There's Still No Sign Of AMD's Low-Cost ARM Development Boards

[torsionbar28]

You know shit of devboards right?
Pricing of 1.5k or even more is normal.

We aren't talking of crappy trash SoCs like tablet stuff.

People confusing hobbyist hardware with commercial dev boards. SMH. $1.5k is cheap even for a full featured dev board.

[Duve]

x2, I'd be willing to bet it's "all hands on deck" at AMD for the Zen product line. They absolutely have to get Zen right, and ASAP. Their survival depends on it.

ARM is indeed promising, but it's a niche market still. You won't see any significant ARM in the datacenter until you get enterprise OS support. That means RHEL and SLES. I'm sure that's in the works, but as of today, it's nonexistent outside of developer previews. Probably because the AArch64 hardware selection is so slim, is anyone selling an ARM commodity server besides HP right now?

If that is the case then their might be a change in optics coming. There seems to be quite a bit going on with RISC-V. Power is a kind of chip that has been slowly building. I would not be shocked if something was to change with AMD withing 5-10 year time with their ARM license.
That said, with CentOS behind them now, I would not be surprised if the community spends some time on a AArch64 port (which I think the AltArch SIG is doing at this point) that gets Red Hat's notice latter.

[ldesnogu]

AFAIK the A1100 was to make sure there there were enough server-class parts out there to finish the ARM server SW ecosystem before K12 shipped, and to give us a seat at the table in case the ARM server market took off as quickly as many were forecasting. We had to do something similar to get SW ready for AMD64, but there we had to do a lot of the early dev work with ISA simulators.

The situation is the same for ARM as it was for AMD64 back then.

AArch64 first ports were also done wih ISA simulators. Tools have been developed with an internal version of QEMU and kernel porting was done with another simulator.

As far as porting goes, fast simulators can do a very good job. OTOH performance tuning requires real hardware. And the issue here is that the A1100 already looks dated.

[Zola]

Yeah, ARM is going after the supercomputing market. Everybody else is already there. That's called "catchup" and we have a graveyard of companies that couldn't even compete with Intel in the low-power microserver market where ARM is supposed to be magically good*, much less in HPC where Intel is selling Knight's Landing systems today instead of talking about an instruction set that doesn't even have a hardware prototype.

*ARM really isn't that special when a 65 watt Xeon-D with 16 cores is practically unbeatable in a energy efficiency metric at this point.

Really? How about doing some reading first sport?

http://www.anandtech.com/show/10353/investigating-cavium-thunderx-48-arm-cores

Now this is with semi costume design based on wider in order cores & they didn't did a stellar job either. It's also based on old 28nm manufacturing process. Still power management didn't work properly for this testing so I should certainly want to see an update or a new article about it whit Ubuntu LTS & resolved power management problems.
How ever you will notice how it's in the worst case on pair (clock per click) with Intel Xeon D's & in some cases significantly more faster regarding multithraded tests, it also costs significantly less money & in some cases even catches up with bigger Xeon brothers. So you see first commercial products are hire (real ones & available) & software is actually catching up good.
Now imagine what would be if A73 OoO cores ware used instead that are 30% faster, cheaper & smaller on silicon (so they would cut price & power consumption). Still so certain about those Xeon D's & how they are a gift from heaven by any metric?

Michal even they are much cheaper sku's then this one 48 core that costs 800$ complying board's are still very pricy so I don't think you will be getting one. Never the less this is just one of many to come so Things They Are Changelog (as the good old song says).

[Zola]

Guess AMD is again betting on a wrong cow & they Chakra is on the wrong side of the real Zen.

[bridgman]

Which do you think is the right cow ? Short term ARM server focus or short term x86 followed by longer term ARM ?

[Nille_kungen]

Which do you think is the right cow ? Short term ARM server focus or short term x86 followed by longer term ARM ?

Is it really black and white like that?
According to previous released information K12 would integrate X86 and ARM cores in the same chip (ambidextrous computing) and let us say that it is then more likely that zen cores would be used instead of previous gen cores (even if the plan was to release previous gen cores it wouldn't make sense anymore to not wait for zen).
If that would be true then K12 needs to wait for zen before it's released, and i would understand if arm isn't the highest priority for AMD.

[Nille_kungen]

AMD's Ambidextrous Computing

[starshipeleven]

This board is nothing special - quad core A57, and it doesn't even come with ram. It is NOT a "special" cpu, it is about the same stuff you see in every decent tablet or phone. That price is laughable, even at 300$. I know enough to know how much it is worth, now if you are willing to pay 1.5k for it - my condolences. Quadruple the cores, and I will get a few at 300$, then it will be a good deal.

dev-fucking-board.

devboards are SMALL production runs, thus their prices are much higher than the price of a consumer product.

Also devboards are sold traditionally to companies so they do keep high prices because they can.

Raspberry isn't a traditional devboard, it is a consumer product sold in high numbers, same for most crappy clones.

[starshipeleven]

Is it really black and white like that?
According to previous released information K12 would integrate X86 and ARM cores in the same chip (ambidextrous computing)

No. Ambidexterous computing is their x86 and ARM SoCs/processors are pin-compatible so you can use boards made for x86 (because OEM are a bunch of cats) to install an ARM processor.
This works ONLY if the UEFI firmware is not arch-specific or runs on common components (like the security cores, that would finally be useful to do something else beside helping NSA), if someone expects the OEMs to actually make ARM UEFI firmwares this comes crashing down.