Graphics card drought

The upgrade to Resolve Studio 17 has become a bit of a mission. The GTX-670 graphics cards do not cut the mustard for Resolve 17. So muggins thinks just go out and buy something more up to date. It was a good thought while it lasted. For love or money, none are to be had.

COVID hermits playing video games must have scooped them all up and production may have been curtailed with COVID running amuck. May have to revert to Resolve 16 unless I want to sit like a loon editing by counting frames.
I had a gutful of that with a home made 16mm film editor and am certainly not looking for seconds. I wonder if Blackmagic Design may yet enter the graphics cards business. RED had a shot at it for a while.

It would surprise me if they did.

That is the weird part. One card did in freebie Version 16 and in freebie version 17.1.1. until after I installed the Studio Resolve version 17.1.1. which asked for the dongle. After that, the CUDA option for the cards was not offered and the system began to play back very slowly and drop out on UHD renders.

For now I have reverted to Version 16. Version 16.1 seems to be the only version which will play with the GTX-670 cards now and I had to chase up an earlier graphics driver from an alternative site as Nvidia has removed the old drivers from the official list.

It may be time for Blackmagic to make and sell their own CUDA cards.

RED had an fpga doing redcode decoding, which is a few orders of magnitude simpler piece than a proper gpu. It isn’t easy to do a gpu, which is manifested by the fact that there are only two serious general purpose gpu vendors in the whole world: nvidia and amd. BMD could try ofcourse, maybe they can make the 3090 equivalent for 200$

A dedicated card just for Cuda/OpenCL tasks would be nice. If that should be done with a fpga or a gpgpu I don't know, but it would be simpler to build than a full blown graphics card. Pretty sure BM is aware of the demand and has been for a while, so it's obviously it isn't simple enough to be economic, despite the demand. I wouldn't hold my breath for such a development.

The force that ultimately killed the big piece or iron was the ability to actually buy off the shelf component for the main hardware.

Making a “dedicated” card is a lost cause at this point.

The issue is not the lone gamer at home, but the Bitcoins miners that can buy literally hundreds of cards because of the insane ROI.....

In a way, if NVIDIA (for example) prevent Bitcoinn Mining, they will lose sales.... so, it is a though game...

A dedicated card just for Cuda/OpenCL tasks would be nice. If that should be done with a fpga or a gpgpu I don't know, but it would be simpler to build than a full blown graphics card. Pretty sure BM is aware of the demand and has been for a while, so it's obviously it isn't simple enough to be economic, despite the demand. I wouldn't hold my breath for such a development.

Why would it be simpler? Who will buy a card that doesn’t work on Fusion page or with other softwares which rely on opengl for processing (like for example Scratch)?

The upgrade to Resolve Studio 17 has become a bit of a mission. The GTX-670 graphics cards do not cut the mustard for Resolve 17.

Your best bet atm is probably an RTX 2060. It's got some fair performance in it.

Newer cards may take until next year to stock up again, sorry to say.

It doesn't help that hardware retailers like Overclockers are throwing 300 - 500 quid on the RRP. I'm never buying from them again.

My RTX 1650 failed in February and I've been without that computer since then. Since it's the only system in-house that runs Resolve, it's been a hell of a drought. Even medium-spec graphics cards have been AWOL since early in the year.

Pure chance allowed me to acquire a 3060 for about 2X the MRSP. A bargain, it turns out, since I can now edit again.

Something needs to be done about crypto-miners. Talk about "unintended consequences of technology".

Apparently, they use more power than Argentina. So far.

Bitcoins miners that can buy literally hundreds of cards

Not that it matters in our context, but to my knowledge Bitcoin is mostly mined with asics, not with gpus. It's etherum that the gpus are being used for.

Why would it be simpler?

We buy GTX cards mainly for their CUDA cores, but they come with a huge overhead of technology that is for gaming. It seems plausible to me that a pure CUDA/OpenCL device is simpler to manufacture. You are right obviously that the market for such a device is too small. Else it would exist.
It is just me dreading that I have to buy a super expensive device that is advertised with enormous powers for realistic 3D physics when all I need is some programable parallel processing power.

Of course that product does it exists. That's what Nvidia sells to Miners and data centers. But contrary to what you'd expect, it's not cheaper.

Of course that product does it exists. That's what Nvidia sells to Miners and data centers. But contrary to what you'd expect, it's not cheaper.

Yeah, that's what makes it a useless product really. Plus targeting lower end cards when they're all buying 3080s and 3070s or higher... it's daft, really.