The first part is a look in to the crystal ball of future ARM cores. As you might remember, the current Calxeda chip is called the ECX-1000 and it is based on the tried and true ARM A9 core. The next generation of product is code named Lago and as you might guess, it is based on the A15 core. Following that is Midway with a CPU core based on ARM V8 cores. Lago is due in 2013, Midway is due in 2014, and brings a new bus. We would love to tell you more about this, but Calxeda didn’t actually put out anything worth writing other than the code names.
Of the two, Midway is the one that matters. ARM based servers are in a tight spot, they are pretty unappealing unless your workload fits the paradigm that they excel at. The main reason for this is not single threaded performance, but that does matter to most buyers, but memory size. A9 is a 32-bit architecture so it is pretty crippled for memory space, and this is a deal breaker for many. A15 is 32+ bits, it can see more memory but each thread is still limited to a 32-bit address space. This means it is just as appealing to a customer as an A9.
Lago will be better at single threaded apps, have a higher memory ceiling, but will not do either of those well enough to make anyone who wasn’t a good fit for the A9 parts care. Midway with it’s V8 architecture will break this barrier and bring not just more single threaded performance, it will uncork the memory bottleneck too. Facebook for example is crippled by the 32-bit limit, it makes ARM a non-starter for their data centers. V8 will change that, the game gets real for Calxeda and ARM server vendors in 2014.
That brings us to the customers, Penguin and HP. We would love to tell you about both, but, err, neither actually have pages up for Calxeda products and the press releases link to, err again, nothing. We will spare you the fluffy awesomeness of the self-congratulatory releases that don’t actually say anything from both companies, it is for your own good.
The one company has not only put out a release, but linked it to benchmarks and has a page up too. That company is Boston and their Calxeda based server is called Viridis. The benchmarks they linked are not comprehensive by any means, but do give you a good starting point. Take them with a big grain of salt until independent and more comprehensive tests come out, but this is a far better start than most new architectures. That said, you can spec a Viridis out, buy one, and get some hard numbers, the competition doesn’t seem to have pages up yet.
Overall, things are really encouraging for those interested in ARM servers. If your workload fits the type of things Calxeda products do well, your servers have arrived. If they don’t, you are going to be using x86 until 2014 at the earliest. Calxeda and ARM servers have gone from curious idea to available now with a path forward in a single day, how can you complain?S|A
Latest posts by Charlie Demerjian (see all)
- ARM adds CMN-600 interconnect and DMC-620 memory controllers - Sep 27, 2016
- Globalfoundries 7nm process isn’t even close to the name - Sep 26, 2016
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016
- Everspin and Globalfoundries team up for embedded ST-MRAM - Sep 15, 2016
- Intel’s Xpoint is pretty much broken - Sep 12, 2016