I had a crazy idea, I wanna make a graphics card company.

[Nena Trinity]

Okay let's say I got a small loan of a billion dollar$ and went to ARM and licensed their latest G77 graphics for the intended use of making graphics chips for laptops and desktops trough PCI-E and got a board partner and a PC laptop maker onboard? Is there any reason why this could not work? Drivers? To weak GPU cores? Chip would be to big to compete with these cores? 

[Levent]

You wont be doing that with measly $1B. Nvidia's 2019 gpu development costs were $2.8B.

[Radium_Angel]

15 minutes ago, Nena360 said:

let's say I got

How to make Chicken Soup:

 

Step 1 - Find a chicken

 

In other words, get the billion dollars, THEN come back and we'll talk fabs and foundry 

[Eigenvektor]

Aside from what @Radium_Angel said, take a look at Intel. They're trying to enter the discrete GPU market. Let's see how much money that is going to cost them and how quickly they can gain market share. And don't forget that they already have a functional company and dozens of years of experience in chip making, marketing, international shipping, international laws etc.

 

Licensing a GPU design is well and good, but you need employees for development (e.g. Windows drivers), marketing, packaging and so much more, all of which you need to factor into your $1B. Then you'd have to see how well that mobile GPU does against a desktop GPU. Unless you can compete on performance and/or price you're screwed. You may be screwed either way, just look at AMD vs Nvidia.

 

And yeah, first step would be to actually find someone willing to hand you $1B without any sort of bullet proof business model.

[Nena Trinity]

18 minutes ago, Levent said:

You wont be doing that with measly $1B. Nvidia's 2019 gpu development costs were $2.8B.

But that is ARM, they made the cores... 

3 minutes ago, Radium_Angel said:

How to make Chicken Soup:

 

Step 1 - Find a chicken

 

In other words, get the billion dollars, THEN come back and we'll talk fabs and foundry 

Its mainly theoretical but the plan was TSMC or Samsung fabs, which ever offer cheaper fabs... Maybe even Global Foundries on 14nm if needed! 

[Nena Trinity]

1 minute ago, Eigenvektor said:

Aside from what @Radium_Angel said, take a look at Intel. They're trying to enter the discrete GPU market. Let's see how much money that is going to cost them and how quickly they can gain market share. And don't forget that they already have a functional company and dozens of years of experience in chip making, marketing, international shipping, international laws etc.

 

Licensing a GPU design is well and good, but you need employees for development (e.g. Windows drivers), marketing, packaging and so much more, all of which you need to factor into your $1B. Then you'd have to see how well that mobile GPU does against a desktop GPU. Unless you can compete on performance and/or price you're screwed. You may be screwed either way, just look at AMD vs Nvidia.

 

And yeah, first step would be to actually find someone willing to hand you $1B without any sort of bullet proof business model.

Yeah but besides the cost, it's mainly a in theory kinda deal... Could the ARM cores work? Or are they simply not able to compete?

[mariushm]

You can purchase a license from ARM or whoever owns Imagination Technologies nowadays to a GPU core, but then you'd have to hire engineers to make the circuit board layouts, and you have to hire programmers to make drivers for Windows and then you'd end up with a video card that's probably comparable with a GT 1030 or a RX 460 in performance, or some integrated graphics.

You would have a hard time adding features and innovating because A LOT of techniques are patented by AMD, nVidia and Intel, and you'll be sued.

So you'll be spending loads of money and you'll have a product that isn't competitive and nobody will buy it.

 

Probably the best one could do to get into video cards as a new company would be to make some kind of new Voodoo card, but with raytracing. For example, make an only raytracing gpu chip ...and make it like AMD did the Ryzens, with chiplets... basically, make it possible to put 16-32 chips on a video card and each chip handles a small portion of the screen.

 

 

[Eigenvektor]

3 minutes ago, Nena360 said:

But that is ARM, they made the cores...

Which means you're dependent on ARM being able to produce a design that is competitive with Nvidia's offerings. You'll be at a disadvantage compared to companies who can design their own chips to fit their needs and innovate.

Quote

Its mainly theoretical but the plan was TSMC or Samsung fabs, which ever offer cheaper fabs... Maybe even Global Foundries on 14nm if needed! 

Those fabs have limited capacities which means you'll be dependent on TSMC/Samsung/GF having enough capacity to even manufacture your chips for you. You'll be at a disadvantage compared to other companies who have existing contracts in place.

4 minutes ago, Nena360 said:

Yeah but besides the cost, it's mainly a in theory kinda deal... Could the ARM cores work? Or are they simply not able to compete?

I've yet to see a real benchmark that compares something like a Mali G77 against something like a 2080 Ti. You have to understand that those are mobile chips, which means they are mostly optimized towards low power usage. They have to work inside a mobile phone which has very limited power available, while the desktop card can draw somewhere upwards of 200 watt. I have no idea how well ARM's architecture scales.

[Nena Trinity]

11 minutes ago, mariushm said:

You can purchase a license from ARM or whoever owns Imagination Technologies nowadays to a GPU core, but then you'd have to hire engineers to make the circuit board layouts, and you have to hire programmers to make drivers for Windows and then you'd end up with a video card that's probably comparable with a GT 1030 or a RX 460 in performance, or some integrated graphics.

You would have a hard time adding features and innovating because A LOT of techniques are patented by AMD, nVidia and Intel, and you'll be sued.

So you'll be spending loads of money and you'll have a product that isn't competitive and nobody will buy it.

 

Probably the best one could do to get into video cards as a new company would be to make some kind of new Voodoo card, but with raytracing. For example, make an only raytracing gpu chip ...and make it like AMD did the Ryzens, with chiplets... basically, make it possible to put 16-32 chips on a video card and each chip handles a small portion of the screen.

 

 

Oh boy you think they will be that bad? Pre-built system builders may still buy em tough... Circuit board partners I know can be obtained in China/Taiwan tough the cost is still a question mark to me... XD

9 minutes ago, Eigenvektor said:

Which means you're dependent on ARM being able to produce a design that is competitive with Nvidia's offerings. You'll be at a disadvantage compared to companies who can design their own chips to fit their needs and innovate.

Those fabs have limited capacities which means you'll be dependent on TSMC/Samsung/GF having enough capacity to even manufacture your chips for you. You'll be at a disadvantage compared to other companies who have existing contracts in place.

I've yet to see a real benchmark that compares something like a Mali G77 against something like a 2080 Ti. You have to understand that those are mobile chips, which means they are mostly optimized towards low power usage. They have to work inside a mobile phone which has very limited power available, while the desktop card can draw somewhere upwards of 200 watt. I have no idea how well ARM's architecture scales.

Scaling and how many Mali cores is needed to make it compete more cores the bigger die, bigger die means more costly & so on... If they scale for shit you may need twice as many as AMD/Nvidia uses and then the plan will fail... X-X

[GoldenLag]

28 minutes ago, Nena360 said:

Maybe even Global Foundries on 14nm if needed! 

good luck taking fabspace on a small node. you will be lucky to start with 28 or 32nm. 

 

in the end, the best you can aim for is a niche market of many display GPUs or Displayoutput GPUs. one of which you can charge a premium for, the other one is literally trying to be as cheap as possible. 

[Nena Trinity]

23 minutes ago, GoldenLag said:

good luck taking fabspace on a small node. you will be lucky to start with 28 or 32nm. 

 

in the end, the best you can aim for is a niche market of many display GPUs or Displayoutput GPUs. one of which you can charge a premium for, the other one is literally trying to be as cheap as possible. 

Mali G77 is designed for 7nm bumping it to 14nm would be bad enough but at 28nm holy bag og *bleep* balls that would suck, still would beat Matrox! 8D And yeah that market may be good along with budget gaming, one thing botherse what about VRAM? O3O 

[GoldenLag]

13 minutes ago, Nena360 said:

And yeah that market may be good along with budget gaming

I dont think you would manage to touch that market. Its covered far to well by other products. 

 

13 minutes ago, Nena360 said:

Mali G77 is designed for 7nm bumping it to 14nm would be bad enough but at 28nm holy bag og *bleep* balls that would suck, still would beat Matrox!

Keep in mind that 150nm is still fairly popular..........

 

28nm is still a very small node that isnt exactly the easiest to get your hands on as a startup.

[Nena Trinity]

1 hour ago, GoldenLag said:

I dont think you would manage to touch that market. Its covered far to well by other products. 

 

Keep in mind that 150nm is still fairly popular..........

 

28nm is still a very small node that isnt exactly the easiest to get your hands on as a startup.

A poop well the Mali G77 probably does not support anything other than LPDDR4 anyway x-x

[Radium_Angel]

Start here:

https://www.matrox.com/graphics/en/

 

Buy these guys out, move onward from there with their designs, etc. Scale up, ramp up and leverage their existing marketshare.

[Nena Trinity]

21 hours ago, Radium_Angel said:

Start here:

https://www.matrox.com/graphics/en/

 

Buy these guys out, move onward from there with their designs, etc. Scale up, ramp up and leverage their existing marketshare.

So what ya saying is try to make some profits with them first then go make chip after? OwO (these guys use AMD chips, replacing them with ARM may prove hard if they signed any long term deals...)

[mariushm]

The x86  patents expire next year or so, so it may be a better challenge to make a cpu...

[Eigenvektor]

15 minutes ago, mariushm said:

The x86  patents expire next year or so, so it may be a better challenge to make a cpu...

Unless that includes AMD64 and other, newer additions, I don't think you need to bother.

[straight_stewie]

On 11/26/2019 at 7:21 AM, Nena360 said:

Okay let's say I got a small loan of a billion dollar$ and went to ARM and licensed their latest G77 graphics for the intended use of making graphics chips for laptops and desktops trough PCI-E and got a board partner and a PC laptop maker onboard? Is there any reason why this could not work? Drivers? To weak GPU cores? Chip would be to big to compete with these cores? 

There are two major roadbloacks here, one technological and one financial:

1. If the Arm G77 IP was useful for your target demographic, then it would already be being used in that demographic. When Arm says it's targeted at mobile they mean phones and tablets, not mobile PCs.
2. If you have access to a billion dollar loan you are either the child of a billionaire or you already have a billion dollars of high performance assets to gain leverage with.

If you've solved the second roadblock already, then I would suggest that you consider entering the RISC-V market instead of the desktop graphics market. The desktop graphics market is rife with competition from two massive and well established companies right now, and that means that those two firms, Nvidia and AMD, are pumping money into their desktop graphics solutions groups: You will not be able to compete with them on a billion dollars credit and no positive revenue, unless you have already made extremely significant innovations at the silicon level. These are 100 billion+ dollar companies that can afford to hemorrhage insane amounts of money on the high performance GPU market to combat their competitors if it becomes necessary to do so. Nvidia and AMD will definitely crush any new startups in the desktop graphics processing space, and if they can't crush them because they've made significant game changing innovations, they will just buy them.

However, the RISC-V space is a market that is just starting out. With a billion dollar commitment you have more than enough money to start a market leading firm, if you can find high quality chip manufacturers that are willing to work with you. You still likely won't be able to manufacturer desktop caliber processors in house with just a billion dollars credit to work with: Advanced semiconductor manufacturing is overwhelmingly expensive, especially when you don't already own the equipment necessary to build the physical chips. But, you might be able to design market leading RISC-V IP. In this business model your primary competition will eventually be ARM, that is, if the market segment actually becomes viable.

[Mira Yurizaki]

Let's step back and start with the basics:

• Do you know how to manage a reasonably complicated project?
• Do you have even a smidgen of engineering experience?

If you answered "no" to both, you'd have zero chance of even getting venture capital. Even if we pretend someone or someones were willing to give you a lot of capital, not having experience in either is just asking for  trouble.