Breaking News
News & Analysis

Red Hat Drives FPGAs, ARM Servers

FPGA summit set for March
2/12/2016 07:50 AM EST
3 comments
NO RATINGS
Page 1 / 2 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
kdahlgren
User Rank
Author
Driver Issues???
kdahlgren   3/14/2016 1:35:26 PM
NO RATINGS
I find Jon's comments bizarre. I regularly test a range of OpenCL benchmarks on systems that have the runtimes from Intel, NVIDIA, AMD, Xilinx and Altera. The drivers are relatively easy to install and play well together. What's the big deal?

ZolaIII
User Rank
Rookie
A rood to nowhere
ZolaIII   2/16/2016 1:41:50 PM
NO RATINGS
As this retracts on my post on earlier article about ARM servers I wanted to say couple of things. As much as general purpose CPU cores are geting less interesting ARM is losing a battle mostly thanks to them self. To start with even ARM V8 cores represent a better choice for integration for future true multi heterogeneous SoC's up to date their is a zero such SoC's commercially available. By true heterogeneous I mean ones integrating up to date GPU, CPU's along with FPGA (with DSP) along with large enough amount of integrated fast lo latency memory to conduct some serious work on it. Actuality their whose one announced last year by Colorado Engineering cold "Hercules" but as they are US military contractors all documentation about it disappear into thin air. Secondly ARM is also shooting it's foot by selling high optimized math libs (& not for petty cash either) even we all know it's NEON™(SIMD) is far from Implementations find on CISC X86 products all do ARMv8 actually showed us how this is mainly because bad not adequate interconnect & cash coherence (just take a look at MFLOPS NEON benchmarks for the same CPUs from ARMv8 architecture conducted at 32 & 64 bit). I actually believe that their is much more room for improvement by better feeding a pore animal (SIMD). We can talk as much as we like how their is a software server software ecosystem for ARM along with tools (by tools I mainly point out to compilers & libs, compute frameworks) for the architecture but we actually all know how the fruit is not yet ripe. It will took a wile to have software and proper tools there on really desired level (at least 3 to 5 year's). So actual question is how ARM can be competitive with out real hardware out there that can meet demands for future computing software or in traditional sector with pore un competitive SIMD's & pricy property libs for it? Now don't tell me it can do it by begin much cheaper in price I didn't see any servere development ARM bord with price like that actually they are usually even pricier than similar x86 ones. So the answer is actually obvious, they can't. & I do know couple of serious scientific project that fallen down because of lack of hardware to be conducted on. As a matter of fact one like that fall in they back alley in Edinburgh University at Advance computing Algebre department. If they continue like this people will simply loose interest soon & then they will be certainly doomed.

ckachris
User Rank
Author
FPGAs in the data centers
ckachris   2/12/2016 9:15:53 AM
If a standard software interface to thes FPGA is defined, it will help a lot for the efficient utilization of the FPGAs in the data centers.

Actually, this is also the target of the EU-funded VINEYARD project: to develop servers coupled with FPGA and dataflow engines from Maxeler. It also aims to develop an integrated programming framework for allowing end-users to seamlessly utilize these
accelerators in heterogeneous computing systems by using typical data-centre programming frameworks (i.e. Spark). http://vineyard-h2020.eu/ 

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed