FPGA

what do we think about FPGAs?
youtube.com/watch?v=rhT6YYRH1EI

Attached: fpga.jpg (1280x1280, 108.62K)

My actual job is programming these shitbricks for high-speed traders to take advantage of the high-bandwidth/low-latency.

The devices themselves are awesome as hell. If you need to have large calculations for a very specific task done in one or two clock cycles then they get the job done.

The issue is the tooling for them is shit. Everything is proprietary and there's no compatibility between different pieces of software. If you come from a systems-programming background where your compiler and system libraries are public and open-source you will absolutely despise the restrictive nature that Xilinx and Altera give.

sounds expensive
>de-10 doesn't ship to even students purchasing until after june

i have one an enjoy it with s-video output on a vintage crt - however its retarded to spend above msrp

I've had a mister for a while now, great little doodad some chad just created a Sparcstation5 core.

Absolute horror to program. I currently do an internship where I have to transfer an algorithm to work on these things. I do have to say that high level synthesis tools make those things actually usable.
The tool chains are broken. The manufacturers supply install guides, but even if you satisfy all requirements, you will have problems with the software.
The last 5 days I had a bug where reading from a valid register would always return false. Today I found out this only happens if you step trough the program in debug mode. Kind of ironic and frustrating...

You will have a hard time making use of these things. You can't just work on data from the RAM, you have to transfer it first to the FPGA. This takes time and maybe even slower than doing everything on the CPU.

There is so much more, but you will never feel like a real FPGA engineer because most of the time you don't know what the fuck is going on and you are just trusting that the IP that other people wrote works.

Attached: 1532705574395.png (787x633, 475.15K)

>trading FPGAs
What's next, actual ASICs?
I heard Intel do CPU auctions for high frequency trading.

Maybe AMD will do something about it.

> What's next, actual ASICs?
We specifically do not use ASICs because we need to be able to keep our current systems up to date with whatever changes the exchanges make in their trading protocols. We also don't need the transistor density that ASICs provide, which is basically their only advantage over FPGAs.

I admire your hope, but this goes beyond a single company. Intel has owned Altera for years and they are even more open-source friendly than AMD. We need toolchains that are both powerful and platform agnostic. That takes significantly more effort for hardware than software projects like GCC or Valgrind.

So you do HTTP in FPGA? And the trading algorithm too?

I agree the tools suck Quartus is a heap of shit. But getting back to OP FPGA is a great tool especially in context of MiSTer and such for retro preservation. With retro systems defined both in emulators and in HDL they will stand the test of time, and HDL gives the advantage of being better suited for creating replacement parts for vintage systems.

No, we provide clients with two normalized trading protocols, either a FIX (dynamically sized and human-readable) or binary variant, and then we use FPGAs to translate that normalized procotol into whatever the exchanges use. NYSE FIX, Pillar, BATS, OUCH, or whatever NIH-bullshit that each exchange wants to have. We don't handle the actual trading algorithms themselves, just the translations. Unfortunately that means we need to manage an entire TCP implementation along with each trading protocol in the FPGA, but the result is we can do the translation in a couple of nanoseconds where a CPU would take tens or hundreds of microseconds at minimum.

Though I'm 90% certain that our clients have their trading algorithms in FPGAs too, and 100% certain we pass the data to risk-calculating algorithms that's are programmed in FPGAs as well. High-speed trading is all about chains of FPGAs now. I really want to leave fintech but the pay is decent.

As nice as retro consoles in FPGAs are, they are never going to get the same mass-adoption that software emulators have unless consumer CPUs start shipping out devices that have FPGAs directly built into them. Depending on what kind of software is developed in the future that might be a possibility, but software emulation provides way too many advantages, one of which is modding support.

Never say never.

>only decent
I get calls from all kinds of fintechs and they offer 500k+ starting excluding the sign-on bonus and matchings and all that shit, in medium cost of living areas. And I have 0 fintech experience.
Wrong specialization?

Depends. When it comes to fintech, the financial stuff doesn't really matter. It's super easy to train someone in that aspect as long as they know basic arithmetic and algebra. I joined as a GPU programmer and I think my training in the financial aspect took 3-6 months?

It's the "tech" part of fintech that they're looking for. If you know how to program FPGAs (and more importantly, debug them) and can demonstrate that knowledge in an interview, then you can easily get a 250k+ position. Not sure about the 500k+ starting though. The only one in my company paid that much is in a contractor role and he a literal wizard.

They're considering me for stats/modeling/medium to long-term trading stuff, not for hft nor for GPU programming/FPGAs/etc. I do have GPU programming experience but they don't even know it.

What does modding have to do with software vs. hardware?

No idea then. It sounds like some sort of quant-trading firm, but I only work in the HFT space, and even then I don't touch the trading algorithms themselves. I would be very surprised if a non-HFT firm would hire someone at 500k+ starting though. HFT makes the most kind of money, especially when the markets or individual stocks become volatile. GME cucks would cry if they realized how much money shitadel made from HFT on that stock alone in the past 2 years.

It's so much easier to mod software emulators with different memory hacks, shaders, or bug fixes than it is in a hardware emulator. If you want to mod a hardware emulator then you need to modify the ROM itself or have modding support built into the hardware. That's probably doable for older systems like NES or SNES, but it's exponentially more difficult for more modern systems.

There's modding and playing games, and there's reproducing logic for creating replacement parts for retro systems. Not the same thing.

I'm not a hardware engineer nor an fpga programmer, but I've been looking at fpgas and prototyping and microcode for a while now and it has me interested. Know of any cost effective development kits for a hobbyist/amateur looking for a crash course?

Some of the workshops I've found call for some expensive purchases (for me)

welp that didn't work. You can have my Megumin as a token.

Fair enough. If you're just trying to replace an IC that is no longer available for purchase then an FPGA can fulfill that role.

If you're trying to replace an entire console though, an FPGA is an expensive and less capable solution compared to a sofware emulator.