|
|
|
|
|
|
by pinewurst
1650 days ago
|
|
|
I think they’re referring to the odd custom-ish embedded parts, e.g. Western Digital usage in mass produced storage devices that can afford novelty. The problem is that there are really no current commodity embedded RISC-V architecture parts - Chinese dime-a-dozen pseudo-Cortex ripoffs don’t count - remotely comparable in performance/power to better current alternatives. Benchmarks of SiFive’s high end are laughable, being grossly sub-RaspberryPi which is hardly ARM top of the line (and doesn’t claim to be such). As SiFive’s P550 hasn’t materialized beyond PR, one can hardly consider it real. Unless some stealth startup appears with a real Zen/NeoVerse analog, I expect this triumph of hype over reality to continue. |
|
|
As is usual in the industry, any serious potential customer will have been able to get the full RTL for the core back in June or July, and evaluate its performance in simulators such as Verilator, or on FPGAs.
Claiming it's somehow not "real" is just stupid. There is no serious possibility that a core that boots Linux and runs SPEC etc in an FPGA won't work the same in a properly designed and manufactured SoC.
Benchmarks of the HiFive Unmatched are exactly what they should be given its A55-like microarchitecture -- a bit better than a Pi 3 and sometimes close to a Pi 4 on normal compiled C code. It doesn't have SIMD or specialised AES and SHA instructions, so of course benchmarks of that kind of thing are worse than a Pi. If that's important to you then of course don't buy one.
RISC-V International last month ratified a number of important extensions including Vector and crypto and those will start appearing in future chips.
There are several documented startups working on "M1 class" RISC-V cores, staffed with highly experience highly credible people such as Jim Keller or people who founded PA Semi and then developed ARM cores at Apple (including the M1).
There's no magic involved in making a top end RISC-V core, just money and good engineering.