Simulating RISC-V Clusters with FPGAs on AWS | Hacker News Reader

Simulating RISC-V Clusters with FPGAs on AWS

aws.amazon.com

86 pointsfabuzaid15 comments

I think it proves AWS will be an alternative and potentially great choice for fabless chip company/startup.

They all need emulation to validate their chip design. However emulator is very very expensive, and unlikely in full use.

Deving on AWS? That seems awfully frustrating and expensive. Now every time I click compile I can remember how much this is going to cost me.

You should compare this to the cost of buying an FPGA. It can be cost-effective.

https://www.digikey.com/product-detail/en/xilinx-inc/XCVU9P-...

The list price of a Vu9P is like $50,000. I'm sure that's not the real price, but still.

https://www.digikey.com/product-detail/en/microsemi-corporat...

This is about half as powerful but comes preassembled and ready to go. Also "only" $500

The 'VU9P has ~2.5M logic elements and 6840 DSP units. The M2S090 has ~90K logic elements and 84 math blocks. It's nowhere near the same class of device.

The AWS FPGA offering only makes sense if you absolutely need a massive FPGA connected to massive amounts of RAM, and don't want to spend the bucks to own it yourself.

Microsemi doesn't make high performance FPGAs. Their niche is aerospace/defense, low-power, and flash/instant-on FPGAs.

Yes, it's expensive. But you have to compare it fairly: think of the cost of racking up servers w/PCI FPGA cards or devboards with FPGAs and JTAG flashing tools. How many of your team can you enable with these solutions, and how long will it take to set it up? What if you're a small team working on a RISC-V compiler and you don't have an IT dept?

If you've got a global team, then it's even more interesting to reproduce these configurations reliably among multiple sites.

Also, if you have to need 10000 for a day to do some sort of dev/QA/integration efforts you don't have to actually own them. Or even physically deal with the things.

Good luck getting 10k same type instances from AWS

To try this, do I need the full 16XL F1 instance, or can I run 1 node on the 2XL (1 FPGA) instance?

You can run a single node on an f1.2xlarge instance.

How much is an f1.2xlarge instance? My google fu seems to be failing me.

$1.65/h on demand

https://aws.amazon.com/ec2/pricing/on-demand/ $1.65/hour in N. Virginia

Way less with spot prices, but for something like this, you probably want on-demand.

Reply on news.ycombinator.com