Revisit: Altera vs Xilinx (NIOS II vs Microblaze)

They're both effectively the the same.

If you want a free CPU and cross Vendor compatibility, try the LatticeMico32.

There's not much in it. All the vendors tools can do the job (most of the time ;) ).

Cheers, Jon

PretzelX,

Xilinx recognizes the investment made when choosing a processor/architecture/language; and has made every effort to follow the "golden rule": Never obsolete your processor.

And, so far, we have never given any customer cause to worry.

There are no plans to (ever) change this policy.

Our track record also speaks to keeping this commitment.

The same can not be said for our competitors.

Austin

How is it possible to obsolete a soft core?

-- Mike Treseler

By offering it in the form of device-specific macros or semi-compiled descriptions.

Only if the processor core is true "source code" written in terms of generic or practically duplicatable library objects do you have immunity from obsolecence. Otherwise, you are dependent on the targeted devices continuing to be available, or whoever has the true source code offering you new versions that work on newer devices, or some third partner creating a workalike.

Mike,

Easy. Just introduce a "new and improved" incompatible version, and remove support for the old one.

Another way of saying "the original soft processor was so bad, that we redesigned it..."

Thankfully, MicroBlaze(tm) soft processors followed the Harvard RISC architecture, and we got it right the first time.

But, enough of this venture into the dark and murky realm of choice of microprocessor architectures and language.

The biggest, and most important reason to choose a processor is what you already have written: can it be targeted at your choices?

Austin

Yes. The device can become obsolete, but if I have the netlist and the device, the "processor" usage of the device cannot be obsoleted.

-- Mike Treseler

Only so long as you can buy the devices. Obsolecence doesn't usually imply that the devices in your company stock room / desk drawer stop working, it implies that there aren't and aren't going to be any more in your distributor's stock room for you to buy. Or it may imply that you are locked into using old technology devices and unable to take advantage of newer ones which may badly needed advantages / ability to keep up with your market.

At that point, the netlist is only going to help you if you have alternate-device versions of all of the components that it instantiates.

Where I live harvard architecture means exit... Having a seperated code and data area causes a lot of overhead in software because each pointer needs to be extended with the memory type it is pointing to.

It also makes the CPU more complicated because you need twice the amount of memory move instructions which wastes opcode combinations which could have been used for other usefull instructions for functions that usually take several instructions (like bit set, clear, and, or, xor).

And why not use an ARM core? That would have made a lot more sense than creating something completely new.

Hi,

I think it would be hard to find an ARM core that isn't harvard architecture.

Harvard architecture means that the pathways from the CPU are seperate for instruction and data. It doesn't mean that the actual memory needs to be two seperate memories. In fact the common usage is that the memory is the same.

What Austin was referring to was that Altera went from a different architecture/ISA when they moved from NIOS-I to NIOS-II. All programs needed to be recompiled for NIOS-II and assembler code had to manually be translated to NIOS-II assembler instructions. MicroBlaze has stayed with the same ISA and all new features are optional. You can take object-code from the first version of MicroBlaze and run it the latest version (and in the future versions).

Göran Bilski

You can, it's just larger and slower. So it's the designer's call. Which would you choose ?

- A slower/larger core, with a license fee hit, or a smaller/faster one, FPGA optimised, and license-free on the silicon. The Lattice Mico32 is free, if you want that option.

Actel are offering two ARM flavours, and like NIOS I/II the ARM7/Cortex M3 are not object/binary compatibe, but need a recompile and asm-recode.

Or, you might choose to not use a FPGA-CPU at all, as the [32 bit + FLASH] offerings become more widespread, and have peripherals the FPGA can only envy.

-jg

Let's see, just where has Xilinx violated your rule? Complete abandonment of XC4000 sortware support when Xilinx switched to XST?

I certainly got stuck with a few hundred new XC4K parts and a Xilinx ISE 4.1 synthesis license that expired and you would not issue.

You claim is outright hot air, an unfailrly damming your competitor with false assertions.

There is a simple answer: Click on

Is there also a simple answer as to get parts you announced two years ago but which still haven't materialised? ;-)

Are you ever going to make the XC4VFX40?

Jon

Peter, show me where the VHDL and Verilog are in that release, exactly the items that were in what Xilinx sold in ISE 4.x and refused to provide me keys for.

Sorry to rain your YOUR parade Peter, but you know just as clearly as Austin that Xilinx dropped support for XC4K synthesis and was unable to issue licenses to customers that had current, valid, ISE registrations. That Xilinx failed to provide a replacement for existing license holders in XST, basicly telling them to go to your

Email at the time, refusing to honor licenses I purchased ....

referring to is the

this was

manager without

of this year we

reasons. Therefor,

the mean to

for a new

that

anymore questions

inform

licenses

FPGA

retiming

support.

like to contact

Synopsys Sales

or

So Austin and Peter, let's just say the Xilinx track record stinks, and you will do or say anything to pump and dump your stock. Xilinx was quick to walk away from existing XC4K customer leaving them high and dry, and HARD obsoleting their current shipping products.

So, when Austin is doing this pump and dump crap:

it certainly didn't recongnize the investment it's customers made in XC4K product, why should we believe they will not walk away from any current product just as easily if they think it will save them a buck.

Open, lie.

It's the existing policy of walking way from XC4K customers when you can save a buck that I'm pretty sure will never change.

yep ... no commitment, just the allusion of commitment.

That might actually be true ... they might actually get product support right. Austin

Probably the truth is somewhere in the middle: Austin does tend to spin, and only see the rosy apple, whilst your example focuses very much on the worm.

The specific example here was a 3rd party tool flow issue, and quite a long time ago. As a 3rd party issue it is not entirely Xilinx's fault. It even seemed there was a design solution, just no longer free.

What we can see since then, is that the bigger FPGA vendors have expanded much of the tool flow in-house, and Xilinx have even done an in-house simulation tool (IIRC?)

EOL dead ends are relatively rare (even the example you quote is not a true dead-end, more an annoyance ) - other EDA sectors have more.

I'd say a far bigger cost issue in the FPGA sector, is Tool quality, and regression. I can understand their issues - the Silicon changes at a rapid rate, in EDA software terms, so they always have to juggle support for the hot-new-chips, with stability on older devices. Things like the open-source cable drivers, look like a very good move.

-jg

And I expect that few besides Mr. Lost mourned the passing of fpga express :)

-- Mike Treseler

Thanks, Jim. I appreciate a well-considered post like this. The discussion has a familiar, unpleasant ring to it and you've shown it in a decent perspective.

Oh, and I liked the sleeping bag instructions from Totally_Lost above:

Open, lie.

- John_H :-)