Custom made GPU miners

[Sjalq]

Here's an idea.

How about we pitch one of the smaller manufacturers of graphics cards to build a cool, power efficient, low memory multi-GPU card?

It won't be profitable for ATI to target a niche market, but it will be for a small graphics card maker.

[1713892557]

1713892557

[1713892557]

1713892557

[1713892557]

1713892557

[Vladimir]

Niche market? I bet 50% of 5970 and 6990 and even 5870 and 5850's will end up in mining rigs. The only reason why AMD does not make cards specifically for bitcoin (and crypto in general) yet is because they are slow.

[bcpokey]

Not particularly profitable for anyone to build just for mining.  The most powerful mining card the 5870 is fabricated on a 40nm process. 40 NANO-meters. Think about that scale and what it implies as to the difficulty of designing and creating a working piece of technology. There is a reason there are literally only 2 gaming GPU manufacturers out there, cost is prohibitive. The only current alternative is FPGA or some other form of ASIC, which still represents thousands of dollars of investment into silicon for calculation power far inferior to gaming GPUs, and still has no code base to support it.

There may be 10 or 20 thousand hardcore miners? I'm guessing that just contracting a fab plant alone would cost more money than all of them would be able to scrape together.

Not to mention that it wouldn't really be profitable for miners either, well, perhaps the few larger groups who could afford such a project (which would further squeeze out the small fry who didn't / couldn't get on board). Higher hashing power does not increase block rate generation, it merely makes it harder to generate.

Niche market? I bet 50% of 5970 and 6990 and even 5870 and 5850's will end up in mining rigs. The only reason why AMD does not make cards specifically for bitcoin (and crypto in general) yet is because they are slow and stupid.

Hrm. I'm curious about this premise. The current network hash power is 4Thash / sec no? 4THash / sec = 4000 Ghash /sec = 4000000 MHash/sec no? A 5870 can do ~400MHash/sec, so 4THash = 10,000 5870s. We know for a fact that the network has tons of 5850s, 5830s, 5770s, 4870s, etc., but even assuming just 5870s that number is quite low.

Are we suggesting that globally there were less than 20,000 5870s created for sale?

[Sjalq]

Not particularly profitable for anyone to build just for mining.  The most powerful mining card the 5870 is fabricated on a 40nm process. 40 NANO-meters. Think about that scale and what it implies as to the difficulty of designing and creating a working piece of technology. There is a reason there are literally only 2 gaming GPU manufacturers out there, cost is prohibitive. The only current alternative is FPGA or some other form of ASIC, which still represents thousands of dollars of investment into silicon for calculation power far inferior to gaming GPUs, and still has no code base to support it.

There may be 10 or 20 thousand hardcore miners? I'm guessing that just contracting a fab plant alone would cost more money than all of them would be able to scrape together.

Not to mention that it wouldn't really be profitable for miners either, well, perhaps the few larger groups who could afford such a project (which would further squeeze out the small fry who didn't / couldn't get on board). Higher hashing power does not increase block rate generation, it merely makes it harder to generate.

I was referring to the folks who build the cards not the GPUs. There are several makers of graphics cards. The expertise is still high but much lower than producing nm ICs.

[Vladimir]

Given 6 Thps it can be estimated that there are around 20 000 chips doing mining right. It can easily be 10 times that by the end of the year, not to mention all the spooky things which are being calculated by FPGA's these days and which can be calculated by GPU's.

Moreover, they even do not need another chip, they just need a different board with better cooling solution, different configuration, BIOS, and less memory.

[bcpokey]

Sigh no multiquote.

Network hashing power peaked at 4THash/sec not 6? To suggest that there would be 10x as many cards hashing it out would require 10x as much return, not impossible but certainly improbable, and one of the scenarios in which the argument "buying coins is better than buying hardware" applies.

If you mean to contact one of the pcb manufacturers, you are still suggesting a design revision which is actually very expensive. I don't think you guys understand the scale of actual production that allows for these things to exist.

http://www.techradar.com/news/computing-components/graphics-cards/amd-shipped-16-million-dx11-cards-in-9-months-703763

In 9 months from launch alone AMD put out 16 million 5xxx series chips. Again I will note that AMD manufactures the GPU not the smaller companies that utilize the chips. They have no ability to fabricate 40nanometer chips. This is an insanely expensive process that is far beyond the scope of any smaller companies. I can only assume that their last gen fabrication line has been shut down but if they had some excess stock I suppose it would be possible for a smaller company to gobble it up and try to make some cards out of them, but again, 20,000; even 200,000 is such a tiny portion of the market that it is not appealing at all. Especially given the healthy appetite people have for the existing unmodified platform that exists.

[Sukrim]

Additional to this is that many miners are happy if they can sell their mining card to games, should Bitcoin crash suddenly (If you don't think BTC could crash and it's price will always rise, buying BTC would be an even better option...).

With custom optiized GPUs this would not be possible any more, so the hardware value would also go down a LOT should mining be no longer attractive.

In this scenario you might rather look at ASICs or FPGAs...

[PcChip]

I don't think you get his meaning; he doesn't mean AMD re-do the chip, he means for someone like sapphire, xfx, his, or diamond to build a board with 4 cypress gpu's and 512MB RAM

[Littleshop]

It is possible that someone could build a custom card out of EXISTING chips such as a card that had a bunch of efficient laptop GPU's in a desktop card package.  The part that would make it a 'bitcoin' specialty item would be a selection of chip that does integer work faster with the lower power usage combo. 

A much easier build (but still probably not happening soon) would be a near stock board but with a custom bios with memory clocks lowered and less ram for less power usage.  It also might have a mod where it pulls almost no power out of the slot and even a PCIE x1 slot connector.

[no_alone]

But what the point in that///???
The difficulty will just go higher!

[bcpokey]

I don't think you get his meaning; he doesn't mean AMD re-do the chip, he means for someone like sapphire, xfx, his, or diamond to build a board with 4 cypress gpu's and 512MB RAM

I got his meaning just fine, I don't think you get my meaning.

1) Cypress chips are not to my knowledge being fabricated anymore. I don't suspect any group is just sitting on a big fat stockpile of unused Cypress chips, so in order to start a new line either AMD would need to return to a last gen fabrication line or have some stockpile to be bought out.

2) A PCB design process costs a lot of money. I'm not an electrical engineer and I'm guessing none of the rest of us are either, but talk to one and they will tell you that it's not a simple matter of slapping a chip onto a piece of silicon to make a functional graphics card. Where is the profit motive?

3) Cypress cards already sell very very well. What is the profit motive to instead of running their already existing designs, creating a custom designed PCB? As already stated there are maybe 20,000 high end cards mining right now maximum, that may seem like a big number to you, but to companies that deal with shipping out millions of units that's a waste of money.

[Sukrim]

I don't think you get his meaning; he doesn't mean AMD re-do the chip, he means for someone like sapphire, xfx, his, or diamond to build a board with 4 cypress gpu's and 512MB RAM

And I say a card like this has a faster value deprecation should BTC crash than FPGAs or "conventional" GPUs.

[JJG]

1. Bitcoin is a drop in the bucket. There are many, many, many gamers out there. Just look at the CoD sales numbers.
2. Developing a GPU is a very long process. Definitely greater than 1 year.
3. Current ATI GPUs are already very good at this. Making them better would entail sacrificing elsewhere.
4. Bitcoin is unstable and new and has a questionable future. If the exchange rate falls, mining hardware interest will fall as well.
5. No company in their right mind would invest in hardware tailored to a small niche with an unstable future.