News Intel GPUs - Falcon Shores cancelled

[511]

Why did they even chose It?

Intel is infamous for this thing their love with HP Libs why do you think Raptor lake clocks 6Ghz they freaking relaxed the CPP from 54nm to 60nm for Golden Cove/Raptor Cove to hit 5+ GHz.

I found this.

49% difference in density. I highly doubt HP allows the same increase in frequency. Actually, B580 clocks as a RTX4060.
But I must say that despite using HP libraries B580 is pretty efficient.

HP allows this performance increase that is why it's HP B580 Is 72 MTr/mm2 for whole die vs 92 for N5 logic I guess the media engine is taking quite the space.

 

[TESKATLIPOKA]

Intel is infamous for this thing their love with HP Libs why do you think Raptor lake clocks 6Ghz they freaking relaxed the CPP from 54nm to 60nm for Golden Cove/Raptor Cove to hit 5+ GHz.

We are talking about GPUs here, not CPUs.
In case of GPUs you can pretty easily widen them for more performance unlike CPU, diminishing returns are not a real issue, If you don't go ridiculously wide.

HP allows this performance increase that is why it's HP B580 Is 72 MTr/mm2 for whole die vs 92 for N5 logic I guess the media engine is taking quite the space.

What performance increase, be more specific? Without HP It would clock at only 2GHz or what do you mean exactly?

B580 has 72 MTr/mm2 density
AD106 has 120.5 MTr/mm2 density
AD107 has 129.5 MTr/mm2 density
Nvidia has 67-80% higher density, that's simply a huge difference.

I believe Intel could have dropped clocks to just ~2GHz(-30%) and put 50% more units inside instead and be smaller and have better efficiency while performing the same by using HD libraries, but this is only my belief.

 

[511]

We are talking about GPUs here, not CPUs.
In case of GPUs you can pretty easily widen them for more performance unlike CPU, diminishing returns are not a real issue, If you don't go ridiculously wide.

We are also talking about Intel here 🤣

What performance increase, be more specific? Without HP It would clock at only 2GHz or what do you mean exactly?

I Meant using more metal layers and taller cells you can gain more performance out of a circuit.

This chart would be better describe this in terms of performance difference in PnP characteristics of HP vs HD the only odd one out is Intel 3 2 Fin lib which has higher performance than the 4 3 Fin Variant and that is the only one I know of.

B580 has 72 MTr/mm2 density
AD106 has 120.5 MTr/mm2 density
AD107 has 129.5 MTr/mm2 density
Nvidia has 67-80% higher density, that's simply a huge difference.

There is and there is no doublt about it

I believe Intel could have dropped clocks to just ~2GHz(-30%) and put 50% more units inside instead and be smaller and have better efficiency while performing the same by using HD libraries, but this is only my belief.

I don't how much clock speed hit they would have taken but it could have been smaller definitely.