SiFive Has A New RISC-V Core To Improve Performance By 50%, Outperform Cortex-A78

[cb88]

Wasn't P550 also the name of Imgtec's highest performing MIPS core before they sold MIPs off?

I seem to remember the same thing... except I think it was P5600... and they were making a big deal about it and the CI20 dev board. Also it wasn't the fastest cores they reelased as they released cores about 50% faster 4 years ago.

[brucehoult]

In my Opinion, we should get this news with *Big* Grains of Salt..because SiFive promised the same before, and its other cores were not even on par with a cortex a53..

That's not true. As a CPU running normal instructions e.g. from compiled C code (or Fortran, whatever), the SiFive U74 in the HiFive Unmatched and the BeagleV outperforms both the A53 and A55.

This is something you can verify yourself, with your own code.

Here's a benchmark I wrote for my own purposes and have been using since before I even knew RISC-V existed: https://hoult.org/primes.txt

Some benchmarks such as GeekBench are heavy on programs that need SIMD and/or crypto instructions (e.g. SHA) to perform well. Currently shipping RISC-V chips including SiFive's simply don't have those because extensions containing those instructions have not yet been ratified. However those extensions are right now in their 45 day Public Comment period before ratification, and will be ratified before the end of the year.

All the major RISC-V companies have cores ready to go as soon as they know those extensions aren't going to change in the ratification process. That includes SiFive, Andes, and updated versions of Alibaba's cores (which have already shipped with a draft version of the vector extension, for example).

There is nothing wrong with SiFive's cores with regard to the things they actually implement.

SiFive's demo SoCs e.g. FU540 FU740 suck a bit. But that's not their core business, just as it's not ARM's business. Both companies sell CPU cores to companies that *do* know how to put them in good SoCs: Qualcomm, AllWinner, Samsung etc

[cbxbiker61]

You gotta remember M1 is impressive but not *that* impressive... its just a node ahead of everyone else because Apple pays more than anyone else for fab access. I'd say thier architecture is on par with monolithic x86 CPUs at this point but thats already late to the game architecturally.

I'm not an Apple fan. I dislike their anti-consumer business practices. I don't think that is likely to change since Apple has a following that gulps down their kool-aid without even a thought as to how Apple takes advantage of them. With that said, the M1 chip is *that" amazing.

It's more than a node advantage. It's an architectural advantage coupled with a node advantage. The M1 chip has 10 instruction decoders due to the consistent instruction code size. AMD indicates they have a upper bound of 4 decoders due to the variable length x86 instruction codes.

The M1 in a notebook can outperform AMD/Intel running on battery power even with the AMD/Intel units plugged into AC.

BTW, there are some good youtube videos discussing the virtues of the M1 chip.

[Alexmitter]

I'm not an Apple fan. I dislike their anti-consumer business practices. I don't think that is likely to change since Apple has a following that gulps down their kool-aid without even a thought as to how Apple takes advantage of them. With that said, the M1 chip is *that" amazing.

It's more than a node advantage. It's an architectural advantage coupled with a node advantage. The M1 chip has 10 instruction decoders due to the consistent instruction code size. AMD indicates they have a upper bound of 4 decoders due to the variable length x86 instruction codes.

The M1 in a notebook can outperform AMD/Intel running on battery power even with the AMD/Intel units plugged into AC.

BTW, there are some good youtube videos discussing the virtues of the M1 chip.

And surprisingly it still performs about as we can expect a modern X86 Mobile chip to perform at 5nm.

Its a fine chip at 5nm, nothing less, nothing more.

[Newb]

As powerful as A78! Translated from chinese, it actually means it takes the same amount of electricity.

[coder]

For most users A78 is more than fast enough

Indeed, Amazon's Graviton 2 and Ampere's Altra both use A76-derived cores.

[coder]

You gotta remember M1 is impressive but not *that* impressive... its just a node ahead of everyone else because Apple pays more than anyone else for fab access. I'd say thier architecture is on par with monolithic x86 CPUs at this point but thats already late to the game architecturally.

It's not true. Apple's cores are in a league of their own. They IPC (instructions-per-clock) are far beyond anyone else's. So much so, that they can basically even overcome the clockspeed difference vs. a Ryzen 5950X.

https://www.anandtech.com/show/16252...le-m1-tested/4

[cbxbiker61]

And surprisingly it still performs about as we can expect a modern X86 Mobile chip to perform at 5nm.

Its a fine chip at 5nm, nothing less, nothing more.

What X86 Mobile chip at 5nm? Sounds like speculation to me.

[Developer12]

You gotta remember M1 is impressive but not *that* impressive... its just a node ahead of everyone else because Apple pays more than anyone else for fab access. I'd say thier architecture is on par with monolithic x86 CPUs at this point but thats already late to the game architecturally.

No, the M1 is SIGNIFICANTLY more architecturally advanced, including in some ways x86, than anything else on the market.

I've beaten this horse to death by now but the M1 is an 8-way decoding core. AMD's zen tops out at 4 and intel's best at 5. And that's just the tip of the microarchitectureal iceberg. The MI can keep FAR more instructions in flight, searching for uops to fuse and optimize out. It doesn't just have a nice process, it has an insane IPC (Instructions-Per-Clock) and that doesn't change even if your chip is made on 180nm (go ask LibreSoC about that).

[Developer12]

I'm not an Apple fan. I dislike their anti-consumer business practices. I don't think that is likely to change since Apple has a following that gulps down their kool-aid without even a thought as to how Apple takes advantage of them. With that said, the M1 chip is *that" amazing.

It's more than a node advantage. It's an architectural advantage coupled with a node advantage. The M1 chip has 10 instruction decoders due to the consistent instruction code size. AMD indicates they have a upper bound of 4 decoders due to the variable length x86 instruction codes.

The M1 in a notebook can outperform AMD/Intel running on battery power even with the AMD/Intel units plugged into AC.

BTW, there are some good youtube videos discussing the virtues of the M1 chip.

It's actually only 8 decoders. 10 cores in the latest chips, but we don't know what their uArch is like yet.