Why aren't we seeing more PCI-E 3.0 x1 dGPUs?

[GodisanAtheist]

AMD just bought Xillinx who are a major player in the FPGA area. Hell, they're one the companies (along with Altera) that pretty much pioneered the technology.

However most of the bitcoin mining is done on ASICs which are dedicated hardware and will always have an advantage over an FPGA for any dedicated task like that. An FPGA would be better in that it could be configured to mine different types of cryptocurrencies depending on shifts in valuation, but I don't know if there's enough valuation shift in the currencies to account for the performance uplift that an ASIC will grant, especially when process improvements will obsolete hardware faster than just about anything else.

Bitcoin is big enough that someone can probably afford the prices of the cutting edge wafers, so even though a company like Xillinx could use a cutting edge process just by virtue of having so much volume and flexibility with an FPGA, they still won't be able to achieve the an economic edge over an ASIC.

- But anything soaking up GPUs to meet crypto demand can certainly benefit more from FPGA's right? Bitcoin is past both GPUs and FPGAs so its not really contributing to the current state of affairs in GPU availability, but stuff like ETH is.

Companies like AMD could take some of the mining pressure off their GPU division by pumping out either drastically cheaper FPGAs that matched performance or equally priced FPGAs that exceed the performance of equivalent GPUs.

 

[MrTeal]

- But anything soaking up GPUs to meet crypto demand can certainly benefit more from FPGA's right? Bitcoin is past both GPUs and FPGAs so its not really contributing to the current state of affairs in GPU availability, but stuff like ETH is.

Companies like AMD could take some of the mining pressure off their GPU division by pumping out either drastically cheaper FPGAs that matched performance or equally priced FPGAs that exceed the performance of equivalent GPUs.

GPUs work so well for mining Ethereum because the algorithm was really tailored to mine well on GPUs and be ASIC resistant. They're commodity parts that benefit from being on a cutting edge node and having hundreds of millions or billions in R&D already dumped into them. I'm sure that if AMD really wanted to they could make a mining FPGA with programmable logic and a massive high speed memory bus that would be better suited for mining than a GPU, but what's the cost of development on that?

Also, a top of the line FPGA like a Xilinx Vertex Ultrascale+ starts at $10,000 for the smallest one and tops out north of $100,000, so you'd have to make a compelling argument for taking resources away from that business to start catering to miners operating on thin margins.

 

[Shivansps]

Actually, you really dont need to do a thing, no need to make physical changes to gpus interfaces, altrought, gpus like the Gt1010 should only come in x1 interface, no point in anything faster than that.

The only thing that is really needed here is is motherboards to start coming with open ended X1 slots, thats it, they do not do that because they tend to cheap on the power delivery.