AMD to Increase Xilinx FPGA Prices by up to 25%

[AleksandarK]

Xilinx Field Programmable Gate Arrays (FPGAs), now part of AMD, are always in demand in the semiconductor industry. Today, AMD has shared a letter to Xilinx customers that the selected FPGA device series will receive an 8-25% price increase. Citing AMD's investment into the supply chain, along with increased prices from the suppliers, Xilinx FPGAs will get more expensive. From January 9, 2023, the cost of the Spartan 6 series will increase by 25%, the price of the Versal series will not increase, and all other Xilinx products will increase by 8%. Interestingly, the older series manufactured on 40-28 nm nodes will increase while the latest Versal series doesn't experience any change.

Regarding lead times, the 16 nm UltraScale+ series, 20 nm UltraScale series, and 28 nm 7 series all take 20 weeks from order to delivery, which will remain until the third quarter of 2023. You can read the entire document below.



View at TechPowerUp Main Site | Source

 

[Bloax]

trololo su tax activated

 

[usiname]

trololo su tax activated

Trolololo, clueless intel fanboy

Intel issues price increase notice for PSG product lines - SemiMedia

www.semimedia.cc

 

[ZoneDymo]

Still dreaming of a future where AMD GPU's have FPGA's that are programmed to run the game of that moment the best, be it RT requirement or Rasterization or both.

 

[Steevo]

Still dreaming of a future where AMD GPU's have FPGA's that are programmed to run the game of that moment the best, be it RT requirement or Rasterization or both.

Too much wasted silicon space, once we master optical interfaces I expect general computing to be a dumb motherboard with optical/power interfaces that we choose the hardware acceleration items we want.

Master optical input of whatever CPU architecture you are running, storage interface is built onto the sold state or mechanical storage unit, memory is built onto the CPU “card”, GPU is essentially the same, generic audio or whatever card you want. Sub processors for dedicated tasks, maybe a few stackable add on cards for FPGAs.

Cheap motherboards as they are essentially power hubs with optical backbones and a few data traces and minimal I-O.

 

[TheinsanegamerN]

Any excuse to price gouge, I guess.

 

[Chomiq]

trololo su tax activated

That's DE10 Nano from Intel. Risen in price by about 300% since 2019

 

[bug]

If Nvidia did something similar, we'd be on the 6th page of "ngreedia this" and "ngreedia that". Just sayin'.

 

[Assimilator]

What fab does AMD use for Xilinx? I know that TSMC increased node prices back in 2021 and the older nodes were hit hardest, so maybe this is just another round of these increases.

It makes sense for a fab's business model, the newer nodes are costing more and more due to the ever-more-expensive EUV machines from ASML, and since the older nodes are predominantly the highest volume the fabs see them as an easy way to make cash to put into the newer nodes.

 

[Aquinus]

Too much wasted silicon space, once we master optical interfaces I expect general computing to be a dumb motherboard with optical/power interfaces that we choose the hardware acceleration items we want.

Master optical input of whatever CPU architecture you are running, storage interface is built onto the sold state or mechanical storage unit, memory is built onto the CPU “card”, GPU is essentially the same, generic audio or whatever card you want. Sub processors for dedicated tasks, maybe a few stackable add on cards for FPGAs.

Cheap motherboards as they are essentially power hubs with optical backbones and a few data traces and minimal I-O.

The size of the transceivers would need to be miniaturized to make that viable which is going to reduce output power of the optics all while being able to have to essentially etch optical cables on motherboards. I have absolutely no idea how hard of a physical problem that is to solve, but I can't imagine that it'll be solved any time soon. Optical is definitely better for longer distance I/O given the advantages over electric signaling. Not so sure about short distances though.

 

[Wirko]

the older nodes are predominantly the highest volume the fabs see them as an easy way to make cash to put into the newer nodes.

It's not just that. TSMC is actively pushing their customers to newer nodes, which means 28 nm or finer. They want to make space for new tech in their old clean rooms. Planar 28 nm is often considered "the last cheap node", not just for manufacturing but also for designing chips.

The size of the transceivers would need to be miniaturized to make that viable which is going to reduce output power of the optics all while being able to have to essentially etch optical cables on motherboards. I have absolutely no idea how hard of a physical problem that is to solve, but I can't imagine that it'll be solved any time soon. Optical is definitely better for longer distance I/O given the advantages over electric signaling. Not so sure about short distances though.

Another big issue is that optics can't do any meaningful signal processing. It's not possible to make a purely optical hub for optical signals which would be able to do even the simplest kind of switching, routing, filtering, or data buffering. Everything must be done in electrical circuits, with lots of fast analogue electronics (which resists scaling to smallest nodes) and lots of electro-optical converters.

That's DE10 Nano from Intel. Risen in price by about 300% since 2019

I don't know those, so can't say if the price is palatable, but the lead time definitely isn't. Do you have to pay in advance?

 

[Unter_Dog]

AMD pushed me to move away from Spartan 6 months ago due to future cap

What fab does AMD use for Xilinx? I know that TSMC increased node prices back in 2021 and the older nodes were hit hardest, so maybe this is just another round of these increases.

It makes sense for a fab's business model, the newer nodes are costing more and more due to the ever-more-expensive EUV machines from ASML, and since the older nodes are predominantly the highest volume the fabs see them as an easy way to make cash to put into the newer nodes.

I know they used to FAB on Samsung 45 nm.
I moved to ARTIX-7 based on heads up from AMD, which is cheaper and more capable but like all new FPGAs has power sequence requirements and more supply rails which are always a pain. This is on 22nm

 

[bug]

What fab does AMD use for Xilinx? I know that TSMC increased node prices back in 2021 and the older nodes were hit hardest, so maybe this is just another round of these increases.

It makes sense for a fab's business model, the newer nodes are costing more and more due to the ever-more-expensive EUV machines from ASML, and since the older nodes are predominantly the highest volume the fabs see them as an easy way to make cash to put into the newer nodes.

Afaik, FPGAs are not made on cutting-edge nodes. But I may be wrong about that, maybe @TheLostSwede could pitch in?

 

[TheLostSwede]

Afaik, FPGAs are not made on cutting-edge nodes. But I may be wrong about that, maybe @TheLostSwede could pitch in?

Depends on the chip, but generally no. Intel seems to be a bit ahead.

FPGAs & 3D ICs

AMD offers the comprehensive multi-node lineup of FPGAs providing advance features, low-power, high-performance, and high value for any FPGA design.

www.xilinx.com

Altera® FPGA Products - FPGA and SoC FPGA Devices and Solutions

Altera® FPGA devices offer a wide variety of configurable embedded SRAM, high-speed transceivers, high-speed I/Os, logic blocks, and routing.

www.intel.com

 

[Steevo]

It's not just that. TSMC is actively pushing their customers to newer nodes, which means 28 nm or finer. They want to make space for new tech in their old clean rooms. Planar 28 nm is often considered "the last cheap node", not just for manufacturing but also for designing chips.


Another big issue is that optics can't do any meaningful signal processing. It's not possible to make a purely optical hub for optical signals which would be able to do even the simplest kind of switching, routing, filtering, or data buffering. Everything must be done in electrical circuits, with lots of fast analogue electronics (which resists scaling to smallest nodes) and lots of electro-optical converters.


I don't know those, so can't say if the price is palatable, but the lead time definitely isn't. Do you have to pay in advance?

I don’t think we will solve the light in a bottle for optical storage, but the transmission speeds and different slots using and sharing a different wavelength of light simplifies a lot of those issues.

All optical needs to do is become simpler and cheaper than copper traces and reduce chip complexity and traces, termination and other issues are already requiring additional SMCs next to PCIe slots.

 

[Wirko]

What fab does AMD use for Xilinx? I know that TSMC increased node prices back in 2021 and the older nodes were hit hardest, so maybe this is just another round of these increases.

It could also be GlobalFoundries. They still have an agreement with AMD, valid until 2025.

Of course, GF is raising its prices too, by (you guessed it) 8%.