Add more context to router idea #10
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Especially of interest is how "wide open" the FPGA accelerator field seems to be.
sethp
commented
May 11, 2024
| - "Acceleration" chips: The best way to do FPGA development seems to be on an FPGA.... but, not "directly," instead using the FPGA to coorindate with the RTL simulator. What's _that_ chip selection look like? | ||
| - Amazon will rent you an [F1 instance with a single (1) attached FPGA](https://instances.vantage.sh/aws/ec2/f1.2xlarge) for $1.6500/hr on-demand (~$0.6488/hr spot), which is about $1200 a month ($500 for spot). They've got a density problem, it seems, since the largest size is an [`f1.16xlarge`](https://instances.vantage.sh/aws/ec2/f1.16xlarge) which comes with 8 FPGAs and a whopping 976.0 GiB of RAM with 64 vCPUs; that and the dedicated 25 Gigabit networking suggest they're renting you the whole blade at that point. But: in our (very limited) experience with RTL simulation, a fairly complex model with no acceleration takes on the order of ones of CPUs and tens of _Megabytes_ of memory. Meaning, a burst-credit model for way oversubscribing the CPU and nano- or micro-scale memory (0.5 to 1 GB) might get us equivalent performance to the `f1.2xlarge` at literally 1/100th the cost of the `f1.2xlarge`. But, to sell us that, assuming their blade is already otherwise "minimally sized" for an AWS DC, they'd need to slot some 512 or 1024 FPGAs per blade. Maybe in the `f2` class, eh? | ||
| - Or, it's possible that _is_ a good deal for running a workload more than a couple hours a month, and I've wildly missed my guesses above. In which case, the FPGA itself makes up a significant proportion of the price, and the kind of hardware that people use to simulate other hardware costs on the order of tens of thousands of dollars. I sure hope not! | ||
| - But, uh, "firesim" (which is the FPGA accelerator underneath chipyard?) [supports](https://docs.fires.im/en/stable/FireSim-Basics.html#choose-your-platform-to-get-started) the Xilinx Alveo U250 (or U280, which is [discontinued](https://www.xilinx.com/products/boards-and-kits/alveo/u280.html)), [Xilinx VCU118](url), and the RHS Research Nitefury II. That's a [$10,000 "accelerator card"](https://www.mouser.com/ProductDetail/AMD-Xilinx/A-U250-A64G-PQ-G?qs=unwgFEO1A6vi%2FTDpEsBygA%3D%3D), a [$10,000 evaluation kit](https://www.xilinx.com/products/boards-and-kits/vcu118.html) (although you can back-order the chip itself inside the evaluation board for a [cool $50,000](https://www.digikey.com/en/products/detail/amd/XCVU9P-L2FLGA2104E/7421960), which, uh 😵💫). But, [RHS Research](https://rhsresearch.com/pages/about-us) to the rescue! Their Nitefury II is a M.2 form-factor accelerator for $150, which is an awful lot more "in budget", even if it only seems to be available via [Amazon](https://www.amazon.com/dp/B0B9FMBF6C). Also of interest: https://www.crowdsupply.com/rhs-research & https://github.com/RHSResearchLLC/NiteFury-and-LiteFury . |
There was a problem hiding this comment.
Should we make an FPGA accelerator board?
There was a problem hiding this comment.
and/or, an m.2 to USB4 adapter, a la https://www.amazon.com/dp/B09XJWH5P1
I'm not exactly sure what I'm saying here, but a Framework expansion bay that "fits" the RHS Research modules (at least the adorable 2230 one) would be cool, especially since slamming the little guy deep inside the laptop makes these I/O connectors pretty permanently inaccessible
Attempts to answer the question "who is this for" & "why do they want it?"
Everything but the price looks really promising; and the cost is ≥50% the PHY chip, for which I did almost no research/cost comparison.
|
ok @dougli1sqrd I think this one is ready for your 👀 ; the margins don't look too great, but everything's looking like it pencils on the back of the envelope to me. What do you think? I have a rough cut at a development plan too that ought to answer the "what is it/what do we want it to do" questions, but I think this one is big enough without throwing that on top at the moment. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Leaving this in draft for now, since I wanted to add some details from the IGLOO2 application note too.