Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

reuters.com

Post reply on HN
Print view View original

221–230 of 243 posts

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#221

Post by inkyoto »

user_7832 wrote: post #207
chasil wrote: post #184

Earlier quoted context omitted.

That is really a loaded question. ARM bigLITTLE exists to address different efficiencies that are exploited in different runtime states. Is big better than LITTLE? Sometimes yes, sometimes no. If there is no definitive answer within the heterogenous cores of a single ARM implementation, then how can it be compared to RISC-V?

Ok, I'll try again. For the same sized single-core chip (x mm squared), running at the same voltage and same clock speed, on the same manufacturing process node, is there a (significant) difference between ARM and Risk-V? (I'm assuming that core size, voltages and frequencies are the biggest differences between the big and LITTLE cores. And yes of course this is a funky theoretical question so a theoretical answer is…

> For the same sized single-core chip (x mm squared), running at the same voltage and same clock speed, on the same manufacturing process node, is there a (significant) difference between ARM and Risk-V?

Very many. And they are at the CPU implementation level.

The instruction decoding buffer depth, the register file size, the implementation of the out of order execution, the TLB depth, the data bus width, separate hardware assist engines etc are amongst many other conscious and decisive design decisions that can differentiate separate designs. Even at the same (citing verbatim) «voltage and same clock speed, on the same manufacturing process node» the performance can vary wildy.

The M1 CPU family is a very wide implementation in nearly every aspect, with a very large L1 cache (both, I and D, caches), reasonably large L2 cache, a (reportedly) 380 register wide register file, 512-bit wide data bus (in the M1 Max) plus extras, hence why it outperforms competing ARM64 designs running at similar voltages and frequencies. The same is true for POWER CPU's (i.e. compare POWER9 vs POWER10 designs) – POWER10 is 2.6x faster than POWER9 despite running at similar frequencies – due to a substantial design overhaul.

Holistic system design matters, not the ISA alone.

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#222

Post by inkyoto »

klelatti wrote: post #138
pm90 wrote: post #129

Earlier quoted context omitted.

Let’s be real. ARM got a chance because it excelled in a very specific use case: low power. Unless RISC-V differentiates itself in a significant way, it won’t replace ARM.

RISC-V seems to me to have two major opportunities. - Where a free ISA allows innovation and experimentation in a way that Arm would not permit. - Where companies can shave cents off mass market products due to lower licensing fees. Will RISC-V replace Arm in your smartphone though - seems like a big ask at the moment.

> RISC-V seems to me to have two major opportunities.

> - Where a free ISA allows innovation and experimentation in a way that Arm would not permit.

I would say that the two major innovations of RISC-V are:

- Composability.

- Extensibility (that also fosters experimentation).

Each ARM version X.Y comes with a license and with a condition: «thou shalt taketh either everything or nothing»; RISC-V has neither a license nor strings attached and is «thou may useth only what thou needeth». Anyone who does not need SIMD or the vector processing can do away with them in the silicon. It saves the die size[0] and costs. This is the composability aspect.

Extensions planned out and/or reserved for future use (e.g. 128-bit extension of RISC-V), and the ability to add custom instructions to experiment with (and possible contribute to the RISC-V forum). This is the extensibility aspect.

Other existing open ISA (MIPS, POWER SuperH J-2? – unsure on the status of the last one) do not offer composability and extensibility capabilities, which leads me to reason that it is the major innovation(s) of RISC-V.

[0] Albeit it is somewhat fraught with the danger of creating a highly fragmented RISC-V CPU market where the software has to accommodate every possible permutation of RISC-V extensions and inject an emulation layer for extensions missing in the die.

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#223

Post by tjoff »

kmeisthax wrote: post #186
tjoff wrote: post #121

Earlier quoted context omitted.

The thing that stands out about ARM is that the user has less control of the hardware. The manufacturers try and succeed in keeping an extremely closed ecosystem. Why would you want that? The Intel+AMD duopoly seems like nirvana compared to the Ampere and M1. In fact, Ampere and M1 are a hint of the most dystopian future of computing I can imagine.

Uh... M1 (at least when Mac-fused) isn't actually locked-down. You do have to jump through a few hoops to boot modified kernels[0], but it's a reasonable level of security for a chip primarily intended to go into laptops. There's an ongoing effort to reverse-engineer the M1 and port Linux to it[1], which has actually gone really well. There's also plenty of ARM hardware out there that really doesn't bother with lockd…

Oh, so I can buy an M1 for a custom board?

Are there several different manufacturers of motherboards for it?

There isn't even an ecosystem to talk about. The fascination about the M1 is quite disturbing. Unless you want to be locked into the apple store it has no value whatsoever. The energy put into running Linux on it is really impressive. But from a broad perspective nothing positive can ever come out of it.

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#224

Post by dralley »

paulmd wrote: post #199
dnautics wrote: post #34

Earlier quoted context omitted.

> And lots of arbitrarily software locking features to extract more money from different market segments. To be fair, AMD does super annoying lock in/binning things too, though not with software locking. For example, there was a Radeon that would have been a crazy awesome server rack mounted GPU for Deep learning 4U blades. You couldn't put it in a server because it was mounted on a chassis that was 4mm too large for…

AMD is also moving towards proprietary interconnects for their add-on cards that provide higher performance links than their competitors are allowed to use. And they won't license them to their competitors of course - it's a "set bonus" if you buy all AMD. It's funny because people were panicking about Intel doing exactly that, and when AMD does it, crickets. They also have moved to locking their CPUs to the motherbo…

> Who knows, and that's the point AMD is going for, they just have to inject enough uncertainty to change consumer behavior.

That's not "the point". The purpose of the feature is legitimate, but you're right that the consequences suck hard for the second hand market.

The thing you're talking about is a security feature where the CPU would be locked to only boot when the firmware is signed by a particular signing key. If you're a high-value target such as a cloud vendor or a government / military, attacks on the firmware are a legitimate threat, and this ensures that the firmware cannot be silently tampered with. This is a big deal to them and that's the reason why AMD implemented this.

But yes, most people don't need or want this. It's entirely optional, it's supposed to be something the customer configures, but Lenovo turned it on by default. Pressure needs to be put on Lenovo to stop doing so.

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#225

Post by tjoff »

floatboth wrote: post #192
tjoff wrote: post #147

Earlier quoted context omitted.

I apologize. I confused Ampere with Amazon's Graviton.

AWS offers Graviton instances in a bare-metal way too, so you can see for yourself what the SoC offers. It's quite clean, there is a Designware PCIe filtering quirk (same as on e.g. the Marvell Armada8k) but other than that it's SBSA-style UEFI+ACPI. Very standard system. In terms of firmware openness, yeah, you don't get the firmware source, but you don't get to touch the machines physically either so you wouldn't h…

>AWS offers Graviton instances in a bare-metal way too, so you can see for yourself what the SoC offers.

Oh, we should be so grateful...

Or perhaps be happy with what we got and not sacrifice everything for some very short term 5% benefits?

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#226

Post by snvzz »

Symmetry wrote: post #146
NortySpock wrote: post #9

Why will ARM succeed where MIPS faded into obscurity? (Well, relative obscurity. I see they are still used in media devices and cheap routers.) https://en.m.wikipedia.org/wiki/MIPS_Technologies

Arm was one of the CISCiest of the RISC processors at a happy medium in instruction complexity that gave them all the simplicity of regular sized instructions while still getting the job done in fewer instructions than a more ideological RISC ISA. Some in depth look at various features here: https://userpages.umbc.edu/~vijay/mashey.on.risc.html Also, getting the job done with fewer instructions meant requiring less R…

>that gave them all the simplicity of regular sized instructions while still getting the job done in fewer instructions than a more ideological RISC ISA.

Thumb2 in 32bit ARM was competitive in code density.

AArch64 code density is abysmal, dramatically worse than RV64GC.

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#227

Post by ant6n »

unfocussed_mike wrote: post #172
Wronnay wrote: post #143

Earlier quoted context omitted.

It's really funny how the market thinks that Intel isn't in the game anymore. They may not be as energy efficient as M1 or ARM chips but they certainly show that they are still a competition and invest a lot of money to improve their performance.

The thing is, the market can be right without knowing why it is right. The market thinks that in some fixed, relatively near point in the future we will look back at now and say "Intel was already hosed, it just didn't know it". It may be that the market is currently only expressing interest in the hopes of others that this particular future will come to pass, but that is the only way it will.

The market used to think AMD was hosed. About 5 years ago, the stock was worth less than 2$. Now its more than a 100$. ¯\_(ツ)_/¯

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#228

Post by klelatti »

tjoff wrote: post #223
kmeisthax wrote: post #186

Earlier quoted context omitted.

Uh... M1 (at least when Mac-fused) isn't actually locked-down. You do have to jump through a few hoops to boot modified kernels[0], but it's a reasonable level of security for a chip primarily intended to go into laptops. There's an ongoing effort to reverse-engineer the M1 and port Linux to it[1], which has actually gone really well. There's also plenty of ARM hardware out there that really doesn't bother with lockd…

Oh, so I can buy an M1 for a custom board? Are there several different manufacturers of motherboards for it? There isn't even an ecosystem to talk about. The fascination about the M1 is quite disturbing. Unless you want to be locked into the apple store it has no value whatsoever. The energy put into running Linux on it is really impressive. But from a broad perspective nothing positive can ever come out of it.

It's a laptop chip. You can run linux and non-app store applications on it. If you don't want to use it - fine - but others are getting a lot of value out of it.

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#229

Post by tjoff »

klelatti wrote: post #228
tjoff wrote: post #223

Earlier quoted context omitted.

Oh, so I can buy an M1 for a custom board? Are there several different manufacturers of motherboards for it? There isn't even an ecosystem to talk about. The fascination about the M1 is quite disturbing. Unless you want to be locked into the apple store it has no value whatsoever. The energy put into running Linux on it is really impressive. But from a broad perspective nothing positive can ever come out of it.

It's a laptop chip. You can run linux and non-app store applications on it. If you don't want to use it - fine - but others are getting a lot of value out of it.

Yes?

And it's still a closed ecosystem.

Re: SoftBank's Sale of Arm to Nvidia Collapses, Arm to IPO

#230

Post by unfocussed_mike »

ant6n wrote: post #227
unfocussed_mike wrote: post #172

Earlier quoted context omitted.

The thing is, the market can be right without knowing why it is right. The market thinks that in some fixed, relatively near point in the future we will look back at now and say "Intel was already hosed, it just didn't know it". It may be that the market is currently only expressing interest in the hopes of others that this particular future will come to pass, but that is the only way it will.

The market used to think AMD was hosed. About 5 years ago, the stock was worth less than 2$. Now its more than a 100$. ¯\_(ツ)_/¯

Right but the key difference is that post-recovery AMD is fabless, because they spun off their fab business.

AMD's stock can jump like this, because its value is based more closely on its designs.

Intel is not taking that approach. One amazing chip design isn't going to massively change their fortunes, and they are heavily sunk in x86, which is not the chip series the market is most interested in.

One meaningful path forward for Intel is to split in two, precisely so that the x86 design business can be liberated by being fabless, and the fab businesses can get on with making unsexy larger-process chips for everyone in America who needs them.

But Intel seem to be resisting this.

Post reply on HN