AMD Announces Automotive-Grade Ryzen Embedded V2000A Series

[cypx]

it looks like ARM is going to increasingly appear in servers and data centers, and in 2024

People keep talking about that since 10 or 20 years...
Seems to me that ARM has already made its way in servers and data center for all relevant usage (routers, specialized compute...) but x86 continue to be the most relevant choices for general usage.

[skeevy420]

Strange they're Zen2 based.

RHEL 8 has support until 2029 and uses GCC 9 that supports Zen 2 and up. Zen 2 is the greatest target AMD can use for LTS EL.

[milkylainen]

Strange they're Zen2 based.

Well. v2000 is a old product. This is just some automotive qualification of a product that has existed for two or so years already?

[skeevy420]

Agreed. I'm sure this was mostly opportunity cost that made them side with Zen2. It could be that at the time that this embedded system was designed it was the best choice that was already proven in consoles. Now it's many years later and there are better solutions. The embedded V1000 took extremely long to be released. It doesn't look like it's a high priority for AMD.



I know AMD tried ARMv-A between 2014 and 2017 https://en.wikipedia.org/wiki/AMD_K12

AMD still has an ARM license but AFAIK the only major use case is the PSP chip. Based on previous discussions, I wonder if AMD will drop legacy x86 from AMD64 before they try the ARM route again. Major speculation either way.

The PS4 hypervisor is ARM, too. PS4 uses both ARM and x86. No reason a car couldn't utilize ARM for some low level stuff and fire up the x86 cores when the car is actually in use and needs to be able to play a video game.

I remember when playing a video game in a car was something out of Pimp My Ride and now it's just what cars are expected to do. There's an analogy between cars and Wayland compositors and how new or niche features become adopted over time to become standard.

[schmidtbag]

Not to the auto industry who dictated they want cheaper smaller die size and cooler to run.

That makes sense, but wouldn't Zen4c potentially have fewer transistors per core while having instructions that the automotive industry may take advantage of? I know it has a lot more per CCD but it also has more cores per CCD.

RHEL 8 has support until 2029 and uses GCC 9 that supports Zen 2 and up. Zen 2 is the greatest target AMD can use for LTS EL.

Does that mean newer architectures have a shorter LTS, or that the automotive industry is interested in using older versions of GCC? Or both?

[skeevy420]

That makes sense, but wouldn't Zen4c potentially have fewer transistors per core while having instructions that the automotive industry may take advantage of? I know it has a lot more per CCD but it also has more cores per CCD.


Does that mean newer architectures have a shorter LTS, or that the automotive industry is interested in using older versions of GCC? Or both?

I'm assuming it's Zen 2 since that seems to be the biggest overlap between what the LTS EL distributions support out of the box and the what AMD can offer.

It's not like Zen 2 sucks. Day to day, not doing super-intensive tasks like games or compiling software, I don't notice any difference between my previous Zen 2 4650G APU with 32GB DDR4 3800 or my current Zen 4 7800X3D with 64GB DDR5 6400, both CPUs paired with a 6700 XT GPU.

[pWe00Iri3e7Z9lHOX2Qx]

People keep talking about that since 10 or 20 years...
Seems to me that ARM has already made its way in servers and data center for all relevant usage (routers, specialized compute...) but x86 continue to be the most relevant choices for general usage.

Your view is several years out of date. While it seemed like ARM was perpetually "almost good enough" in the datacenter up through the Cavium Thunder days, ARM's Neoverse platform definitely pushed things across the goal line. All major cloud providers have been offering instances based on some Neoverse v1 derived design. And most of them will be offering some in-house tweaked version of Neoverse v2. Total perf / cost on these instances is quite attractive.

[gentoofu]

That makes sense, but wouldn't Zen4c potentially have fewer transistors per core while having instructions that the automotive industry may take advantage of? I know it has a lot more per CCD but it also has more cores per CCD.

Yeah, but price. They don't want to fight for 6nm allocation while 7nm node machines are freely available and can be had for much less. Also, I think Synopsys just released auto-grade design rule guidelines for LPDDR5X/5, PCIe 5.0, and etc. back in September for AMD to be able to use. We might see another 2-3 years for AMD to get certification on functional safety and verification for reliability standards.

[mihau]

you've heard of skylake+++++, now get ready for renoir+++++

[AmericanLocomotive]

Oh look. AMD just released another new Vega part after they already have started discontinuing driver support for it.

https://www.phoronix.com/news/Mesa-24.0-Faster-RADV-Vega