Linaro & Co Are Working On An ARM 24-Core Desktop / Developer Box

[grok]

You conveniently ignored the two quotes saying that this is meant for developing ARM on ARM - which means, let me translate it for you, actually doing your development work on the ARM platform, AKA using it as a workstation, and this is why it has a PCI bus, so it can support a desktop graphics card. Don't believe me? Read the article. Follow the links in the quotes. I went to some effort to help you understand this. It would be great if you put as much effort into trying to understand it as you have into (blindly) refuting it.

Well the 16x slot can be used for 4x and 8x boards, or even an 1x one, or for a computing GPU (would you really want to try some CUDA or OpenCL there) or for a display or a GPU with monitors attached if you really want to use it as a desktop.

I think the point is, it's just a computer and you do the computer things you want to do with it? And you're not stuck with a few GB of soldered LPDDR memory. Likely most customers will ssh into, even from a Windows computer, while two or three people will make it their main workstation as it's better than a Richard Stallman's MIPS laptop and cheaper than a POWER8 or POWER9 motherboard.

[schmidtbag]

You conveniently ignored the two quotes saying that this is meant for developing ARM on ARM - which means, let me translate it for you, actually doing your development work on the ARM platform, AKA using it as a workstation, and this is why it has a PCI bus, so it can support a desktop graphics card. Don't believe me? Read the article. Follow the links in the quotes. I went to some effort to help you understand this. It would be great if you put as much effort into trying to understand it as you have into (blindly) refuting it.

I didn't ignore that. I'm saying that whatever big projects you're thinking of likely aren't developed on it, or at least don't cause performance issues. Again, it's best as a test development platform, not a full-blown workstation. But, go a head and get angry at how much you think this is a dumb product - no my problem you see things so black and white.

[Zan Lynx]

wrong again, moore's law doesn't mention process size either

Moore's Law is about the number of transistors. You can either increase the chip size or reduce the size of the transistors. Chip size is limited because the bigger they get, the more likely it is to have serious errors. That leaves size and density.

So yes, Moore's Law is DIRECTLY related to process size.

[gnufreex]

wrong again, moore's law doesn't mention process size either

So what it mentions according to you? Color of computer case?

[linuxgeex]

Well the 16x slot can be used for 4x and 8x boards, or even an 1x one, or for a computing GPU (would you really want to try some CUDA or OpenCL there) or for a display or a GPU with monitors attached if you really want to use it as a desktop.

I think the point is, it's just a computer and you do the computer things you want to do with it? And you're not stuck with a few GB of soldered LPDDR memory. Likely most customers will ssh into, even from a Windows computer, while two or three people will make it their main workstation as it's better than a Richard Stallman's MIPS laptop and cheaper than a POWER8 or POWER9 motherboard.

I agree that consumers can do with it what they want, I'm just saying that if they planned to market it as a workstation - which is what they have done - then they made a poor choice in their selection of cores. I am criticising that decision, and their decision to keep marketing it as a workstation, because it will result in elevated costs to consumers (due to consumers not understanding what this product is actually good for resulting higher per-unit costs due to smaller production runs) and possibly the failure of this product in the marketplace. They need to give up calling it a workstation and start calling it a build box or server, which is what it is actually good for.

[andreano]

Will ARM DynamIQ make it easier to have 23 small cores + 1 big core?

Current big.LITTLE has some limitations (something about "CPU clusters", and is why you don't see smartphones with 7+1 cores) which DynamIQ is supposed to solve.

So when Android phones have had these new DynamIQ [a55 + a75] processors … for 5 years … maybe then something great will come out.

[grok]

Well we have to look at what this little server chip does, it does not necessarily use the same "clusters" as on phones and similar consumer products. Perhaps (probably) the consumer stuff doesn't scale to many cores.

And so I looked up the chip :

https://en.wikichip.org/wiki/socionext/sc2a11

It looks like twelve dual core CPUs, sharing a bus and L3.
I don't know if the block diagram sums up every I/O, perhaps that's all there is - I suppose a set of 4x PCIe goes to the big slot, then there's another set of 4x PCIe - two for the 1x slots, one for SATA and one for USB 3.0 I presume.

Max RAM size is supposedly 16GB, but with ECC. So if you thought to load it down with tons and tons of firefox and chrome processes (silly example) well you might run out of memory anyhow, if you don't watch it.

I wonder if it's somewhat limited by the on-die bus when trying to load all the cores (would surely depend on the workload)

Still looks nice compared to e.g. old Geode x86 boards or routers with 32MB RAM. 8 to 16GB RAM doesn't sound bad then.

[andreano]

This is beginning to make sense now in the first week of 2018: A53 (and A55) cores are immune to Spectre & Meltdown—they don't execute speculatively.