NUVIA To Make Serious Play For New CPUs In The Datacenter, Hires Linux/OSS Veteran

[starshipeleven]

Oh yes, let turn ARMs to what x86 is.

That's not what I was talking about, so wtf?

This said it seems it probably going to be focused on server things and ppl managing large deployments aren't exactly happy about that state of things, you can see how many alternate boot solutions pop here and there

If there is something that server people don't want, is yet another hardware-specific bullshit unsafe implementation for network boot or system boot.

[coder]

As concrete example: Nvidia is blob-only something, does not runs on anything that isn't x86 (yes, China wanted something to pair with their MIPS designs for example - and Nvidia had nothing to offer at this point; these can't use x86 either I guess).

Nvidia is not x86-only - they also support POWER, and most recently ARM.

I thnik there is some headroom in high-performance computing area for smart guys who can get it right: ... More or less stable (and ideally without dropping support of HW at speeds AMD/Nvidia do - enterprises wouldn't buy new CPUs, GPUs and so on every couple of years unlike stupid gamers).

I'm sure AMD and Nvidia's support window is more than long enough to fit the typical deployment life of their datacenter hardware. These guys rotate out HW faster than you think, because they know what they're paying for power & cooling and when it's more cost-effective to swap-in a more efficient generation.

Also, these things actually do wear out with use, of which they get a lot. So, the failure rate of older GPUs will become quite substantial, if they hang onto them long enough. Unfortunately, I'm not finding good data on that, so I can't say whether it's a limiting factor, in practice.

[coder]

we need a Open-source GPU based on Open-POWER ISA

Why not use an ISA that's fit for purpose? POWER is not going to result in a very good GPU.

If you take a general-purpose ISA and try to build a GPU around it, you get a debacle like Larrabee or Xeon Phi. GPU ISAs have some important differences that enable them to achieve far higher levels of throughput and power-efficiency, at the expense of flexibility and single-thread performance.

[torsionbar28]

Transmeta Crusoe 2.0

[cybertraveler]

From the article:

... I decided not to let my Dreams be only Dreams. Together, we are going to change the world! ...

Beautiful mindset.

[pkese]

ARM? hell no....
we need a Open-source GPU based on Open-POWER ISA without closed source firmware.

I.e. take a fat bloated ISA and pretend it is light and efficient and shove it into a GPU?
Let me enlighten you - it's called lipstick on a pig, it had been done before and didn't succeed. It was called Intel Larrabee.
Open-POWER ISA is approximately as bloated as x86 (or at least way above all other RISC ISAs including Sparc or ARM).
You certainly don't want to put *that* on a GPU, unless your intent was to fail anyway.

RISC-V also would be nice.

Possibly...
Definitelty if you need a simple ISA that will let you put thousands of cores on a single piece of silicon,
however there may be reasons why contemporary GPUs are closer to VLIW than RISC architectures.

[coder]

From the article:

... I decided not to let my Dreams be only Dreams. Together, we are going to change the world! ...

Beautiful mindset.

Unless your dreams are other people's nightmares!
; )

[coder]

no one will try to build POWER-ISA compatible GPU cores. you use it as a start to build a gpu isa derived from the power isa

in the end you have incompatible ISA but a light and efficient GPU-ISA (but derived from power ISA)

Well, if you're allowing it to completely break compatibility, then probably doesn't much matter where you start.

The point is that what you really want is an open GPU. Fair enough. I suggest just leaving it at that, without getting overly concerned about what the specific ISA is inspired by, or most resembles. What really matters is that the ISA is suited to the task at hand.

[cybertraveler]

Unless your dreams are other people's nightmares!
; )

True!

[Space Heater]

Unfortunately so. The ARM ecosystem is a non-coherent mess.
Which is fine for the odd embedded system, but non-functional in the COTS buy/replace field.
I do welcome more players to the SoC arena, but in reality,
this is nothing more than a slapping-together-yet-another-soc type business.

Edit: I do hope I'm wrong and I do hope this effort will bring some sanity into the ARM ecosystem without loosing much of its flexibility.

Fortunately you're completely wrong, ARM servers already have standards (by the way Jon Masters worked on ARM server standards while at Red Hat).

SBSA (Server Base System Architecture)

• https://github.com/ARM-software/sbsa-acs
• http://infocenter.arm.com/help/topic..._DEN_0029B.pdf
• https://lwn.net/Articles/584123/

SBBR (Server Base Boot Requirements)

• https://github.com/ARM-software/arm-...ee/master/sbbr
• http://infocenter.arm.com/help/topic..._DEN_0044C.pdf
• https://community.arm.com/iot/b/blog...ation-released