Intel SoCs it to 'em with new D: Tiny but powerful After some teasing late last year, Intel has taken the full wraps off the Xeon D: a system-on-chip (SoC) version of its data-centre darling that it hopes will excite hyperscale operators and those keen on very dense server rigs. The Xeon D will come in four and eight CPU core models, D-1520 and D-1540. Each will run two …
And the price differential. Intel is currently slightly ahead in the 14nm process with Samsung and TSMC very close behind. While ARM in the server is still missing important parts of the software eco-system, it benefits even more from economies of scale than Intel. That might be the big difference in comparison with previous Intel vs. AMD, et al. battles. But the holes in the software side will also need fixing.
All 14nm processes are not created equal :-)
http://www.extremetech.com/wp-content/uploads/2014/12/Cell-SizeComparison.png
And when was the last time YOU bought a server with the primary goal of it being $200 cheaper than the power-guzzling slower alternative?
In the past year the "ARM Server" redux has gone from "low power" to "more integration" to "Cheaper" to "Just wait til the software catches up".
This whole ARM server thing is riding the 4 year old myth of "superior smartphone power levels".
It sucks, its always sucked and its time for it to die.
Mainframes were displaced by Mini Computers. Mini Computers were displaced by workstations. Workstations were displaced by x86. Why do you insist that x86 is immune?
These Xeons are 45W parts. Current technology ARM 64 chips draw about 4-5W for a 4 core CPU with onboard IO. 40 physical cores on the same energy budget as one 8 core Xeon?
All they need do to take a large chunk of market share is to be fast enough, have software support and be cheaper.
The reason they draw 1/10th the power is that they do less than 1/10th of the work.
"Fast enough and cheap" isn't going to overcome faster and cheaper parts like this at 2W per core.
http://ark.intel.com/products/77980/Intel-Atom-Processor-C2530-2M-Cache-1_70-GHz
Which has been on the market for 18 months and can be made for $7 (Its a glorified tablet chip with the peripherals removed).
Would you invest your pension pot in a head to head against that knowing that something nearly 2X as good just got launched yesterday.
Unfortunately for you and Intel that number is closer to 1/2 than 1/10, and the newly anounced A72 core reduces that lead significantly. The other, rather large factor that you're ignoring is that companies can get ARM based chips made to specification, with precisely the onboard IO support that they require rather than Intel's one size fits all approach. This makes the complete system cheaper by more than the cost of the CPU. I do hope you aren't resting your pension hopes on Intel.
And when was the last time YOU bought a server with the primary goal of it being $200 cheaper than the power-guzzling slower alternative?
People don't buy individual servers any more, data centres buy heaps of them and people buy or rent capacity there. The owners of data centres, therefore, have a huge interest in the TCO of what they buy: price, density and energy demand are very important.
Most of the software stack is now available and for proprietary stuff ARM is the better platform anyway as it's easier to add dedicated stuff in silicon to boost performance and reduce running costs. ARM still lacks real oopmh for some jobs but it's current problem in the data centre is not having a standard firmware to make it easier to swap bits of kit in and out. We'll have to see if what ARM has promised on this works. On power/performance ARM is still ahead of Intel and the servers can be denser – the years of developing for mobile phones really do matter.
At the end of the day I don't really care what hardware my stuff is running on as long as works reliably. That Intel takes ARM seriously can be seen by the various products it's released over the last couple of years culminating in this. As with AMD's x64 this shows the market working. Once ARM-64 systems are available in number we can expect to see Intel reacting on price.
ARM's advantage over Intel remains the different business model. Instead of going after just AMD (and maybe Cyrix), Intel is facing a bunch of well-funded competitors and ARM itself is insulated somewhat from the struggle. Of course, the diversity has also held back the move into the data centre and Intel has some top notch people but I find the developments in the ARM architecture and manufacturing over the last few years far more impressive than Intel's rearguard action.
Intel will indeed react on price and "minimum price" is dictated by silicon area.
As an example of how the economics work against people like Applied Micro:
The world's most efficient server chip (Intel's Avoton C2730) is about 90sqm on 22nm and it costs about $7 to make. The moment Xgene2 appears Intel will bomb the price of that part to $7.50.
(600 good die from a $3000 wafer plus $2 assembly/test). Its above "marginal cost" so its legal.
Xgene invested in order to achieve 'server class margins" of 80 percent and they will in fact achieve "break-even" if they're lucky.
All they will achieve is to keep Intel "honest" at the low end and will fail to recoup their investment.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
