality or graphics functionality would need specific extensions to perform that task more efficiently. But you could still start with maybe a RISC-V as the core architecture and extend off that. That’s their vision.

KEVIN KREWELL:是的。那將簡化很多軟體發展。現在,像 DSP功能或繪圖功能這樣的每個功能性需要特定的擴展才能更有效地執行該任務。但是你仍然可以從RISC-V作為核心架構開始,然後擴展它。那是他們的願景。

BRIAN SANTO: Cool, cool. RISC-V has a political component, or one that was imposed on it. In the grand geopolitical way of looking at things, China’s been isolated from certain of the most advanced Western technologies. But it doesn’t look like their access to RISC-V technologies is cut off, and a lot of people were expecting that China would embrace RISC-V. Is there evidence that that is, in fact, the case that they’re doing that?

BRIAN SANTO:酷。不過RISC-V具有政治成分,或是被強加於其上的。從宏觀的地緣政治來看待事情,中國與某些最先進的西方技術被隔離開來,但他們對RISC-V技術的取得看來並未被切斷,很多人都期待中國會擁抱RISC-V。是否有證據表明,事實上,他們正在這樣做?

KEVIN KREWELL: There’s definitely a lot of enthusiasm for RISC-V in China. Alibaba was a premier sponsor involved in the Summit. In addition, when I asked a specific question to the board, about a third of RISC-V activities is in APAC, which they wouldn’t be more specific China versus say Japan and other countries, you know, India, whatever. And about a third is North America. And about a third is EMEA, Europe and Africa. So it’s definitely an appeal.

KEVIN KREWELL:在中國肯定有很多對RISC-V的熱情。阿里巴巴就是該廠峰會的主要贊助商之一。此外,當我向主辦單位提出一個具體問題時,大約有三分之一的RISC-V活動是在亞太地區,當然他們不會更具體的比較中國與日本,或是印度等其他國家。總之大約三分之一是北美,大約三分之一是EMEA、歐洲和非洲。所以這絕對是一個訴求。

The RISC-V International is based in Switzerland. So it’s neutral.

RISC-V International 總部位於瑞士,所以該組織是中立的

BRIAN SANTO: [LAUGHS] Do you think they were trying to convey something?

BRIAN SANTO:[笑]你認為他們是想傳達什麼嗎?

KEVIN KREWELL: I think there was a message being sent there.

KEVIN KREWELL:我想這已經透露了一個訊息。

BRIAN SANTO: Yeah. There’s definitely an implication.

BRIAN SANTO:是的。肯定是有寓意的。

KEVIN KREWELL: And also, the implementation of a core by, say, SiFive, might still fall under restrictions of technology transfers. But the instruction set itself is open, and there’s no way to limit anybody from implementing that instruction set.

KEVIN KREWELL:而且,例如 SiFive 的核心實現可能仍會受到技術轉移的限制,但指令集本身是開放的,沒有辦法限制任何人使用該指令集。

BRIAN SANTO: Interesting. Anything else about the ecosystem, the trade implications, or any specific RISC-V developments that we haven’t touched on that are worth noting?

BRIAN SANTO:這很有趣。關於生態系統、貿易影響或我們尚未涉及的任何特定RISC-V發展,還有什麼值得注意的嗎?

KEVIN KREWELL: I think we covered a lot of territory, actually. This was a big summit for them. So they had a lot of announcements, and I haven’t been able to dig into the specifications yet that they’ve ratified. But expect the 2022 year coming forward, they’re gonna focus on getting the software tools and developing the software ecosystem around what they’ve now ratified. So that’s their big thrust for next year. And I think that’s where their biggest gap is right now they’ve got to close.

KEVIN KREWELL:我想我們已經談到了很多。這是一場對他們來說很重大的峰會,所以他們有很多宣佈,我還沒有能夠深入研究他們已經批准的規格,但預計到 2022 年,他們將專注於完成軟體工具並圍繞他們現在批准的內容開發軟體生態系統。這會是他們明年的主要目標。我認為這也是他們現在必須趕上的最大差距。

You got a number of different performance cores and different levels. You’ve got lots of different types of design activity going on in the ecosystem, some very bright people involved. So I think it’s very robust at this point.

你有許多不同性能和不同等級的核心,在生態系中有許多不同類型的設計活動在進行,其中有一些非常聰明的人參與其中。所以我認為它在現階段非常強韌。

BRIAN SANTO: All right. Kevin, thank you very much.

BRIAN SANTO:好的。非常謝謝Kevin。

KEVIN KREWELL: No problem.

KEVIN KREWELL:不客氣。

BRIAN SANTO: That was Kevin Krewell, principal analyst at Tirias Research. Krewell was just back from the RISC-V Summit in San Francisco, held last week.

BRIAN SANTO:以上是Tirias Research 首席分析師 Kevin Krewell的分享。Krewell才剛從在舊金山市舉行的RISC-V Summit回來。

(完整內容請參考原文連結)

 

發表評論