Altera or Xilinx

For low-end projects that need things like ethernet or other high-end macros I would look at Lattice. I haven't had the opportunity to use any of their hardware or software but it seems that they offer more to lower-end customers.

---Matthew Hicks

Lattice! How could I forget? Used their CPLDs (isp1032?) some ten years ago. It indeed looks like they have some nice devices and tools seem comparable although the free version is only for evaluation (no sim?). Payed tools seem a bit cheaper than actel/xilinx though.

Any experiences with device and tools here?

No, those are the two big fish. There are other smaller FPGA vendors such as Lattice and Actel.

How long is a piece of string? Should I buy Intel or AMD processors? A GM, Ford, or Toyota car? At the end of the day, it doesn't matter which you use as long as it gets the job done and is cost effective.

If you really think that all else is equal, decide on the basis of price. It seems unlikely that there is no difference in price between the smallest devices that meet your requirements.

Eric

If your local distributors are really equivalent, then the only significant difference to me is vhdl synthesis.

I prefer Rob Dekker's vhdl front end to brand X. The numeric_std library also gets better coverage in the docs and the rtl viewer is cleaner.

However, if I preferred verilog, or synopsys style vhdl, or if I just wanted to wire up some cores, then it would still be a wash.

-- Mike Treseler

I'm under the impression that Altera parts are much easier to obtain in small quantities. Most 'hobbiest' projects seem to use Altera parts. Every time I buy Xilinx parts, I'll have to meet minimum order values so sometimes I must buy more devices than I actually need.

Also if you really want to push an FPGA to its limits (space/speed) it is almost impossible to write FPGA independant code. So whatever you choose, you'll probably get stuck to it because changing vendors will require re-learning the quirks, tricks and basic fpga elements.

On the other side of the coin, it is easy to write FPGA independent code if I'm not near the limits and I study the synthesis templates and learn some verilog or vhdl.

-- Mike Treseler

I've bought small quantities of Xilinx parts from DigiKey.

-Dave Pollum

Xilinx (and Actel?) have free linux version of their programming tool. Which Altera doesn't.

You can download the Altera Quartus II Programmer for free.

Petter

The download page says:

"Quartus® II Stand-Alone Programmer Version 7.1 Service Pack 1 Windows XP and Windows 2000 (Note: Solaris, Linux, and HP-UX are not supported)"

-- Mike Treseler

customers.

I've used the Lattice ECP and ECP2 series, although I haven't had the pleasure of using ECM2/M yet (the ones with the high-speed serial I/O). We bought the tools for a few hundred dollars. The front-end for synthesis is Synplify or Precision (your choice, they're both bundled in). If you're not used to anyone elses user interface yet there's probably no reason to pick Lattice vs Xilinx vs Altera. Many of those posting here gave up on the GUI front-ends long ago. That being said, a lot of Lattice flow is similar to Xilinx, mostly because the tools both originated at NeoCad. However they're just different enough that you'll need to learn the quirks, especially if you need to push the limits of the part. On the back-end, I'd say Xilinx is well ahead on the mapping and place&route, but again you can get good performance from Lattice if you know what to tweak and are not afraid to use multipass place and route. I haven't installed Lattice version 7 tools yet, but there's talk of improvement on the back end for that release.

Both Lattice and Xilinx bundle branded versions of ModelSim with the paid tools. It may just be my imagination but it seems that the Xilinx version runs slower than the Lattice version. I think the Xilinx version inserts waits if your design exceeds a certain number of statements or non-Xilinx "leafs". I haven't run into that wall on the Lattice version, but I'm not sure if it's due to differences in the license or in the size of my designs. If you're going to do large designs and long simulations or if you do mixed language designs, I'd suggest getting a license for ModelSim PE.

Lattice surprised a lot of people by releasing the low-cost ECP2/M series when Xilinx and Altera only offered similar features in their higher end products. That is no longer the case, and I would expect both Xilinx and Altera to beef up their low-cost offerings if they see erosion in their customer base due to Lattice.

For many designs also realize that the I/O options are becoming the limiting design criteria rather than the FPGA fabric. Lattice parts have some nice features for using DDR I/O, but again you need to look at the quirks to see if it does what you want for your design.

HTH, Gabor

Thank you all for your input. To sum it up:

Altera and Xilinx are indeed the major brands to look at. Lattice has some nice stuff, but is smaller and does not support VHDL (at least not in the low-end tools).

There are no real advantages of one over the other (maybe there is for some specific designs). So the choice should be made based on the less technical stuff like distributors, support, prices.

I've asked purchasing to get some indication qoutes of the low-end devices, eval kits and tools from both. The Xilinx distributor got back the next day, still waiting for Altera. :-(

Xilinx seems to have a bit nicer eval kits, especially if we need ethernet. This is not available on the low-end Altera eval kits.

On the other hand, the sales rep for our altera distributor used to come in for a cup af coffee on a regular basis in the old days, and he's still there.

As my most current and extensive experience is with Xilinx (CPLD), I am currently leaning to Xilinx, but no disicion yet.

Sorry, I missed the Linux part. As a Linux user I should have noticed...

Petter

Consider running a known-good design through both sets of tools and getting quotes from both vendors for 5000 pieces of the most likely device for delivery one year from now.

-- Mike Treseler

Are you sure ?

The web page states this : [Supported HDL languages include; VHDL, Verilog 1995, Verilog 2001.]

and they release the Mico8/Mico32 in both Verilog and VHDL.

You can download their Starter for free, and try a design.

-jg

This really irks me...."no real advantages"... I will not get into a propaganda tirade, but rest assured that there are significant technical differences between the two competitors. I am of course biased, and much more familiar with the Xilinx devices and their special advantages. And -as promised- I will not elaborate them here. There are differences, for better or worse, between Obama and Clinton, between the US and Canada, between BMW and Mercedes, between Apple and Microsoft, and also between Xilinx and Altera. You may prefer one brand over the other, but don't call it a toss-up! We worked hard to make them different. It is only when you ignore all the interesting and important fine details that everything looks the same grey mush. "Once you have seen one lake or one mountain, you have seen then all. California is just like Ohio with some mountains and beaches thrown in!" Nonsense! Peter Alfke

Peter,

As a long time Xilinx user and fan, as a convert from Altera (first by need, then by price and familiarity) I see both points.

If there's a niche algorithm that makes or breaks the design or if there's a critical path that's impossible to implement in one vendor's devices, then the differences are extreme. The silicon from each vendor has its sincere strong points in certain corners. No one vendor can declare total domination, however, just greater market share.

As an engineer tasked with coming up with the best price/performance solution to meet our Quality, Cost, and Deliverable (QCD) requirements, I choose one vendor over another. A design can be implemented in so many ways that a good engineer can come up with a superb implementation in one vendor's silicon that won't pass muster in the other. But turn it around and require the engineer to target the other device and that same implementation will be tweaked to the new silicon and not run so well on the original "vendor of choice" silicon.

If any vendor stopped innovating, the market share would dwindle for that manufacturer and the whole industry would suffer; without the competition, the drive to excel is lessened. Because of the continuing innovation, the solutions stay close to the same level of performance with some algorithms or implementations winning hand over fist in one brand of silicon versus the other.

QCD targets have to be met. If cost were identical, the quality factors can come into play. If new devices are promised but not delivered, the project's deliverables are in jeopardy. It's all an engineering trade-off. If Xilinx had only free parts with no delivery or software issues to consider, I would only use Xilinx.

Any good engineer can produce an average design that meets the quality requirements. It's only the exceptional designs (or the sub-par engineers) that fail in this task.

If I had a QCD requirements that made Ohio the "right" choice, I'd hope the experience could be just as fine in the end as the California story.

I appreciate your passion.

- John_H

Peter,

I'm not being a wise guy by an stretch. I am seriously inquiring for my own edification.

Name some things that Xilinx can do that Altera can't or more specifically, is there an application out there in the world today that can only be solved with a Xilinx part? Would it be a fair statement to say that >90% of the designs out there can be done effectively with either company and the real advantages lie more with:

1. Service
2. Tools
3. Price
4. Delivery
5. Familiarity: meaning that once someone starts with a certain vendor, uses them for a while, and understand the hardware and how to flex it, they're less likely to switch--especially since most of us are put on ridiculous delivery schedules.

Again, not trying to raise ire, just humbly asking a question.

Best regards, Rob

The temptation is there to answer with: FIFO, IDELAY, SRL16, DSP48 efficiency, PPC, etc., but that would just invite an endless series of acrimonious responses. Hell, there are many people who do vote for the party that I despise, and lots of people who seem to prefer the PC over the Mac. Just don't tell me there are no differences in user friendliness and elegance...

FPGAs (and operating systems) are not (yet) a commodity, where only price and availability counts. Let's hope it stays that way for a while. Peter Alfke