I am looking at using the SAM9263 and noticed that they use a 16.36766 MHz crystal rather than the 18.432 they use on the SAM7 eval boards. Anyone know why they picked this value? Do they use this value on the other SAM92xx eval boards?
In general, what are your experiences with these parts?
Anything is possible, but the two frequencies are not related by an integer combination I can find. This seems odd to me since they specify the frequency so exactly to 7 decimal places. This is about the limit of what a reasonable crystal can provide.
I also noticed that the eval boards for the SAM9 devices are around $1,000 US. I think the only one significantly less is for the SAM9260 which is about $600. Why are they so expensive?
It is really volume dependent and it helps negotiating with your CEM. If you use the same components, on a low volume board as on a high volume board then you still get the same price of components.
I looked at the cost of producing a board, and if you have a low volume, then you run into problems with minimum order value of many components, and at AT91RM9200DK volumes that about doubled the price of the components alone.
The AVR Gateway was designed to be low cost and will be produced in numbers where minimum order volume will not come into play.
Atmel has been reorganised so that the AT91 team and the AVR team is now in a single organisation, and it is not unnatural to expect that they will try to become more efficient producing development boards.
When you have STK500 volumes, you are a little bit more leverage, than when you only have AT91RM9200DK volumes
Why would the AT91RM9200DK have been planned for low volume? Are you saying that the AT91SMA926x-EK boards are expected to be low volume? Of course this is self fullfilling prophesy when you price one board at $69 and the other at $1000! At $69 people will buy it just to play with it and see how it runs even if they don't have a need. Even better would be a design contest. Is Atmel not marketing the ARM9 parts as hard as they are the AVR32?
Of course not. I have several project in 100ku or higher volume.
Surprisingly we sold 100s of the AT91RM9200DK at $5000 just in my region. Sales volume improved with the AT91RM9200EK but not dev tools revenue. You do have people thinking a lot before making that investment but I am only aware of one project which I lost mainly due to toolcost. The decision was delayed and delayed and then something happened which I believed would not have happened if they have had the tools in house.
I bet that people coming to Atmel Seminars will be able to go home with an AVR32 kit. This is by far the best way to distribute the kits.
The ARM9 parts at least get a lot of *my* attention. I quite often go into a customer which has *almost* decided to go with an LPC; and then show the SAM9260.
If a customer is planning to use external memory, then it will be very hard for them to resist that little goodie. The pin compatible flash version, the SAM9XE, should make NXP and ST cry and trash their performance graphs. While the flash is not significantly faster, you can loop in the cache, and use the TCM for 4-5 x performance boost.
NXP does not have anything which comes close to the 9260 The LPC3xxx does not have ethernet and a lot of other goodies present on the 9260 and the ST ARM9 part is a two chip solution which should not have too much higher performance than an ARM7.
What??? You expect to sell 100k units of the AT91SMA926x-EK boards??? How can you call that low volume??? I think you are confusing the EK with the chip.
That is not the only measure of the business you can loose from having high priced evaluation tools.
BINGO! I can assure you that there is very little difference between the LPC2xxx and the SAM7xx parts. But much of the LPC business came from the very low cost units that are out there. I watched it grow and much of it was due to the feedback between the availability of cheap eval tools and the grassroots popularity of the chips. Each one fed the other with very rapid growth. There are still any number of vendors who only provide LPC eval boards and not Atmel. I only know of one vendor who provides Atmel support and not NXP.
You may get wins at the customers who show on your radar. But there are any number of customers who select a part for a design before you know anything about them and it is not infrequent that these projects are with large customers. I know because I have seen it happen. Decisions are made without input from the vendor largely based on what the engineer is familiar with. Low cost eval tools help a great deal in getting the customer familiar with the parts with a minimum of management review.
So you are saying that you don't provide low cost solutions because you don't need to?
Reading too carelessly. The SAM9260EK is low cost from the AT91 point of view. This is based on their culture which is coming from the ASIC world. The AVR32 guys are coming from the 8 bit world and have a different culture.
I expect that the reorganisation moving both groups under one hat, will result in cross-pollination. This should result in lower cost AT91 tools.
I know, but if the processor is right for the job, even $5000 is not a big deal for the right project.
I think that you find that the majority of chip sales are made to very few companies, and those companies will get what they need for development.
It was important for the AVR and is important for the AVR32 to promote themselves using low cost tools, because they need to have a lot of users to make it interesting for tools makers to support the parts. The AT91 team does not have to invest in promoting the ARM architecture.
The approach to have cheap tools means that you can win project where basically any part will do. When you have an edge in the chip, people are prepared to pay for the tool.
Where it is very important, to have low cost tools, I.E at consultants who need to be familiar with tools in anticipation of you probably have a discount program for tools.
I am well aquainted with your thinking, which I personally share. I am working hard to move Atmel to a lower cost structure for ARM tools.
I think that Atmel can live with $500 board for the ARM9 at the moment but the reason is really the underlying cost structure. If/when competition catches up, then this will become a more important decision factor.
I think that prices will come down before they are needed for this reason though.
A large part of that is NXP was first with single chip FLASH, whilst for a while all Atmel had were the external memory ARMs.
I think Atmel's "release rate" is now above NXP.
Yes, and the $69 AVR32 kit, looks a good way to do that.
Another important aspect of such low cost Eval kits, is as "hardware data sheets"
- as they come with tools, it can be quicker to load and measure performance, than trying to find the info-detail you want, in all the literature sources.
Looks to me like they are being careful to release the FLASH models at exactly the same time: ARM9 and AVR32
Ulf said the ARM and AVR divn's were separate until recently, and you'll see the AVR32 is a bottom-up growth chip - that Divn has the AVR 8 bit volumes, and mindset, and they apply it to the AVR32 - with the new FlASH AVR32's I expect this will accelerate. The AVR32 part is still very new.
The ARMs from Atmel have been a top-down product. ASICs, then Microprocessors, and more recently Microcontroller versions - so that's a different mindset.
It takes time in large companies for these mindsets to cross-pollinate, and for the lower price kits to get on the radar.
Probably not much engineering cost to swap the AVR32 in their Eval PCB, to an ARM9 ?. Identical PCB with swapped CPUs could be a nice benchmarking tool...
I won't argue all the little details even though I disagree. But your ARM9 eval boards are not $500. I would consider that marginally acceptable. But they are much higher. The single lowest cost board is $600, the rest are $1000 or more.
You (or your company) assume that your chip will get the chance to be evaluated on all major projects. There are any number of engineers who pick parts based solely on familiarity even for large jobs at large companies. My last position required that I learn to push vendors away when I did not need info. I learned that all the larger companies learn to do that.
Another factor is the threshold of approval. If I need a manager to sign off on a purchace it makes it that much harder to do. If I need a second level manager that means I typically have to justify it in writing. Likewise, if I ask my rep for a freebie, it is much more likely to happen with a $200 board (I got those often) than with a $1000 or even a $600 board.
My point is that the view from inside your company is not necessarily the same as the view from inside the customer's company.
Yes, this is the result of the high cost base. Need volume, and I have been working on a proposal which will allow volumes to increase, and thus price should go down.
I know that I don't get a chance to be evaluated on all major projects. I also know that if I get a chance to discuss with the customer before decision, Atmel gets about 50-75% hit rate on AT91 parts. I.E: visits results in design wins more often than not. Quite often it is due to selling ARM9 vs ARM7.
I am well aware of the TV-Shop paradigm...
If you get a kit free of charge, do you really care if it cost $200 or $5000? You assume that pricey kits cannot be had free of charge. I dont understand why not, if the business case is right.
The problem only occurs if you don't get the kit free of charge.
Well, nearly. I'd imagine Rick and many others would be less keen to accept a $5000 'freebie' than a $69 'freebie'
- why if they are both Free, you ask ? Well, there is clearly a higher 'obligation' on the $5000, and so the 'brush off' is harder.
Some companies may even have firm policies on such things, to prevent 'abuse by leverage' cases, from either end of the sales chain.
So a $69 kit, grabbed from the discretionary monthly spend budget via Digikey (et al) gives a nice low profile way to get a real working device on your bench - and real devices are becomming more important, (as I mentioned before), where the data is falling behind....
No, it is not an issue with "accepting" an expensive kit. It is the willingness of the vendor to offer. My point is that the more familiar with the devices the engineer is, the more likely he is to use the part. If a project is in the early stages and an eval board is provided, that part is much more likely to be used. The early stages of any project carries a lot of risk, including spec changes that eliminate a type of processor and even cancelation of that portion or the entire project. My experience is that salesmen weigh offers for cost vs. benefit and risk. So with a significantly less expensive kit, the cost factor makes it possible to toss them around to engineers if they just say the word ARM vs. having to weigh the likelyhood of a project taking off and selecting the part in question.
I have seen this in action. It is not uncommon that engineers consider parts from different vendors to be equivalent and base their selection on other factors, availability of tools, support from disits and the manufacturer and even perceived cost of project startup. Having a eval board in hand is one of those things that provides a natural slant, even more than taking the engineer to lunch ;^)
The raisonance kits seem to be a good compromise. They have a standard motherboard which they presumely can manufacture in higher volumes. They then have a memorystick type CPU board which plugs into this motherboard. A different stick for each type of processor. When I was looking for an STR9 dev kit, theirs were by far the best value for money.
The STR9 does not have a MMU, and runs at < 100 MHz, No cool Linux port available.... Have to be cheap ;-) - 89 Euro / single qty.
You can get the real thing from
The I/O tech has 8 MB Flash and 16 MB SDRAM and sells for 68 Euro@1k You can get it with up to 16 MB flash and 64 MB SDRAM There is also an FPGA option with Spartan III (up to 1.2 Mgates) and another 64 MB SDRAM, all for 149 Euro@1k
The fact that low cost modules are available, and people still buy the more expensive dev kits is interesting.
The new AVR32 board at $69 should be real good value. This is not just a module, it has a lot of connectors and expansion possbilities.