Parallella $99 board now open hardware on Github

Although Adapteva are still fulfilling their Kickstarter committment, their shop is already open for preorders of the 16-core Epiphany board for November delivery.  Three options appear to be available:

If "No GPIO" means none, zero, zilch, that doesn't appear very enticing, I must say.  If this describes the situation accurately, the range of application of the basic board will be a lot narrower than expected.  And if the Zynq-7020-based Parallella-16 costs $199, then the price of the Parallella-64 is probably going to be very unfriendly.

Although Adapteva are still fulfilling their Kickstarter committment, their shop is already open for preorders of the 16-core Epiphany board for November delivery.  Three options appear to be available:

If "No GPIO" means none, zero, zilch, that doesn't appear very enticing, I must say.  If this describes the situation accurately, the range of application of the basic board will be a lot narrower than expected.  And if the Zynq-7020-based Parallella-16 costs $199, then the price of the Parallella-64 is probably going to be very unfriendly.

Does anyone have a ballpark price for the Xilinx XC7Z010 or XC7Z020-1CLG400C?

I ask because if the price difference between the Zynq 7010 and 7020 versions of Parallella-16 is $80, then it's highly likely that the cost of the 7010 is a large part of the BOM cost of the entire board.

Combining the gigabit Ethernet which is not a low-cost option with the (probable) high price of the 7010 device, this suggests that the Epiphany-16 chip itself is not very expensive because somehow it all has to add up to $99.  Since the board is open hardware, this in turn suggests that far cheaper Epiphany-powered boards could be created by open hardware enthusiasts by replacing the extortionate Xilinx hardware with something else.

If as seems likely, the Epiphany-16 really is inexpensive, 4 of them on a board fronted by a cheap TI or Allwinner ARM SoC and interfaced through low-end programmable logic sounds a lot more appealing than multiple hundreds of dollars for a Parallella-64 which isn't yet in development anyway.

Morgaine Dinova wrote:

 

Does anyone have a ballpark price for the Xilinx XC7Z010 or XC7Z020-1CLG400C?

 

I ask because if the price difference between the Zynq 7010 and 7020 versions of Parallella-16 is $80, then it's highly likely that the cost of the 7010 is a large part of the BOM cost of the entire board.

At Avnet.com, which is where I usually price Xilinx, the XC7Z010-1CLG400C is US$53, quantity 100.  The XC7Z020-1CLG400C is US$108.  So it's about US$55 difference.  Remember that the 7020 Parallella includes the GPIO connectors.

The Parallella pricing is for "preorder", so may go up in November.  The original Xilinx marketing materials for Zynq talked about US$15 chips, but that's probably a target, with absurd quantities.

I'm sure Adapteva will happily sell Epiphany chips to whoever wants to buy them.

John Beetem wrote:

 

At Avnet.com, which is where I usually price Xilinx, the XC7Z010-1CLG400C is US$53, quantity 100.  The XC7Z020-1CLG400C is US$108.  So it's about US$55 difference.

Very interesting!

It was only a rather unsafe inference that the 7010 was a large part of the cost, but it seems that I guessed right, or at least not entirely wrong.  It seems then that the Epiphany-16 costs comparatively little, and most of the $99 price of Parallella is going to feed poor starving Xilinx.

Since it's open hardware, alternative designs without the Xilinx tax beckon.

Morgaine Dinova wrote:

 

If "No GPIO" means none, zero, zilch, that doesn't appear very enticing, I must say.  If this describes the situation accurately, the range of application of the basic board will be a lot narrower than expected.  And if the Zynq-7020-based Parallella-16 costs $199, then the price of the Parallella-64 is probably going to be very unfriendly.

Given there's an 'optional upgrade' for the GPIO connectors it seems likely that the difference is simply down to installing the connectors.  Any volunteers to hand solder four of those ?

In some ways you can see the reasoning, not having them will not prevent you doing software things on the Epiphany processor.  If you really want gpio, and don't care so much about the Epiphany there are probably better boards.

Am I correct in thinking that the only difference between the 7010 and 7020 is more FPGA space ?  If so, what's this board really meant to be, a dev board for parallel processing on the Epiphany, or an FPGA dev board ?

selsinork wrote:

 

what's this board really meant to be, a dev board for parallel processing on the Epiphany, or an FPGA dev board ?

If Adapteva had asked themselves that question very clearly and seriously, I suspect that Parallella would have a very different design and a very different cost.  As it stands, the main effect of the board will be to promote the Zynq range to a far greater number of people than Xilinx would normally expect, but to a far smaller number of people than Adapteva would like as an audience for Epiphany.

It's a design from heaven ... for Xilinx.

I agree that the choice of Zynq seems a bit odd - is it that the Epiphany chip is not seriously useable with out the support of a 'big' ARM processor and  an FPGA to glue them together.

I just had  a quick look on their website and it seemed that all theri applications diagrams showed the Epiphany connectedd to an FPGA.

MK

Morgaine Dinova wrote:

 

As it stands, the main effect of the board will be to promote the Zynq range to a far greater number of people than Xilinx would normally expect, but to a far smaller number of people than Adapteva would like as an audience for Epiphany.

My feeling was always that Epiphany would have a very small audience. Both the RPi and BBB have shown there's a market for a low priced board, but that a large section of the buyers are thinking 'media center'.

Price wise, $99 doesn't compare well. The circuitco page is showing 74020 BBB shipped as of today.  So you have to wonder if the Parallella can ship enough to get to Michaels prices for 100K Zynq devices. Maybe they can, or maybe Xilinx have given them a good deal up-front, but either way I feel you're right and the Zynq will end up overshadowing the Epiphany.

selsinork wrote:

 

Price wise, $99 doesn't compare well. The circuitco page is showing 74020 BBB shipped as of today.  So you have to wonder if the Parallella can ship enough to get to Michaels prices for 100K Zynq devices. Maybe they can, or maybe Xilinx have given them a good deal up-front, but either way I feel you're right and the Zynq will end up overshadowing the Epiphany.

Whereas if Adapteva had mounted the Epiphany on a barebones Arduino shield or BeagleBone cape or Pi plate with minimal glue logic, the board could have ridden the huge wave of established ARM and AVR enthusiast communities and at Pi-type prices.  This seems clear from our ballpark cost examination above.

Note that if 100k volumes would make Zynq prices plummet, they would do the same to the cost of the Epiphany chip, and so the price imbalance would remain.  Volume does not change the overall picture of a fundamentally misplaced choice of host pairing.  And with the greater volumes, Adapteva would even be making a profit in this early experimental phase, instead of having to pay the bulk of proceeds from sales to Xilinx.

Just imagine if the Pi or BBB contained an additional component that is many times as expensive as their main SoCs.  The "Pi price niche" would not have materialized, and hence neither would have the enthusiastic mass adoption of Pi and the boards that followed it.  Adapteva may have made a big mistake.

Michael Kellett wrote:

 

is it that the Epiphany chip is not seriously useable with out the support of a 'big' ARM processor and  an FPGA to glue them together.

Epiphany has no dependency on its host processor except as a means of downloading code to it and then supplying the data for that code to process.  The Epiphany cores have neither instruction set nor architectural similarity to ARM and can work well with any host system that can supply the code and data sufficiently quickly.  There is no single definition for "sufficiently quickly" here --- in parallel programming, that is always going to be application-dependent.

The Zynq's two Cortex-A9 cores are each not hugely faster than our BBB's Cortex-A8, and they're being clocked at only 800MHz according to the Parallella Gen-1 Reference manual so computationally they're each pretty much in the same speed ballpark as the BBB's single core.  In other words, the Zynq's ARM is not providing any significant speed advantage over much cheaper devices, at least per core.  Presumably both cores can feed the Epiphany simultaneously, and if so then there is a throughput advantage gained by using a dual-core ARM over a single-core device.  There is no shortage of multi-core ARMs these days though, and most of them probably cheaper than the Zynq because they're made for the consumer market.  (*)

The host CPU(s) aside, there is also the question of interfacing to the Epiphany.  I expect that the reason why Adapteva have chosen to interface through an FPGA is because they haven't yet tied down the optimum way in which to interface to Epiphany.  After all, if they knew this already then there would be no need to use programmable logic and the extra costs associated with it.  Choosing a host SoC that combines ARM and FPGA looks like an ideal combination for their purposes, but this is true only if the cost of combining the two functions doesn't adversely affect the desired goal of bringing Epiphany into widescale use.  It seems to me that the Xilinx device is doing exactly that, because its huge price must be limiting Epiphany uptake.

The FPGA design files have been made available by Adapteva so we'll be able to see what the interfacing requirements of Epiphany really are --- see the quite detailed "Parallella Platform Reference Design" white paper which also includes a (broken) link to the HDL, now available here.  Given that knowledge, I bet that a much more cost-effective host design can be found, one that can bring Epiphany to a far wider audience, more cores at a given price, and higher profits for Adapteva through focusing more strongly on their own chips rather than on costly distractions.

===

(*) Addendum:  For example, the unit price of the dual-core Allwinner A20 is 12 euro from Olimex, and 9.60 for 50+.

The Epiphany is a sort of co-processor - it doesn't have peripherals of its own so it's always likely to need glue logic to fit it into a system that does anything useful. The really nice thing about the Zynq is the tightness and quality of the coupling between the FPGA and the ARM cores - nothing else (that I know of) comes close. So if you want the Epiphany to do the hard work for an ARM the Zynq is about the best solution on offer (in terms of performance) so it's a reasonable choice if the main goal is to show off the E at it's best.

The problem with a great many low cost ARM (and other) processors (the Broadcom and Allwinner A20s included) is that they don't offer low latency high bandwidth data paths in and out of the core. You might get sata, pcie etc with quite reasonable bandwidth but awful latency.

So if you want the Epiphany to do a lot of talking back and forth with the rest of the system you'll need an FPGA and if you want that to communicate well with the processor the Zynq is as good as it gets.

The A20 at about £8 with a £10 fpga isnt going to be a lot cheaper than the Zynq but the performance will be a lot worse.

So having thought about a bit more I think the Zynq does make sense - it lets their board address the widest range of applications and probably will enable them to demonstrate just how fast their chip can go.

MK