What is the cost of an FGPA unit in the 6000 series and how does it compare to a current high end Intel I-7 based computer?

  • 1
  • Question
  • Updated 3 years ago
  • Answered
  • (Edited)
Hello everyone.

Based upon another recently discussed topic, I was curious to find out what the approximate cost is of an FGPA (currently used in the 6000 series) is and how powerful it is in computational power compared to a current high end Intel I-7 based computer?  The reason I ask this is to try and understand if it was technically possible to have created the 6000 series in a manner that would have allowed all the heavy computing tasks to have been done by a powerful personal computer rather than the hardware unit; much like the previous generation of Flex radios. Quite honestly; if the same; or similar; performance could have been achieved using an external computer, I would have preferred that scenario to keep the cost of the hardware unit in check and allowed me to upgrade the computer over time instead of the radio.  It just seems that if what I am suggesting were possible, it would have been cheaper to upgrade computer systems over time rather than hardware units.

The secondary question that arises in my mind is; if a single powerful external computer were not powerful enough to accomplish the task, would it be technically possible to use the idle computational capacity of a number of computers on a home based LAN system to accomplish the same task; much like the Sony high powered gaming consoles do in terms of crunching data for the SETI project or companies that use networked computers to do frame by frame restoral of classic movies?  I realize that any cost savings would go out the window beyond a certain point, but many people do have several computers in a household.  I also suspect that latency in terms of processing data in real time might also be a limiting factor; but was just wondering what is and isn't technically possible? 

Forgive me if my lack of knowledge may seem to make these questions appear dumb, but someone once told me that there is no such thing as a dumb question in the pursuit of knowledge.

Regards:  John  VE3INH
Photo of John M

John M

  • 34 Posts
  • 11 Reply Likes

Posted 3 years ago

  • 1
Photo of K7NXT

K7NXT

  • 37 Posts
  • 27 Reply Likes
John, the hardware cost of an FPGA is easily found online. But, I would suggest that the actual "cost" relates to the value of the programming it holds, which is much greater.  But as far as computers go, an FPGA is more like hard coded logic chips where propagation delays are measured in nanoseconds.  This allows for the very high performance. There is no comparison to a PC program, where 10s or 100s of background tasks are running simultaneously within an operating system that also has significant overhead.  Thx, Bill  K7NXT
Photo of Al / NN4ZZ

Al / NN4ZZ

  • 1678 Posts
  • 573 Reply Likes
Hi John,
IMHO Another factor (maybe the most important factor) in the 6000 design is the architectural decision to keep the data processing in the radio and only transfer the data needed by the client to the client.  

This is a big advantage especially when considering WAN remote.  Even with the much smaller data transfer, latency may be a challenge for WAN operation.

Regards, Al / NN4ZZ  
al (at) nn4zz (dot) com
6700 - HW......... V 1.5.1.70
SSDR / DAX...... V 1.5.1.152
CAT................... V 1.5.1.0
Win10 

 
Photo of John M

John M

  • 34 Posts
  • 11 Reply Likes
Thank you for taking the time to respond and explain that.  What you say makes a great deal of sense.

Regards: John VE3INH
Photo of Bob Craig, K8RC

Bob Craig, K8RC

  • 199 Posts
  • 80 Reply Likes
And lets not forget, outside-the-box software is much more susceptible to reverse-engineering and re-purposing.

Think of how many copies of PowerSDR are running other manufacturer's hardware at no profit to Flex vs. the SmartSDR client software which is of little use to anyone or anything except a 6x00. 
Photo of Duane, AC5AA

Duane, AC5AA

  • 382 Posts
  • 84 Reply Likes
Another concern would be the latency requirements to generate high-speed QSK CW - itwould be a lot more difficult in a separate box. The 5000 never got there, but the 6x00 series does a really decent job of it.
Photo of Simon Lewis

Simon Lewis

  • 495 Posts
  • 117 Reply Likes
John your comments around cost are getting repetitive and boring. The question you asked in previous threads is NOT going to change the situation!

Bottom line is regardless of what you think the cost should be or what development route FRS took, or the approach they use to develop, hardware etc this is not going to change FRS's approach, create you a 50 buck radio, or deliver something different to what YOU think it should be. Buy whats available or look at a different supplier.

And ... btw ... The reason why the heavy lift is done in unit rather than the PC is that the results using in unit processor are MUCH more reliable and reproducible. Timing for CW etc is fixed and manageable, the only processing one needs is something to manage the client end which you can run on light weight PC's or tablets. heavy weight processing in PC means a hefty powered PC with HIGH band width connections.  now all I need is a simple network cable, I can do it remote, locally connected, in house, shack or from my hotel room. Why would anyone want to chain themselves to a high end processing box that limits all this!

You see all the comments and questions around PC software here .. now imagine the problems if all this was in PC... hams do not like to invest in PC hardware to run radios and many are not PC experts. You don't need to be with SSDR, the lift is done in unit, its controlled reproducible and managed.

FRS made the PERFECT choice going in unit for the heavy processing lift and in a swoop killed the vast majority of issues found with poor PC performance, drivers etc in PSDR.

I hated my Flex 1500 ... I LOVE my 6500's 

Asking the same question around architecture and development won't change the position of FRS etc - they have the platform they need. and the 6300 for example is superb VFM ... if you want an SDR and want cheap .. buy a sec hand Flex 3000 or 5000, but stop poking around  with dumb questions ... the answer is not going to change.

And sorry there are dumb answers and dumb questions ;)

Rgds

Simon ZL4PLM



 
(Edited)
Photo of EA4GLI - 8P9EH - Salvador

EA4GLI - 8P9EH - Salvador

  • 1725 Posts
  • 530 Reply Likes
I am with Simon on this one. It is getting boring with the money questions and the constant complaining about Flex price. If you are so interested in pricing and how they run their business email FRS, heck, you can even call them. They will probably answer your phone call.

You should get an Anan, fits your limited budget, and runs off a pc.

I don't understand you ulterior motives but your posting history present a fairly clear picture.
(Edited)
Photo of John M

John M

  • 34 Posts
  • 10 Reply Likes
Salvador; please understand that I; as well as the vast segment of the amateur community at large; are not nearly as fortunate and/or wealthy as you appear to be (from what I have seen on your QRZ page).  As a result; cost does matter.  My questions and comments thus far were never intended as complaints against Flex, but rather an attempt at understanding the reasoning behind the large jump in price from the previous series and whether or not more economical means have been explored or are being explored to bring forth a radio system that could potentially open up a large new market that would both benefit Flex's bottom line and average amateurs with modest financial resources. 

I may very well be forced to take your advice and see if an Apache Labs Anan product could be acceptable, however; I hope to hold out for a more economical version of a Flex; if and when it becomes available. My Flex 1500 has served me well despite its limitations and I look forward to examining FRS future products to see if they also may match my needs.

Also; with all due respect; as a relatively new member within the Flex community, 3 conversations and 24 replies hardly qualifies as a "history" and certainly does not denote ant sinister ulterior motives.....chuckle!

Regards:  John  VE3INH
(Edited)
Photo of Tim - W4TME

Tim - W4TME, Customer Experience Manager

  • 9000 Posts
  • 3408 Reply Likes
John - thank you for question and your patience while you received constructive answers to your question.  We understand that the price point of our radios may be out of reach for some people and wanting to know the reasoning why we are more expensive than others is valid knowledge when making a purchasing decision.  While our engineering approach is more expensive than others, we feel that is is greatly superior to the other SDR competitors and that the value intrinsic to our engineering decisions will manifest itself now and much more in the future.  We have only begun to realize the potential of utilizing the compute power we put inside the radio.  There are great things to come while we realize this potential.

Salvador - As I said to Simon, you have made your point.  There is no need to continue.
Photo of John M

John M

  • 34 Posts
  • 10 Reply Likes
Thank you Tim for your patience and taking the time to explain the rationale behind what FRS is doing.  I look forward to seeing what the collective efforts of your staff at Flex generate in the future. 

Best regards:  John  VE3INH
Photo of John M

John M

  • 34 Posts
  • 11 Reply Likes
I just wanted to thank K7NXT, Al/NN4ZZ, Bob Craig/K8RC and Duane/AC5AA, for your well thought out and polite responses.  Everything you mentioned made a great deal of sense.  I appreciate your thoughts and input on the questions I posed.

Best regards:  John  VE3INH
Photo of Simon Lewis

Simon Lewis

  • 495 Posts
  • 117 Reply Likes

Insanity: doing the same thing over and over again and expecting different results.

Photo of Mark - WS7M

Mark - WS7M

  • 959 Posts
  • 337 Reply Likes
Well I don't know the history of the questions and really I don't care.

I have to echo what was said above... In general you have the three following ways to implement complex logic:

fpga
microcontroller
microprocessor


You might wonder what the difference is between a micro controller and a micro processor.  In my use here a micro - control contains on board flash that you flash the program into.  A micro processor generally loads this into some kind of RAM from a storage device.

I realize this is a very crude analogy but my point is simply this:

fpga - You burn the logic directly into the chip and it looks like a gang of logic chips doing a specific task and the time to do this is extremely fast.  You can reprogram this to change the logic.

micro-controller - a little slower.  It executes instructions from a program you plug in and burn into flash and it can do many of the things an fpga can do but probably some what slower, then you have the issue of how to update the firmware.

Micro processor (i7).  Very cool, very powerful chip but to make it work you need a clock, ram, storage perhaps some kind of boot loader or bios etc.  Not trivial where as a micro controller or fpga has everything it needs.  Once loaded the program is fast the power is great but still not as fast perhaps as an good fpga or micro-controller.

I am currently working on a project where we have a cyclone pga and we've implemented 25 stepper controller logic units in the fpga.  All we have to do is connect the step and direction lines from the pga to the actual motor drivers and with a single chip we can drive 25 steppers around.   This would be difficult to do with a micro-controller and possible with an i7 but perhaps tough to do as you have to come up with interfacing schemes.
Photo of John M

John M

  • 34 Posts
  • 11 Reply Likes
Great answer Mike.  Thank you for taking the time to chime in with your explanation.

Regards:  John  VE3INH
Photo of Peter K1PGV

Peter K1PGV, Elmer

  • 541 Posts
  • 315 Reply Likes
A forth alternative would be to create an ASIC.

The first one might be a tad expensive, but after that each one would be MUCH less expensive than an FPGA ;-)

Peter
K1PGV
Photo of Mike Meeks

Mike Meeks

  • 7 Posts
  • 2 Reply Likes
But redoing it when an update is needed, it's very expensive again .
Photo of Jay / NO5J

Jay / NO5J

  • 1401 Posts
  • 209 Reply Likes
I can't really find the price without the exact part number, and asking "What is the exact part number of the FPGA" likely wouldn't get an exact answer. What sort of price would you like them to be? Xilinx sells to Digikey but minimum order size for similar Xilinx Virtex 6 parts seems to be 36 parts which Digikey doesn't stock, at $31860.00. so $885 each but that leaves 35 more you'd need to sell to even come close to breaking even. You could probably skip the middleman by buying direct from Xilinx in larger quantities. The pricing above is Complete BS without a part number though. Google might give you different numbers if you ask it nice. If this sounds cheap to you, why worry about the price of a cheap part. Oh by the way, FPGAs don't come with any programming, But you might know someone locally that can do the job. My daughter used to babysit part time, she didn't charge to much. Your question's not dumb, but I suspect most of the answers will be.
 .
73, Jay - NO5J 
Photo of John M

John M

  • 34 Posts
  • 11 Reply Likes
Thank you Jay for taking the time to research that aspect of my query and provide your feedback.  It is very much appreciated.

Regards:  John  VE3INH
Photo of k3Tim

k3Tim

  • 816 Posts
  • 155 Reply Likes
I have seen the part #'s of the fpga and various other big ticket items on postings here that linked to presentations. I did the cost estimate exercise for the FPGA and Jay is on the mark at around $900 each. If purchased in qty 100 or more the discount was real money but that ties up a lot of capital in only 1 item.  I was at the $2k mark with only a handful of parts.  Don't forget the low phase noise oscillator is also very expensive. 
I've pulled the cover on my 6500 to check the build quality and the PWB is commercial++ meaning it looked better than most commercial boards I've seen. The cost estimate didn't include the PWB or even the final transistors. Given more parts / assembly (automated and manual /calibration / burn-in / testing / software / FPGA / DSP and support team additional costs the 6k series is "aggressively" priced in my opinion.


k3Tim
Photo of Simon Lewis

Simon Lewis

  • 495 Posts
  • 117 Reply Likes
the VE OP should ask Icom why their 7850 is so expensive .. jeez surely the TMS DSP chip isn't that expensive . how come its so costly

ROFL
Photo of K7NXT

K7NXT

  • 37 Posts
  • 27 Reply Likes
I'm not understanding the underlying point here.

There are many costs in a product:  Parts cost, R&D costs, design costs, software costs, manufacturing costs, testing costs, warranty costs, taxes, etc. etc.

The raw FPGA cost in a complete system like the FLEX seems akin to asking what the paint cost in a museum quality masterpiece painting. 

Or am I missing something?

K7NXT, Bill
Photo of Simon Lewis

Simon Lewis

  • 495 Posts
  • 117 Reply Likes
thats just it Bill .. there is no point!

He isn't happy that Flex has a price it demands .. wants to keep open conversations around pricing and why it is what it is despite numerous long conversations that cover the same ground time and again

its meaningless and pointless exercise ...

As I said ... it's not going to change things from FRS!
Photo of Tim - W4TME

Tim - W4TME, Customer Experience Manager

  • 9000 Posts
  • 3408 Reply Likes
Simon - you have made your point.  Let's move on please.
(Edited)
Photo of Rob Fissel

Rob Fissel

  • 270 Posts
  • 47 Reply Likes
Ignoring the fact that this post likely has more to do more with complaints of cost vs. an actual hardware performance question, you'd be well served to do some basic research on Google and Wikipedia about FPGAs and ASICs, and the major differences between using an application specific integrated circuit (or field programmable gate array in this case) vs. a CPU. I'm no engineer by any means, but I can certainly understand the fundamentals and reasons why FPGAs and ASICs are preferred for certain tasks compared to slaving to a computer and CPU.

As Steve, N5AC, has mentioned on several occasions on this forum, the sheer computing power of Xilinx's top end Virtex FPGA's, coupled with their (now reasonable prices for consumer applications) prices, makes the use of these IC's in their radios a no brainier compared to a high end, high horsepower PC. Thin pipe vs. fat pipe, internal DSP, and low overhead thin client software... 

Hardware talk aside, I feel that there is tremendous value in my 6300, which outperforms many rigs costing twice as much. 
Photo of John M

John M

  • 34 Posts
  • 10 Reply Likes
Hello there Rob and thank you for your detailed explanation.  I do appreciate it.

My questions and comments thus far were never intended as complaints against Flex, but rather an attempt at understanding the reasoning behind the large jump in price from the previous series and whether or not more economical means have been explored or are being explored to bring forth a radio system that could potentially open up a large new market that would both benefit Flex's bottom line and average amateurs with modest financial resources. 

Thank you again for shedding some light as to the advantages of using FGPAs and ASICs over that of conventional computers.  By the way;  after examining a 6300 at a local retailer, I have to agree with you; the 6300 does appear to outperform many pricier radios.  Unfortunately the 6300 is just out of my budget and it is my sincere hope that Flex may have a more economical scaled down version available in the future.

Regards:  John  VE3INH
Photo of Rob Fissel

Rob Fissel

  • 270 Posts
  • 47 Reply Likes
I secured my 6300 for $2k shipped and insured on QTH. Flex will likely also be offering certified used 6xxx series radios in the near future as a fringe benefit of their trade up program. Don't lose all hope yet!
Photo of KK9W - Steve

KK9W - Steve

  • 523 Posts
  • 170 Reply Likes
What does it matter what a FPGA costs. They put a value on their hardware with R&D and development and manufacturing and profit. You know the old saying "you get what you pay for" If you don't like that don't buy it.
(Edited)
Photo of John M

John M

  • 34 Posts
  • 11 Reply Likes
I hear and understand what you are saying Steven.  I guess I'm just going to have to wait and see what Flex comes out in the future that may be in my price range; that is; if I'm not tempted by some other SDR based radio.

Regards:  John  VE3INH
Photo of Tim - G7GFW / F4VQP

Tim - G7GFW / F4VQP

  • 110 Posts
  • 36 Reply Likes
The cost of individual components in the 6XXX series is irrelevant to anyone but FRS. What is relevant is the price you are prepared to pay for the radio of your choice.

As others have pointed out, one of the biggest costs to FRS is the software (and the cost of that is another discussion/argument). Software is expensive to write, test, debug and then do it all again.

As far as I am concerned, I looked at several radios before buying a Flex. By comparison to many other top end radios, my 6300 looked like a bargain and I have to say that the reality is even better than I had hoped for. Without a doubt, FRS have done an amazing job with the Signature series and I for one, will be a loyal and very satisfied customer for many years to come.

What you are paying for when you buy any product is the sum of ALL the parts, be they hardware, software or reputation. If you think that the product is expensive, vote with your wallet and don't buy it. Simple as that.

By the way, I am 69, have been a Ham for over 50 years and have owned or used many of the 'top radios' and the Flex has them all beaten by a long way.

 To revisit the cost of parts for a second, I am sure that the cost of many parts has come down a lot since FRS launched the 6XXX series - but the cost of writing software has gone up commensurately - as software becomes more complicated, the cost rises exponentially.

To sum up, by any standards FRS have produced a product that many people want to own, many people do own and most owners think is superb. 

If anyone thinks the radios are over priced, then don't buy them. If you buy and don't like it, sell it, the second hand market is strong - proof that other people want the radio.

Tim
(Edited)
Photo of Mark - WS7M

Mark - WS7M

  • 959 Posts
  • 336 Reply Likes
Perfectly stated in everyway.  Bravo!
Photo of Bob Craig, K8RC

Bob Craig, K8RC

  • 198 Posts
  • 77 Reply Likes
Long ago (Win95 was new) I sold a software package that I had spent 100's of hours writing. More than once, in the booth at Dayton, I was challenged on pricing. How did I justify asking $29.95 for the disk when, in another booth, floppy disks were 10/$5?

Except for cases of philanthropy (my favorite, DX Lab Suite comes to mind) everything is worth far more than the sum of its parts.

If you think it's easy, you build it.
Photo of Norm - W7CK

Norm - W7CK

  • 745 Posts
  • 151 Reply Likes
I think the approach Flex took when then migrated the processing from the PC to the radio was a very good one.  Like others have said, it has several benefits.  One is that it can run a striped down version of Linux which has much less overhead than Windows.   I'm sure this alone substantially increases the amount of processing the FGPA can perform compared to an I-7 based CPU.  And like others mentioned, when running remote, very little data has to be transferred across the Ethernet cable if the processing is done internally to the radio. 

The OP did have one question that I thought would be interesting to know and that was how the FGPA (currently used in the 6000 series) compares in processing power to a high end Intel I-7 based CPU.  I happen to have the 6700 and it would be interesting to have some sort of comparison.

I love this rig by the way and can't wait for the next release of SmartSDR.  I hope they address some of the problems I'm experiencing while using 2 meters.  Band switching and such.

 
(Edited)
Photo of Peter K1PGV

Peter K1PGV, Elmer

  • 541 Posts
  • 315 Reply Likes
how the FGPA (currently used in the 6000 series) compares in processing power to a high end Intel I-7 based CPU

This is a difficult question to answer intelligently, because FPGAs and microprocessors are inherently different beasts.  It's a little bit like comparing a passenger car and a passenger train. They can both move people from one place to another, but they excel in very different domains. While both of them can be used to get you from (say) Boston to New York, they are developed, used, and targeted at very different types of solutions. 

FPGAs are "slower" (in terms of clock rate) than general purpose CPUs, but because they are inherently parallel, they can get multiple things done in one clock cycle (unlike a single general purpose CPU).

You'll find an interesting (technical but approachable) discussion here.

Peter
K1PGV.

Photo of EA4GLI - 8P9EH - Salvador

EA4GLI - 8P9EH - Salvador

  • 1725 Posts
  • 530 Reply Likes
Let ́s humor this question which only took a few minutes of online research.
You measure the performance of the FPGA in Gigaflops. 

A Core2 Quad Q8400 gives you about 33 gigaflops

A Core i7 4770k at 4.6 GHz about 100 Gigaflops.

The Flex 6700 Virtex-6 121 Gigaflops.

All pale in comparison with gaming console. The Xbox 360 has about 300 gigaflops of floating point performance, on a theoretical level. The PS3 is about 400 gigaflops.

The Xbox One and PS4 are about six times more powerful on graphics. The Xbox One can do 1230 gigaflops, while the PS4 can do about 1840 gigaflops.

Therefore we should all run our radios out of a Sony PS4.

Or maybe just run it off a video card... Nvidia’s GTX Titan produces 4500 GFLOPS

Or maybe, just maybe, you just can't measure it this way and all of this is just dumb answers to a dumb question....
Photo of Frank  WA3NHK

Frank WA3NHK

  • 56 Posts
  • 10 Reply Likes
Hi Salvador,

You might want to check the source of your numbers.  I'm seeing Linpack benchmarks of about 3.8 (i.e., under four) Gigaflops for a Core i7 4820K overclocked at 3.9 GHz using Turbo Boost.  The benchmark includes some Linux overhead and consists of a double-precision multiply of a 100x100 matrix.  

Stepping back and doing a simple reasonableness check on your 4.6 GHz Core i7 number as an example I'm not sure how the 4.6 GHz CPU can perform 22 floating point operations on each clock cycle, which it would have to do in order to achieve 100 gigaflops.

In general, comparing FPGAs to GPPs is like comparing chalk and cheese.  They're totally different animals with different missions in life.  A DSP engineer would never attempt to develop a true SDR such as the 6x00 without doing the bulk of the processing work in an FPGA.  It isn't a cost issue at all.  It's physics.

The original post had valid questions on "why an FPGA?", although "price management" shouldn't have been intimated as the possible driver.  I could make this thread even more nonlinear by mentioning there's also a TI DaVinci processor (containing a DSP and a GPP) that's in the pipeline after the FPGA.  So I won't mention that.
Photo of Tim - W4TME

Tim - W4TME, Customer Experience Manager

  • 9000 Posts
  • 3408 Reply Likes
"It isn't a cost issue at all.  It's physics."

BINGO!!!
Photo of George - AB4FH

George - AB4FH

  • 63 Posts
  • 17 Reply Likes
A microprocessor takes several clock cycles to complete one of its standard instructions.  A Fast Fourier Transform  (most often a radix 2 decimation...can be looked up on Wikipedia) used in digital signal processing takes many standard instructions.  A Field Programmable Gate Array or FPGA,
can be firmware programmed to execute a custom instruction in a single clock cycle.  So, the FPGA will always be hundreds to thousands of times faster than a general purpose microprocessor for the same clock speed.

Oh, and another very important distinction, the FPGA will usually be configured as multiple parallel circuits, each executing their instruction at the same time.  So, it is a custom configured, heavily parallel, firmware circuit.
(Edited)
Photo of Norm - W7CK

Norm - W7CK

  • 745 Posts
  • 151 Reply Likes
So its true multiprocessing as apposed to Intel based processor running Windows which in reality more or less is only processing a single instruction at a time per processor.  Actually I don't think many Windows apps actually take advantage of multiprocessors yet.  Still doing the round-robin thing.....
Photo of SteveM

SteveM

  • 236 Posts
  • 39 Reply Likes

John,

If you would like to know why there is an expensive FPGA in the 6000 series, here is your answer:

http://www.xilinx.com/products/intellectual-property/duc_ddc_compiler.html#documentation

FPGA manufacturers have already developed and, therefore, provide much of the IP required for SDR (as well as many other applications). This is how much of the engineering development occurs today. In this age, no company can spend the time or money necessary to re-invent the wheel.

Photo of John M

John M

  • 34 Posts
  • 11 Reply Likes
I just wanted to relay a final thank-you to the vast majority of kind individuals who responded to my queries in a patient and understanding manner. Your excellent responses have given me a wealth of information that has furthered my understanding of the advantages of FGPA's and their functionality over that of traditional computers.  This forum is an amazing and wonderful informational resource!

Best regards:  John  VE3INH
(Edited)
Photo of EA4GLI - 8P9EH - Salvador

EA4GLI - 8P9EH - Salvador

  • 1725 Posts
  • 530 Reply Likes
You are welcome. 73s.

This conversation is no longer open for comments or replies.