The traditional FPGA market is abandoned

76 点作者 jamesbowman将近 9 年前

8 条评论

lvh将近 9 年前

I'm not sure I agree with the idea that Intel is going to make these folks work on something else because FPGAs don't grow that much. That makes little sense; Intel isn't really hurting for more employees, and the FPGA engineers can't necessarily be made to be productive on the higher-growth sections cited.Instead, it makes a lot more sense that Intel is going to try and make that market more lucrative and larger instead. As a major chip manufacturer, they're in a great position to ship FPGAs to tons of new customers. Chicken-and-egg problems (you don't FPGA until you need to because it's not available) have made FPGAs a niche element. However, OpenPOWER/CAPI have demonstrated that you can get huge benefits from slapping FPGAs to existing general purpose compute.So; TL;DR: I don't think it makes sense to assume Intel is just going to assume that market will stay what it is. Instead, they think they can make that market better, and do more than Altera/Xilinx can do individually. In that light, a purchase makes perfect sense.(Disclaimer: I work for a company involved in OpenPOWER/OpenCompute and has shipped hardware that does this.)

评论 #11780320 未加载

PaulHoule将近 9 年前

People are still scratching their head over this acquisition but here is my take.Intel has failed at phones not because "x86 sux" but because Intel makes a small number of SKUs that are often different bins of the same design -- these are heavily tuned in every way. Phone chips are not so optimized in terms of process and design, but instead they are true "systems on chip" where a number of IP blocks are integrated to produce something which is optimizable by adding functional blocks.Something FPGA based, whether it ends up in the datacenter or elsewhere, could have a highly optimized FPGA block which is field configurable, so this gives Intel something to compete with SoC vendors on their own terms.One detail is the nature of the memory and I/O systems. FPGAs can move a huge volume of data, so unless you upgrade the the paths in and out of the CPU/FPGA area, the FPGA would be starved for bandwidth.It would take one of two things for the FPGA market to expand based on these developments.First, if there was some "killer app" that sold a huge number of units, that would help. The trouble here is that if you sell a huge number of units, you might be better off making an ASIC functional block an integrating it into a true SoC.The other one is that (I think) FPGA development today is constrained by the awful state of tooling. Pretty soon you will be able to pop a FPGA hybrid chip into a high-end Xeon workstation and have one hell of a development kit in terms of hardware, but without a big improvement in the tooling, very few people will be able to really make use of it.

评论 #11781725 未加载

评论 #11787298 未加载

评论 #11781670 未加载

评论 #11783277 未加载

dmytroi将近 9 年前

Ehm, FPGA/CPLD are everywhere, it's hard to say, but my educated guess is that hardware market consumes similar amount of SoC-like and FPGA/CPLD chips, look, even Pebble smartwatch uses Lattice iCE40 FPGA, MacBooks also usually have few FPGA's here and there. FPGA/CPLD's are essential to implement "glue-logic", which does what it says - glue different pieces of hardware together. Of course you can do "glue" in software, but than your customers will enjoy poor battery life time and random glitches here and there. And nobody gonna spend big-money (it's actually depends, sometimes it 1kk+$, sometimes you can get away by 30-50k$) by running their own ASIC's when you can just buy jellybean FPGA's.Indeed phone/desktop market might move to more one-chip-for-everything solution, but even then we need glue-ish logic to control something like screen backlight DC-DC converters, charging IC's, etc, which is much more easier done from FPGA/CPLD-like devices. On the other hand FPGA/CPLD's are essential in some classes of devices, for example test instruments: modern oscilloscopes usually have 3+ FPGA's in them, companies like Keysight usually only run custom ASIC's when they hit limitation of current silicon tech, like their N2802A 25GHz active probe (starting from 23500$) uses amplifier IC made with indium phosphide process (InP), which is kinda far away from whatever current consumer product companies are doing, you can check the teardown of this beast here <a href="https://www.youtube.com/watch?v=jnFZR7UsIPE" rel="nofollow">https://www.youtube.com/watch?v=jnFZR7UsIPE</a>So in my opinion FPGA/CPLD market will live long and strong, players might change, but demand is enormous. The only problem in my opinion is that whole market is more B2B-like (FPGA's are usually just a humble IC's inside end customers products, you don't see stickers "Altera inside" or anything on products themselves), so it's kinda hard to get grasp what's going on.

kevin_thibedeau将近 9 年前

XESS is biased toward the segment of the market they play in which is centered around computational applications where having a micro on board is usually desirable. There are many more places where simple glue logic is needed and a small, cheap FPGA/CPLD fits the bill. One would hope that Altera's (and possibly Xilinx's) new masters won't fuck this up and shutter their low end devices but that doesn't mean there won't be demand for them into the future. People with a need for these types of FPGAs aren't XESS customers so they are blinded by the limited scope of their own business.

kqr2将近 9 年前

In case the link is down:<a href="http://webcache.googleusercontent.com/search?q=cache:http://www.xess.com/blog/extinction-level-event/" rel="nofollow">http://webcache.googleusercontent.com/search?q=cache:http://...</a>Also, I believe that cheap microcontrollers have been able to replace FPGAs in some cases.

评论 #11780699 未加载

JoachimS将近 9 年前

The article is based on the premise that Altera will stop developing FPGAs for general availibility. Nothing I've seen (and would love to be shown otherwise) so far indicates this.FPGAs has been used as test designs when qualifying fabs for volume production since they are regular, but more complex and closer match to general ASICs than memories.FPGAs are also often used more and more either with internal CPU cores (hard or soft) or as a companion to CPUs providing acceleration, esp data plane processing.This means that the aquisition of Altera is mor of adding a business that complements the CPU business, not remove a competitor from another market segment. Intel can sell CPU+FPGA solutions for data center, big data. But it can also sell more chips and increase utilization in their fabs.And for FPGA users the FPGAs coming out of these fabs will probably be better with higher density, lower power consumption that what Altera managed to design themselves. And getting a Stratix or a Cyclone SoC with an Atom core inside running Linux would be a very neat solution.

petra将近 9 年前

Or there's another alternative. The main reason it was hard to make money from the industrial etc segment - was low volume combines with a lot of support costs.But what ifa. xilinx built c-like tools that enabled embedded software engineers develop easily ?b. they released those freely to some segment of the market ?c. they've built an external support and IP ecosystem, either open or closed or both ?Those actions can increase margins for xilinx, and they seem to be doing a,b.As for the hardware, maybe the article is right. Also ,recent industrial chips are using 28nm, and going beyond that is extremely expensive and might not fit the industrial scenario anyhow, maybe there's not a lot of innovation left in the industrial FPGA market,