3 ms·

This was actually a bright-spot in Intel's lineup that had a chance to be a moonshot. I don't know why Intel is giving up this early especially when AMD acquire

by 1-6 3y ago

This was actually a bright-spot in Intel's lineup that had a chance to be a moonshot. I don't know why Intel is giving up this early especially when AMD acquired Xilinx recently. Is intel trying to emulate Nvidia? I don't think they should try to become another Nvidia. They should be building programmable AI chips with FPGA. Heck, LLMs are able to code Verilog. There are many many possibilities.

tester756 3y ago
Why "giving up"? It's just spinoff and their people are at executive levels
brucethemoose2 3y ago
> They should be building programmable AI chips with FPGA Eh, people have been saying this for over a decade, and I think that opportunity has passed. FPGAs are not going to beat GPUs anytime soon due to the software ecosystem (among other things), and ultimately they are not going to outrun ASICs that are now economically viable (especially in the embedded space).
- crotchfire 3y ago
  > FPGAs are not going to beat GPUs anytime soon due to the software ecosystem Completely true as long as the chip companies stonewall open-source toolchains. If Xilinx or Altera published the bitstream format for their highest-volume high-end chip (i.e. one or two generations back from the bleeding edge) you'd see the effect on the entire AI space within twelve months. I'm not kidding. The interest in this kind of access from developers is enormous, and the problems in this space are extremely regular. There are massive opportunities for FPGAs to avoid spilling to DRAM or even to SRAM -- you have an ocean of tiny register-speed memories in LUTRAM mode. But it will never happen. And so FPGAs will continue to be trinkets for weapons manufacturers and not much else.
  - someguydave 3y ago
    If they adopt open-source tool chains Xilinx & Altera would be signing their own death-warrants. The hardware would be reproduced by third parties (the open source code would contain all you need to know to produce the chips) and it would become a commodity like DRAM. Profit margins for the incumbent companies would go to zero and the users would profit immensely. Instead, FPGAs are sold to big entities like governments and big corps who can throw bodies at the crappy toolchains until they produce what they want.
- btown 3y ago
  Now I'm wondering about a world where OpenAI can treat their compute not as "dedicate this time slice of an A100 to generating tokens for customer A's query" but rather "there is a custom pipeline at a gate level that exactly matches our model architecture, hardcoding weights at the gate level, and can treat parallel requests the same way a processor would pipeline instructions." I don't think any human could write the TensorFlow compiler for this, but theoretically a generative AI could learn how to, experimenting on its own compute hardware? I don't know that we've seen the last of FPGAs.
MBCook 3y ago
They gave up on Xscale right before mobile demand shot to the moon.
contrarian1234 3y ago
They acquired Altera in 2015, which in the tech world is when the dinosaurs roamed the earth. How much more time would they need to see a profitable synergy? Bearing in mind they can probably see what's in the pipeline for the next couple of years. I'm not an expert... But in my naiive opinion it seems entirely reasonable to expect an acquisition to start paying off within ten years? Would love to hear counterexamples
- crote 3y ago
  A big issue is that sticking an FPGA onto the CPU die itself doesn't really make sense. Those are some really expensive transistors, and you're essentially wasting them when they are not being used for cores or cache. Besides, you'd be spending significant amounts of money building a niche product. It would be a lot more viable to use a chiplet approach: bake the FPGAs on a slightly-cheaper node, and just glue them right next to the CPU itself. Unfortunately Intel is still miles behind AMD when it comes to chiplets, and their tech is only just now barely starting to ship.
adrr 3y ago
Because intel messes up everything with their bureaucracy. Look at the NUC, they made it extremely hard to get because of their convoluted distribution strategy. They could have easily done direct to consumer and made it easy to get. Till they fix their culture problem, they should be unloading these business units and profit by holding majority stakes in them.

by 1-6 3y ago

tester756 3y ago

Why "giving up"? It's just spinoff and their people are at executive levels

brucethemoose2 3y ago

> They should be building programmable AI chips with FPGA Eh, people have been saying this for over a decade, and I think that opportunity has passed. FPGAs are not going to beat GPUs anytime soon due to the software ecosystem (among other things), and ultimately they are not going to outrun ASICs that are now economically viable (especially in the embedded space).

crotchfire 3y ago
> FPGAs are not going to beat GPUs anytime soon due to the software ecosystem Completely true as long as the chip companies stonewall open-source toolchains. If Xilinx or Altera published the bitstream format for their highest-volume high-end chip (i.e. one or two generations back from the bleeding edge) you'd see the effect on the entire AI space within twelve months. I'm not kidding. The interest in this kind of access from developers is enormous, and the problems in this space are extremely regular. There are massive opportunities for FPGAs to avoid spilling to DRAM or even to SRAM -- you have an ocean of tiny register-speed memories in LUTRAM mode. But it will never happen. And so FPGAs will continue to be trinkets for weapons manufacturers and not much else.
- someguydave 3y ago
  If they adopt open-source tool chains Xilinx & Altera would be signing their own death-warrants. The hardware would be reproduced by third parties (the open source code would contain all you need to know to produce the chips) and it would become a commodity like DRAM. Profit margins for the incumbent companies would go to zero and the users would profit immensely. Instead, FPGAs are sold to big entities like governments and big corps who can throw bodies at the crappy toolchains until they produce what they want.
btown 3y ago
Now I'm wondering about a world where OpenAI can treat their compute not as "dedicate this time slice of an A100 to generating tokens for customer A's query" but rather "there is a custom pipeline at a gate level that exactly matches our model architecture, hardcoding weights at the gate level, and can treat parallel requests the same way a processor would pipeline instructions." I don't think any human could write the TensorFlow compiler for this, but theoretically a generative AI could learn how to, experimenting on its own compute hardware? I don't know that we've seen the last of FPGAs.

MBCook 3y ago

They gave up on Xscale right before mobile demand shot to the moon.

contrarian1234 3y ago

They acquired Altera in 2015, which in the tech world is when the dinosaurs roamed the earth. How much more time would they need to see a profitable synergy? Bearing in mind they can probably see what's in the pipeline for the next couple of years. I'm not an expert... But in my naiive opinion it seems entirely reasonable to expect an acquisition to start paying off within ten years? Would love to hear counterexamples

crote 3y ago
A big issue is that sticking an FPGA onto the CPU die itself doesn't really make sense. Those are some really expensive transistors, and you're essentially wasting them when they are not being used for cores or cache. Besides, you'd be spending significant amounts of money building a niche product. It would be a lot more viable to use a chiplet approach: bake the FPGAs on a slightly-cheaper node, and just glue them right next to the CPU itself. Unfortunately Intel is still miles behind AMD when it comes to chiplets, and their tech is only just now barely starting to ship.

adrr 3y ago

Because intel messes up everything with their bureaucracy. Look at the NUC, they made it extremely hard to get because of their convoluted distribution strategy. They could have easily done direct to consumer and made it easy to get. Till they fix their culture problem, they should be unloading these business units and profit by holding majority stakes in them.