AMD/Xilinx announced Spartan UltraScale+ family

[asmi]

https://www.xilinx.com/products/silicon-devices/fpga/spartan-ultrascale-plus.html#productTable

It covers the range from 11K cells to 218K cells, so pretty much entire existing range of Spartan-7 and Artix-7. I found a couple of things curious - first off, some of higher-end parts will have MGTs and PCIE4 blocks (up to PCIE4x8 or two PCIE4x4), which means that they will compete directly with Artix US+ devices. Also, in addition to typical soft memory controllers for DDR4, some parts will contain hardened controllers for LPDDR4 and LPDDR5 with bandwidth of up to 4266 Mb/s. And of course they are going to have support for all modern physical interfaces like MIPI.

Looking at the specs, I see a potential that Spartan US+ (what an acronym! ) will make Artix US+ obsolete, but of course the pricing will be crucial here. Current Artix US+ begin (XCAU10P) at about $200 per device, so it would be interesting to see how they will price new devices.

Oh, and also forget about convenient 1 mm pitch BGA packages - they are no more (except for FSVG1156 with 1156 balls), and even 0.8 mm begins with 529 balls, with smaller packages having 0.5 mm pitch. In general, these devices have A LOT of IOs, the smallest package has 361 balls.

[glenenglish]

and no medium density 0.8mm easy to use packages in the 200-350 ball range. just giant ones  or tiny ones.

I have complained... made suggestions....

[colorado.rob]

By only supplying parts in such fine-pitch BGAs, they are gently telling price-sensitive customers "shop elsewhere."

[SiliconWizard]

Yes, but do you really think price-sensitive applications would use the UltraScale+ family to begin with?

[asmi]

Yes, but do you really think price-sensitive applications would use the UltraScale+ family to begin with?

It depends on the level of price sensitivity I guess. Or their strategy regarding "old" families (7 series, US). I remember when Xilinx really wanted customers to move from S6 to 7 series devices, they priced them to be cheaper than corresponding 6 series device. If they would want to do the same here, they might price them quite aggressively. Though I personally doubt it.

[SiliconWizard]

Well, unless they they become *very* aggressive specifically with this Spartan US+ series, judging from the current price for the Artix-7 US+, I doubt it as well.

[chickenHeadKnob]

I don't see any logical reason they would cannibalize the sales of other product lines. Corporate managers are averse to doing this. When that kind of thing happens it is because competition has forced them through loss of market share.

[ali_asadzadeh]

Good news, I think Chinese FPGA makers cause them this direction, and I think we could get them cheap enough soon, prices like 100~200$ is insane for these and now one would buy them for real products, my estimate is 5$ for the lower ends to 50$ for higher ends, like spartan 6 device prices

[langwadt]

Yes, but do you really think price-sensitive applications would use the UltraScale+ family to begin with?

It depends on the level of price sensitivity I guess. Or their strategy regarding "old" families (7 series, US). I remember when Xilinx really wanted customers to move from S6 to 7 series devices, they priced them to be cheaper than corresponding 6 series device. If they would want to do the same here, they might price them quite aggressively. Though I personally doubt it.

I'm sure the reason they really wanted people to 7 series instead of S6 was that they didn't want to add S6 support to Vivado, and didn't want to support ISE

[asmi]

Good news, I think Chinese FPGA makers cause them this direction, and I think we could get them cheap enough soon, prices like 100~200$ is insane for these and now one would buy them for real products, my estimate is 5$ for the lower ends to 50$ for higher ends, like spartan 6 device prices

Keep dreaming

[asmi]

I don't see any logical reason they would cannibalize the sales of other product lines. Corporate managers are averse to doing this. When that kind of thing happens it is because competition has forced them through loss of market share.

I don't think they really care about low-end market to be honest. Their cash cows are space, comms and military, and those folks tend to not care about component prices as much.
Just look at what they called "cost-optimized portfolio" and you will see that they have no idea what cost optimization is in real market outside of customers with blank cheques.

[ejeffrey]

By only supplying parts in such fine-pitch BGAs, they are gently telling price-sensitive customers "shop elsewhere."

It's really only price sensitive *and* low volume that cares about that.  It's not surprising that they don't cater to that market.

Also if you are low volume you can probably use one of their partner SOMs.  At least for what they call "midrange" and above, you can get an entire SOM with voltage regulators, DRAM, flash memory, and board to board connectors in single unit quantities for less than the FPGA itself in unit quantities.  Obviously in higher volumes you don't actually pay the digikey price, and for "low end" parts like the spartan US+ the cost of the SOM parts is a bigger fraction but it still can be an attractive option for low volume.

[asmi]

It's really only price sensitive *and* low volume that cares about that.  It's not surprising that they don't cater to that market.

Even that becomes less of a problem thanks to the folks like JLCPCB, which make higher-layer count PCB tech more accessible. If someone would tell me two years ago that now I can have 10 layer PCBs manufactured with via-in-pad for ~$200, I wouldn't believe it, and yet here they are. The only thing that's holding me from going all-in to 0.5 mm BGAs is their lack of support for 0.08 mm traces, which is a requirement for fitting a trace between via field, and/or of course blind-buried vias, but I can't see how they can implement this with a shared-panel approach.

Also if you are low volume you can probably use one of their partner SOMs.  At least for what they call "midrange" and above, you can get an entire SOM with voltage regulators, DRAM, flash memory, and board to board connectors in single unit quantities for less than the FPGA itself in unit quantities.  Obviously in higher volumes you don't actually pay the digikey price, and for "low end" parts like the spartan US+ the cost of the SOM parts is a bigger fraction but it still can be an attractive option for low volume.

SOMs have their own problems in a sense that they limit what you can implement. If your application happens to fit with what SOM offers, that's good, but in my experience most of my designs relied heavily on the fact that I can customize memory interface parameters (bus width, capacity), in one case I actually needed two separate DDR3 interfaces (which you will never find on a SOM). And of course fully custom designs tend to be more compact (high-speed connectors are typically by far the largest parts on a board, so you have a ton of PCB space by just NOT having them), which sometimes is also important.

[nctnico]

I don't see any logical reason they would cannibalize the sales of other product lines. Corporate managers are averse to doing this. When that kind of thing happens it is because competition has forced them through loss of market share.

I don't think they really care about low-end market to be honest. Their cash cows are space, comms and military, and those folks tend to not care about component prices as much.
Just look at what they called "cost-optimized portfolio" and you will see that they have no idea what cost optimization is in real market outside of customers with blank cheques.

Xilinx' prices are highly flexible. In some cases they'll slash 80% off the list price of a 4k dollar FPGA just to get the project.

[SiliconWizard]

Yes, but do you really think price-sensitive applications would use the UltraScale+ family to begin with?

It depends on the level of price sensitivity I guess. Or their strategy regarding "old" families (7 series, US). I remember when Xilinx really wanted customers to move from S6 to 7 series devices, they priced them to be cheaper than corresponding 6 series device. If they would want to do the same here, they might price them quite aggressively. Though I personally doubt it.

I'm sure the reason they really wanted people to 7 series instead of S6 was that they didn't want to add S6 support to Vivado, and didn't want to support ISE

That was probably at least part of the reason, yes.
But while initially they "gently" pushed people to the 7 series, now the pricing of the 6 series has become so absurd that it's absolutely impossible to keep producing devices with it unless you have a lot of stock, or you have secured a tight agreement with XIlinx years ago. (The third option being probably to buy it from dodgy asian sources.) Or, of course, if it's part of a very expensive device in which the FPGA is only a tiny fraction of the cost, even at the current prices. That basically is just niche products.