48 cores

I see you have given up trying to discuss the issue. Thanks for the debate.

You quickly learn in any seriously parallel or multicore code that the task with the absolute highest priority after realtime data acquisition is the one that allocates the work to keep all the other cores busy.

Processes start, run, then stop when they have finished what they were doing. So on a recently started Linux system, the highest process number is a rough estimate of the number of processes that have run since startup.

I think FPGA's turn up in quite a number of TV's. The high-end TV's get high-end FPGA's, the mid-range ones get cheapo FPGA's, and the low-end ones wait for the ASICs to come out.

How relevant the TV FPGA market is depends on your viewpoint. In terms on TV's sold in total, the number is tiny. In terms of profits for TV's, it's significant - the margins on the high-end devices is far greater than for the cut-throat low-end market. And for the FPGA manufacturers, it is huge.

MS is always very slow to see new ideas - if they think ARM servers will take off, it is because everyone else in the market /knows/ they are already taking off.

ARM servers already outweigh Itanium servers by a factor of thousands. Most small NAS systems, which are highly popular, have an ARM cpu. And they are not uncommon on larger NAS/SAN servers in the business world.

What is new is increasing use of them in web servers, database servers, file servers, etc. All that is needed is the 64-bit ARM cores (because of the unjustified idea that server cpus need to be 64-bit) to reach full production, and for the big names to start selling.

I would expect that within a year or so, ARM servers will be a mainstream choice - not dominant, of course, but a solid choice.

On 09/06/14 13.15, David Brown wrote: ...

"Something wonderful is about to happen":

September 12, 2013, My experiments with ARMv8 server software stack:

Glenn

Meanwhile Intel is still innovating:

05/27/2014, Intel Core i7-4790, i5-4690, i5-4590, i5-4460, i3-4360, i3-4350 and i3-4150 Review. Meet Haswell Refresh!:

Glenn

Google arm server and see what's actually happening.

If you have the luxury of allocating a new core for every interrupt source (such as a serial/ethernet line), the system gets quite clean, when you don't have to mess with top/bottom ISRs etc. :-)

Who are using Itaniums these days ?

This is an honest question.

I know that some VMS (OpenVMS) customers are running Itaniums for their Itanium virtual machines, but what else ?

Are they using VAXclusters or perhaps some Tandem systems on Itaniums?

Redundant CPUs, redundant disks and redundant interconnects are nothing new, DEC had some good redundant systems available in the

Sounds like you are describing the GA144, a chip with 144 processors. Each one is very fast, very small and low power with only a very small amount of memory. The idea is to allocate processors to small tasks in an analogous way to how logic in an FPGA is allocated to small tasks. It is not likely to ever run Linux, but it can do a great job of real time tasks.

That proves my point. The ARM server market is in its infancy with most of the articles being about "firsts" and "sampling" and other marketing hype... but wait, here is another opinion from the Motley Fool...

There's nothing magical about ARM servers Marketing headlines that ARM server players typically try to push are usually of one or more of the following flavors:

ARM is disrupting x86 servers ARM servers will offer choice ARM is lower power

Unfortunately, while these marketing points may be catchy, they're almost patently false.

This is what you wanted me to read, right? Read the rest of the article instead of drinking the marketing Koolaid. :)

Turns out the current crop of ARMs from AMD and Intel just can't stack up against the x86 favorites.

The good news is that we will all reap the benefits in lower power devices across the board, laptops, desktops, tablets, etc., while still running the bulky, inefficient OS that we all know and... well know.

I'm curious. What do you base this on? Do you have any figures for how many FPGAs are sold in TVs?

"Taking off" or just "showing up"? I don't see any evidence they are even making a dent. See my reply to John's other post referencing the Motley Fool article.

Again, what do you base this on? Even if it is true, it is pointless. Itanium was never a big seller.

Full production is just an order away. My understanding is that all of the important software is not there yet. In other words, renovating the server market won't happen overnight even once the ARMs are the better solution.

Bingo. Not dominant for some time to come. Not sure what a "mainstream" choice means really.

out

out

People that got locked in with specialty software that they are too cheap to replace.

One of the great things about benchmarks is that they are almost entirely useless - except to make whatever positive or negative point that interests you at the time.

In particular, benchmarks like SPEC are of almost no relevance for servers. In most server applications, the emphasis is on pushing data around - not on calculations. And there will be multiple parallel threads, not one single thread like in SPEC. Will ARM be better at this than x86 in performance per watt? I don't know - but I /do/ know that SPEC benchmarks tell us absolutely nothing.

Another issue is that what matters is total power - not the power of the cpu chip. In x86 systems, there is traditionally quite significant power in the peripheral chips - in ARM systems, peripheral interfaces are usually part of the main SOC. Which system will be lower power in /total/ ? Again, I don't know - and again, a power rating for the cpu chip tells us very little.

Then there is the issue of standby or idle power compared to full power. Many servers spend a lot of time idling, and so low-power modes can be more relevant than performance per watt. Many server applications don't /need/ high performance - it doesn't take much cpu power to saturate GBit Ethernet on a file server or static web server. A cpu with lower performance per watt might easily have lower total power consumption.

I don't see ARM taking over the high-end server space in the foreseeable future - but I think there is plenty of variety in the server market where they could quickly be a good solution.

There are some who believed the promises and expectations from Intel and HP, and invested heavily in Itanium machines and software (note - I am not implying this was a con - Intel and HP believed in what they were doing). Changing architectures later is expensive, so some users kept their Itaniums.

IBM have mainframes that have run for decades without a second of downtime, and which have no original parts except for the frame itself - /everything/ is redundant and hot-replaceable.

Several high-end cpu architectures, such as Power (and Itanium) support redundant and hot-pluggable cpus. I believe Power is the most popular cpu for people looking for that sort of reliability - but Itanium does have a few adherents. Of course, if you want top-tier hardware and you need to use Windows (an oxymoron, I know), then Itanium is your only choice - Power and other cpus are Unix and Linux only.

ARM servers will remain a nice market, until you get Windows 20xx Server support for ARM. This is not likely to happen very soon, since the Itanium support was dropped in Win 2012 server. Windows has been supported for quite a few platforms during the years, but all have been dropped in favor of x86 and x64.

It's based on accumulated background information (press releases, number of application notes on FPGA websites, chatting with distributors, etc.). So no, I have no concrete information, numbers or sources to point you to - it's merely my opinion, and I'll change it if I see information to the contrary.

"Taking off" as in "big names are investing heavily, making commitments, and demonstrating prototypes". "Taking off" as in "just leaving the runway". So yes, that also means "just showing up" as real-world devices.

Changes in the server world take time. First, the ideas and benefits need to be figured out and sold to the big players. Then the software needs to be in place, and hardware base needs to be designed. Then the big names need to put them into systems with surrounding support, extra value, etc. Then they must be rolled out to early adopters and customers for trials. Finally, they start becoming mainstream.

ARM v8 (64-bit ARM) servers are well on their way here - several big names have made their first ARM server families and they are starting to roll out prototypes to early customers.

/You/ brought up the Itanium.

But the point is that small NAS servers are very popular, and most are ARM based, especially the smaller and cheaper ones.

These sort of things are ARM servers:

It will take time - of that there is no doubt. By the middle of next year, I expect you will be able to order a Dell ARM server as easily as a Dell x86 server - that's what I mean by "mainstream choice".

You will also start to see ARM in specialised servers - customers like Google or Facebook will use the flexibility of ARM to make SOCs that have exactly the hardware mix they need to get chips optimised for their datacenters.

I did some googling and Itanium was still supported in Windows 2008 server but no longer in Windows 2012 Server.

This is really quite expensive, if you have to rewrite much of your in house applications.

I have maintained some VAX systems with dozens or thousands users in the past, I used to shut down one non-clustered VAX each year, let it cool down and then restart it to detect any weak components. At this stage some OS upgrades might have been done at the same time. One problem with this policy was getting calendars for the next few years, so I had to print my own calendars for the next year :-) in order to agree with the customer on boot dates the next year.

I sometimes forgot these dates or could not arrange an HW engineer with spare parts on site, but no problem, we will do this reboot the next year :-).

Of course, with non-clusterd systems, a system disk head crash would shut the operations down.

On an other clustered VAX site, the cluster was newer down during the decade I worked with them. Individual CPUs were momentary down for rolling OS version updates and the mirrored disks did not cause total system outages.

Sounds like a nuclear power plant replacement plan :-) when you are not allowed to build "new" NPPs.

Itanium is no longer supported on Win 2012 Server.