By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - PC Discussion - AMD Hawaii GPU R9-290X ‘Volcanic Islands’ PCB Leaked – 512-Bit Interface and Massive Die Size?

TheJimbo1234 said:

512-bit? Sweet jebus...........
Though I have to say, I'm not surprised and it would explain a lot about the new consoles (especially the PS4's absurd memory throughput) and if they have been working with MC and Sony, then they will have learnt a lot on how to customise their gear - one being that they would need gpu's to be able to access the vast system memory rather than having it's own independent memory.

 

 

fatslob-:O said:
haxxiy said:
fatslob-:O said:

If you take into account all the process node shrinks and how prone to error the silicon is when attempting photolithography I'd be willing to bet the fastest GPU achieveable isn't going to be more powerful than the PS4 by a factor of 8.

Edit: To reach the smallest process nod your going to have to wait 10 years.

Do not underestimate the capabilities of AMD and Nvidia to release increasingly bigger cards in order to keep fooling clueless enthusiasts they are seeing real, Moore law-like progress as far as shrinking and efficiency goes. I mean, remember that once upon a time this is what we called high-end... nowadays it wouldn't pass as the integrated video card of a low end laptop:

Do note that the smaller the process node the more likely the bigger dies are prone to error's. So I wish the manufacture's good luck for being able to get any decent yield beyond a 370mm^2 at a process node of 5nm.

Edit: BTW the TITAN has some disabled units due to how prone that thing is to error's.


They are good upto 2026 iirc, as in they know the techniques to mass produce the dies at an affordable price, however beyond that, they are stuffed - everyone at Silicon Valley has come to that agreement. The good news is that they had a breakthrough with graphene and there is enough money to go into that. However, when graphene dies come in, Moores law will go tohell as graphene can clock well over 400GHz, and that is not also taking into account that less logic gates are used to perform logic functions (due to the gates being hybrid gates).

Moore's law has to do with the doubling of transistors so how does frequency mean the doubling of transistors ?

But how are they going to cool off that type of heat ? (The forumula for cpu power is (C)(V^2)(F)) 

Can you explain to me what a hybid logic gate is. (I know what a logic gate is but how can a gate possess more than 2 logic types or do you mean you can create a bybrid logic gate that is not an AND, NOR, OR, XOR, NAND, or NOT gate.)

Edit: I forgot XNOR.



Around the Network
fatslob-:O said:
TheJimbo1234 said:

512-bit? Sweet jebus...........
Though I have to say, I'm not surprised and it would explain a lot about the new consoles (especially the PS4's absurd memory throughput) and if they have been working with MC and Sony, then they will have learnt a lot on how to customise their gear - one being that they would need gpu's to be able to access the vast system memory rather than having it's own independent memory.

 

 

fatslob-:O said:
haxxiy said:
fatslob-:O said:

If you take into account all the process node shrinks and how prone to error the silicon is when attempting photolithography I'd be willing to bet the fastest GPU achieveable isn't going to be more powerful than the PS4 by a factor of 8.

Edit: To reach the smallest process nod your going to have to wait 10 years.

Do not underestimate the capabilities of AMD and Nvidia to release increasingly bigger cards in order to keep fooling clueless enthusiasts they are seeing real, Moore law-like progress as far as shrinking and efficiency goes. I mean, remember that once upon a time this is what we called high-end... nowadays it wouldn't pass as the integrated video card of a low end laptop:

Do note that the smaller the process node the more likely the bigger dies are prone to error's. So I wish the manufacture's good luck for being able to get any decent yield beyond a 370mm^2 at a process node of 5nm.

Edit: BTW the TITAN has some disabled units due to how prone that thing is to error's.


They are good upto 2026 iirc, as in they know the techniques to mass produce the dies at an affordable price, however beyond that, they are stuffed - everyone at Silicon Valley has come to that agreement. The good news is that they had a breakthrough with graphene and there is enough money to go into that. However, when graphene dies come in, Moores law will go tohell as graphene can clock well over 400GHz, and that is not also taking into account that less logic gates are used to perform logic functions (due to the gates being hybrid gates).

Moore's law has to do with the doubling of transistors so how does frequency mean the doubling of transistors ?

But how are they going to cool off that type of heat ? (The forumula for cpu power is (C)(V^2)(F)) 

Can you explain to me what a hybid logic gate is. (I know what a logic gate is but how can a gate possess more than 2 logic types or do you mean you can create a bybrid logic gate that is not an AND, NOR, OR, XOR, NAND, or NOT gate.)

Edit: I forgot XNOR.


Yes, that is right, but when the trans double, normally the proc power doubles at the same time. Graphene would demolish this as you could have a chip with less transistors still utterly outstrip a current proc. making Moores Law useless.

Graphene requires far less power and is highly resistant to heat, also you don't need mass power to clock at that speed as you are not looking at fermi levels to overcome (uses "negative resistance").

Basically using this negative resiatance, they made something that works essentially the same as a XOR gate, but takes up far less room than a current XOR and they only needed 3 of these negative resistance transistors to make one.

http://www.extremetech.com/extreme/164301-graphene-transistors-based-on-negative-resistance-could-spell-the-end-of-silicon-and-semiconductors

Essentially, we are all going to have to go back to uni to relearn fundemental circuit or proc design when these bad boys come out.



I'll remain skeptical until I read proper reviews.

I don't know if the 512-bit is true or not, not that the 7970 were bandwidth limited, but that Titan killer an the numbers leaked and posted by zarx show increases of even more than 40%, which to me seem to be a lot given that the architecture i basically the same and they are till using 28nm.

But I hope it's true. That will give us a price war that can only benefit us!



Please excuse my bad English.

Currently gaming on a PC with an i5-4670k@stock (for now), 16Gb RAM 1600 MHz and a GTX 1070

Steam / Live / NNID : jonxiquet    Add me if you want, but I'm a single player gamer.

Those benchmarks

And the price!!!

This will be awesome if true



Intel Core i7 3770K [3.5GHz]|MSI Big Bang Z77 Mpower|Corsair Vengeance DDR3-1866 2 x 4GB|MSI GeForce GTX 560 ti Twin Frozr 2|OCZ Vertex 4 128GB|Corsair HX750|Cooler Master CM 690II Advanced|

TheJimbo1234 said:
fatslob-:O said:
TheJimbo1234 said:

512-bit? Sweet jebus...........
Though I have to say, I'm not surprised and it would explain a lot about the new consoles (especially the PS4's absurd memory throughput) and if they have been working with MC and Sony, then they will have learnt a lot on how to customise their gear - one being that they would need gpu's to be able to access the vast system memory rather than having it's own independent memory.

 

 

fatslob-:O said:
haxxiy said:
fatslob-:O said:

If you take into account all the process node shrinks and how prone to error the silicon is when attempting photolithography I'd be willing to bet the fastest GPU achieveable isn't going to be more powerful than the PS4 by a factor of 8.

Edit: To reach the smallest process nod your going to have to wait 10 years.

Do not underestimate the capabilities of AMD and Nvidia to release increasingly bigger cards in order to keep fooling clueless enthusiasts they are seeing real, Moore law-like progress as far as shrinking and efficiency goes. I mean, remember that once upon a time this is what we called high-end... nowadays it wouldn't pass as the integrated video card of a low end laptop:

Do note that the smaller the process node the more likely the bigger dies are prone to error's. So I wish the manufacture's good luck for being able to get any decent yield beyond a 370mm^2 at a process node of 5nm.

Edit: BTW the TITAN has some disabled units due to how prone that thing is to error's.


They are good upto 2026 iirc, as in they know the techniques to mass produce the dies at an affordable price, however beyond that, they are stuffed - everyone at Silicon Valley has come to that agreement. The good news is that they had a breakthrough with graphene and there is enough money to go into that. However, when graphene dies come in, Moores law will go tohell as graphene can clock well over 400GHz, and that is not also taking into account that less logic gates are used to perform logic functions (due to the gates being hybrid gates).

Moore's law has to do with the doubling of transistors so how does frequency mean the doubling of transistors ?

But how are they going to cool off that type of heat ? (The forumula for cpu power is (C)(V^2)(F)) 

Can you explain to me what a hybid logic gate is. (I know what a logic gate is but how can a gate possess more than 2 logic types or do you mean you can create a bybrid logic gate that is not an AND, NOR, OR, XOR, NAND, or NOT gate.)

Edit: I forgot XNOR.


Yes, that is right, but when the trans double, normally the proc power doubles at the same time. Graphene would demolish this as you could have a chip with less transistors still utterly outstrip a current proc. making Moores Law useless.

Graphene requires far less power and is highly resistant to heat, also you don't need mass power to clock at that speed as you are not looking at fermi levels to overcome (uses "negative resistance").

Basically using this negative resiatance, they made something that works essentially the same as a XOR gate, but takes up far less room than a current XOR and they only needed 3 of these negative resistance transistors to make one.

http://www.extremetech.com/extreme/164301-graphene-transistors-based-on-negative-resistance-could-spell-the-end-of-silicon-and-semiconductors

Essentially, we are all going to have to go back to uni to relearn fundemental circuit or proc design when these bad boys come out.

A couple of things to keep note about graphene cricuits is that they don't provide enough bandgaps for digital signals.



Around the Network
fatslob-:O said:
TheJimbo1234 said:
fatslob-:O said:
TheJimbo1234 said:

512-bit? Sweet jebus...........
Though I have to say, I'm not surprised and it would explain a lot about the new consoles (especially the PS4's absurd memory throughput) and if they have been working with MC and Sony, then they will have learnt a lot on how to customise their gear - one being that they would need gpu's to be able to access the vast system memory rather than having it's own independent memory.

 

 

fatslob-:O said:
haxxiy said:
fatslob-:O said:

If you take into account all the process node shrinks and how prone to error the silicon is when attempting photolithography I'd be willing to bet the fastest GPU achieveable isn't going to be more powerful than the PS4 by a factor of 8.

Edit: To reach the smallest process nod your going to have to wait 10 years.

Do not underestimate the capabilities of AMD and Nvidia to release increasingly bigger cards in order to keep fooling clueless enthusiasts they are seeing real, Moore law-like progress as far as shrinking and efficiency goes. I mean, remember that once upon a time this is what we called high-end... nowadays it wouldn't pass as the integrated video card of a low end laptop:

Do note that the smaller the process node the more likely the bigger dies are prone to error's. So I wish the manufacture's good luck for being able to get any decent yield beyond a 370mm^2 at a process node of 5nm.

Edit: BTW the TITAN has some disabled units due to how prone that thing is to error's.


They are good upto 2026 iirc, as in they know the techniques to mass produce the dies at an affordable price, however beyond that, they are stuffed - everyone at Silicon Valley has come to that agreement. The good news is that they had a breakthrough with graphene and there is enough money to go into that. However, when graphene dies come in, Moores law will go tohell as graphene can clock well over 400GHz, and that is not also taking into account that less logic gates are used to perform logic functions (due to the gates being hybrid gates).

Moore's law has to do with the doubling of transistors so how does frequency mean the doubling of transistors ?

But how are they going to cool off that type of heat ? (The forumula for cpu power is (C)(V^2)(F)) 

Can you explain to me what a hybid logic gate is. (I know what a logic gate is but how can a gate possess more than 2 logic types or do you mean you can create a bybrid logic gate that is not an AND, NOR, OR, XOR, NAND, or NOT gate.)

Edit: I forgot XNOR.


Yes, that is right, but when the trans double, normally the proc power doubles at the same time. Graphene would demolish this as you could have a chip with less transistors still utterly outstrip a current proc. making Moores Law useless.

Graphene requires far less power and is highly resistant to heat, also you don't need mass power to clock at that speed as you are not looking at fermi levels to overcome (uses "negative resistance").

Basically using this negative resiatance, they made something that works essentially the same as a XOR gate, but takes up far less room than a current XOR and they only needed 3 of these negative resistance transistors to make one.

http://www.extremetech.com/extreme/164301-graphene-transistors-based-on-negative-resistance-could-spell-the-end-of-silicon-and-semiconductors

Essentially, we are all going to have to go back to uni to relearn fundemental circuit or proc design when these bad boys come out.

A couple of things to keep note about graphene cricuits is that they don't provide enough bandgaps for digital signals.


And that is the key - you don't use the bandgap as graphene is not a semiconductor .

  This was the problem they had with graphene as it either conducts really well, or really really well and no doping could solve it. The breakthrough is the negative resistance. You use that instead of you standard bandgap semiconductor method. The result is something which has logic output due to a large enough potential difference at the output. 



TheJimbo1234 said:
fatslob-:O said:
TheJimbo1234 said:
fatslob-:O said:
TheJimbo1234 said:

512-bit? Sweet jebus...........
Though I have to say, I'm not surprised and it would explain a lot about the new consoles (especially the PS4's absurd memory throughput) and if they have been working with MC and Sony, then they will have learnt a lot on how to customise their gear - one being that they would need gpu's to be able to access the vast system memory rather than having it's own independent memory.

 

 

fatslob-:O said:
haxxiy said:
fatslob-:O said:

If you take into account all the process node shrinks and how prone to error the silicon is when attempting photolithography I'd be willing to bet the fastest GPU achieveable isn't going to be more powerful than the PS4 by a factor of 8.

Edit: To reach the smallest process nod your going to have to wait 10 years.

Do not underestimate the capabilities of AMD and Nvidia to release increasingly bigger cards in order to keep fooling clueless enthusiasts they are seeing real, Moore law-like progress as far as shrinking and efficiency goes. I mean, remember that once upon a time this is what we called high-end... nowadays it wouldn't pass as the integrated video card of a low end laptop:

Do note that the smaller the process node the more likely the bigger dies are prone to error's. So I wish the manufacture's good luck for being able to get any decent yield beyond a 370mm^2 at a process node of 5nm.

Edit: BTW the TITAN has some disabled units due to how prone that thing is to error's.


They are good upto 2026 iirc, as in they know the techniques to mass produce the dies at an affordable price, however beyond that, they are stuffed - everyone at Silicon Valley has come to that agreement. The good news is that they had a breakthrough with graphene and there is enough money to go into that. However, when graphene dies come in, Moores law will go tohell as graphene can clock well over 400GHz, and that is not also taking into account that less logic gates are used to perform logic functions (due to the gates being hybrid gates).

Moore's law has to do with the doubling of transistors so how does frequency mean the doubling of transistors ?

But how are they going to cool off that type of heat ? (The forumula for cpu power is (C)(V^2)(F)) 

Can you explain to me what a hybid logic gate is. (I know what a logic gate is but how can a gate possess more than 2 logic types or do you mean you can create a bybrid logic gate that is not an AND, NOR, OR, XOR, NAND, or NOT gate.)

Edit: I forgot XNOR.


Yes, that is right, but when the trans double, normally the proc power doubles at the same time. Graphene would demolish this as you could have a chip with less transistors still utterly outstrip a current proc. making Moores Law useless.

Graphene requires far less power and is highly resistant to heat, also you don't need mass power to clock at that speed as you are not looking at fermi levels to overcome (uses "negative resistance").

Basically using this negative resiatance, they made something that works essentially the same as a XOR gate, but takes up far less room than a current XOR and they only needed 3 of these negative resistance transistors to make one.

http://www.extremetech.com/extreme/164301-graphene-transistors-based-on-negative-resistance-could-spell-the-end-of-silicon-and-semiconductors

Essentially, we are all going to have to go back to uni to relearn fundemental circuit or proc design when these bad boys come out.

A couple of things to keep note about graphene cricuits is that they don't provide enough bandgaps for digital signals.


And that is the key - you don't use the bandgap as graphene is not a semiconductor .

  This was the problem they had with graphene as it either conducts really well, or really really well and no doping could solve it. The breakthrough is the negative resistance. You use that instead of you standard bandgap semiconductor method. The result is something which has logic output due to a large enough potential difference at the output. 

I thought negative resistors were theoretical ?

What happens if they don't find a real solution to the bandgap ? (Well there's bilayer graphene to do it to solve the bandgap issue.)



No 120fps no buy.



They need 20nm.

At least this release will make nVidia reduce the price or their cards.



Shinobi-san said:
Those benchmarks

And the price!!!

This will be awesome if true


God can you imagine the massive price drops?!  Think if you bought a Titan for $1000, and then the next week it was $700!!!



Prediction for console Lifetime sales:

Wii:100-120 million, PS3:80-110 million, 360:70-100 million

[Prediction Made 11/5/2009]

3DS: 65m, PSV: 22m, Wii U: 18-22m, PS4: 80-120m, X1: 35-55m

I gauruntee the PS5 comes out after only 5-6 years after the launch of the PS4.

[Prediction Made 6/18/2014]