BTW, I love the idea of running SERV from SPI flash. I was working on a SPI-oriented CPU a while back, but became disenchanted because immediately-useful projects like SERV already existed.
Running SERV code directly from SPI is actually one of those things I haven't gotten around to do yet. It would be really cool to see if we could feed instructions serially into SERV from an SPI memory
@@p23q Just found this comment and had to try it myself. And you're right. Doesn't work on the LP384. Main problem is that there is no on-chip RAM on these devices. Does look like it's possible to fit three cores in the lp1k though
so, if iCE40 can hold 8 cores at 50Mhz, and it costs 38 bucks, then you need 36 cores to match the performance of a single core on ARM A53 processor (1.4-1.8Ghz) which is used in Raspberry Pi, but there are 4 of them, so you would need 36 x 4 = 144 cores. So 144 / 8 = 18 , you would need 18 FPGA boards of $38 each (and that would be $684 bucks) to match the performance of Raspberry Pi which costs around $35 bucks in retails stores and $15 if bought from China in bulk. Now, these Raspberry Pi processors have DDR3 , so I better forget about this comparison as it would obviously be in favor of today's ARM RISC processors, which will be hundred of times faster than today's FPGAs. You would need to make 500 cores of 50Mhz each on each FPGA to compete with today's powerfull ARMs , and that would require an FPGA with around 1.5M or 2M LUTs , and these devices cost thousands of dollars. Moral of the story: FPGAs are completely useless and not competitive with today's ASICs , they can only be used as toys for playing on the weekends.
Absolute Zero my friend FPGAs can fit cpus with custom instructions, that the user define, to make an ARM with a custom instruction you will need millions, unless you’re shipping millions of them you cannot justify it, and if you can’t make it to the market by the time you need, then that’s an FPGA use case, There’s a FPGAs in your phone in your motherboard, in your laptop, and right now in the cloud
FYI, you are comparing mcus to mpus. Fpga has its place. When you don't have 100s of millions to fork out on a tape out, you use fpgas. How do you think arm or Intel development happens? These are initial development. Arm was also expensive when it came out. It fell from $20+ to about $0.5 over time. Cortex m4 was over $5 just 2 years ago. Today it's < 1.
The ARM license joke was gold
Too bad that the audience was eager to go for lunch. All the SERV names are awesome! Kudos to the presenter.
Well, to be fair, I didn't have much more to say so we were all happy to have lunch :)
BTW, I love the idea of running SERV from SPI flash. I was working on a SPI-oriented CPU a while back, but became disenchanted because immediately-useful projects like SERV already existed.
Running SERV code directly from SPI is actually one of those things I haven't gotten around to do yet. It would be really cool to see if we could feed instructions serially into SERV from an SPI memory
This guy has good humour.
First time ever I see a "cool" electronics engineer
So again: which year is going to be RISC-V year? ;)
So you could fit one of these into an LP384? Wow!
Actually not. I tried it, and its a bit too small after all. No periphery anyway then. :-)
@@p23q Just found this comment and had to try it myself. And you're right. Doesn't work on the LP384. Main problem is that there is no on-chip RAM on these devices. Does look like it's possible to fit three cores in the lp1k though
How much is an ARM license ? :D
An arm and a leg
RISC V needs more GPL
so, if iCE40 can hold 8 cores at 50Mhz, and it costs 38 bucks, then you need 36 cores to match the performance of a single core on ARM A53 processor (1.4-1.8Ghz) which is used in Raspberry Pi, but there are 4 of them, so you would need 36 x 4 = 144 cores. So 144 / 8 = 18 , you would need 18 FPGA boards of $38 each (and that would be $684 bucks) to match the performance of Raspberry Pi which costs around $35 bucks in retails stores and $15 if bought from China in bulk. Now, these Raspberry Pi processors have DDR3 , so I better forget about this comparison as it would obviously be in favor of today's ARM RISC processors, which will be hundred of times faster than today's FPGAs. You would need to make 500 cores of 50Mhz each on each FPGA to compete with today's powerfull ARMs , and that would require an FPGA with around 1.5M or 2M LUTs , and these devices cost thousands of dollars.
Moral of the story: FPGAs are completely useless and not competitive with today's ASICs , they can only be used as toys for playing on the weekends.
Absolute Zero my friend FPGAs can fit cpus with custom instructions, that the user define, to make an ARM with a custom instruction you will need millions, unless you’re shipping millions of them you cannot justify it, and if you can’t make it to the market by the time you need, then that’s an FPGA use case, There’s a FPGAs in your phone in your motherboard, in your laptop, and right now in the cloud
you are a troll and have no clue what you are talking about
You can always choose a better and more expensive FPGA
Someone should really tell this to the billion dollar fpga industry. They don't seem to have noticed everything they do is actually useless.
FYI, you are comparing mcus to mpus. Fpga has its place. When you don't have 100s of millions to fork out on a tape out, you use fpgas. How do you think arm or Intel development happens? These are initial development. Arm was also expensive when it came out. It fell from $20+ to about $0.5 over time. Cortex m4 was over $5 just 2 years ago. Today it's < 1.